Came here to say this. Or, if you are just even slightly handy, it's quite easy to open the camera case and just flip the port. Closing it back together can be snapping in place or just a few drops of adhesive.
dual cam with telephoto makes the most sense, wouldnt be too complex.. i'm surprised its not done already. smartphones have 2-3 camera setups for years now.
@@Tom-sg4iv Most of the high end ones already got multiple sensors video modes. Theres even one from Axon that lets you use all 5 simultaneously, including the selfie one.
@Phillip Banes - I remembered that I saw a vid showing that it's possible to record from two back lens on a phone (wide + telephoto) at a very good quality recently. I don't remember what phone it was, but I found this video made two years ago about an app doing it. ...tho, what I saw a few weeks ago been native in the phone (no need 3rd part app). I wish I could find what phone I saw exactly but here's what I found for now: th-cam.com/video/dsjRHhx9zb0/w-d-xo.html
Maybe the future is to ONLY have a telephoto camera, but have 4 units covering each corner of the car. Stitching the image together gives you an Ultrawide footage anyway. Except the image is sharper. In fact, I'd be completely happy if Western Nations made it mandatory for all new cars to have this as default. It should make insurance claims, and really MANY issues very solvable thanks to unbiased footage. Old cars on the road can get them retro-fitted, but with time slowly they will be replaced with new cars that have these installed. Cameras, Lenses, Processors, Storage, and Connectivity has become far cheaper, abundant, and better around 2017 and beyond... compared to a decade before that, and the decaded before that.
Dude nice work, this vid could very well change the industry. You did most of the leg work for the manufacturer's. If I was one come Tuesday this would be the new direction of the company.
Hmmm, providing prof of concepts testing off the shelf parts and putting it all in a well organized presentation such as a TH-cam vid seems like quite a bit of leg work. Implementation is of course gonna be difficult that's why the ball lands in the manufacturer court as it will require a team of people.
@@toddstratton9197 Proof of concept is usually the stage where someone actually integrates the parts together as a starting point for development of a real product.
Great idea. Could be an opportunity here for one of the dash cam manufacturers to put a dual front-facing camera out to test the market. As long as it's not too expensive it could potentially be a winner.
Linus Tech Tips covered a similar issue with dash cameras very recently. He showed even the more pricer cameras image quality was not much better than the sub $100 ones. The reasoning is they are all basically using the same System on a Chip, and those are mostly at least five year old technology. We are probably better off making our own with something like Raspberry Pis
Most current dashcams are still using processors that came out 5+ years ago. Typical story of racing to the bottom. The tech is there, SoC with built-in high-res computational videography optimized for tough lighting conditions. There's just not enough market demand for higher end dashcams.
And that's why we need to get more people on board with it. Insurance companies should require them. Bigger market, higher demand, more units sold, higher budget and more competition equals better products
Another great video. I left almost this exact idea in my comment on your last video 😀 If you want to do some additional testing most of the rear camera setups that I've tested are actually still utilizing standard analog security camera wiring (AHD/CVI/TVI). You could get an analog security camera with a telephoto lens and with a little splicing you could plug it into the rear camera input of most dashcams (the ones with the TRRS connectors would be the easiest). As far as exposure on the plates, the tech you are looking for is called "HLC" or highlight compensation. Basically the algorithm works so that there are no clusters of pixels on the image that have a value of 255 and the camera will dial down exposure until that block goes down to 254 or less. HLC would need to be built into the dashcam though, so that would need to be a firmware change.
Yep, Google popularized this in their smartphones, and Samsung and Apple have copied them. It's actually brilliant. First you capture high framerate, underexposed video. Then you combine ~4 frames into 1, compensating for anything that moved. Now every pixel has more precise color/brightness, so you can crank up the brightness and colors without making the image noisy. Obviously you need a lot of RAM and a powerful ISP/DSP.
@@runed0s86 Yea, that's something I've already thought about several times while watching dashcam footage on TH-cam... you can stack multiple frames from existing videos to enhance the license plate and make it readable. It would take both stacking and adjusting position, zoom, and skew to line each frame up to the reference frame, but it does work. I've worked on image processing systems that have done the same for identifying images printed on a printing press.
What about taking the dash cam that has the optional plug-in rear facing camera and using an HDMI cord between the main camera unit and the optional camera so that you can turn and face that rear facing camera towards the front?
Was thinking the same. An HDMI Extender (HDMI Female to HDMI Male) would probably work. There are some on Amazon that are as small as 15 cm (just under 6 inches).
For a higher performance LPR capture while on the move, we need a higher frame rate camera. Many of the purpose built LPR cameras see in the infrared, so headlights won't affect 'seeing' the license plates, and because plates are reflective, the cameras have IR emitters and are constantly sampling at different exposures (similar to photography and exposure bracketing). The color/visible image is simply just for context; maybe the Axon dashcam/LPR cam relies on the visible image and does OCR against it. Edit: jumped the gun, commented while I was still watching :)
BRILLIANT!!! I would have suggested taking the tiny camera apart, turning the connector over and reassembling the camera. Your suggestion is so much simpler!!!!
Great work! I would hope this gets to the manufacturers because this is exactly what we need. You need to buy an HDMI reverse adapter lol. To answer your question though, I'd pay $200 for a 2k dashcam, then $100 - $150 for the license plate add on. That's if I knew it worked very well. It'd be nice to have options to add a rear onto that especially since 3 channel dash cams are out there. So yeah, $350 for a dashcam + license plate reader.
Honestly, I have been in 4 accidents with a dashcam. Only one, was a hit an run, at night, and in the rain. So, anecdotally, this was only needed 25% of the time for myself. As expected the plate was unrecognizable between headlight blow out and the rain on the windshield. In the other cases, the only sufficiently needed information was the make, model, color, and the circumstances of the collision. While I wish there was a consumer level product that could assist in these incidents. I'm not sure if the additional expense is really needed. That premium would really be a luxury. Especially considering I have only ever run sub $100 dash cams. I still run two Yi Dash cams today. They are going on 7/8 years old and still perform the same. I can't fathom spending $400 on a dashcam that gives me the same story as a $60 cam does, the Mid 2000s white Ford Escape failed to stop resulting in a collision and drove away. Of course my eyes told me the same story; but, it was nice to provide to the Highway Patrol and the insurance company to prove it wasn't an at fault collision. For what ever that was worth to my insurance rate.
These past couple videos have been some of the most helpful videos you’ve done. I think you are really on to something with multi-lens dash cams facing forward. I saw your video on the GoPro, are you going to review the new GoPro 11 Heroes that are coming out that are supposed to have better cooling & even higher res + frame rate? P.S. You have to remember heavy mark-up for government agencies like police departments because of paperwork & compliance costs, plus limited market. Mass marketing can drive price down by harnessing a larger audience. You should try to Zoom interviews manufacturers & engineers. I’d be curious to hear what they have to say. I don’t care about size of camera that much so long as I can mount it high on my windshield.
Apparently there needs to be someone that can produce a gimble mount IR laser to lock on to police license plate readers and saturate the CCD. Get on this one now Uniden!!
You can get covers for your plates that show your plate fine when looked at straight on but blurs your plate when viewed at an angle. Technically, they are illegal in many states because there are blanket laws about putting covers on your plates at all but since you can see the plate fine when viewed from head on...if you were pulled over, it looks normal from a cop behind you.
@@MrAgentEcho A) I never said anything about pointing it at anyone. B) Since you seem to know what the power output from an IR laser to distort the capability of a CCD is, I would be curious to know what that would be, are we talking micro watt or straight watts of energy? There are lasers in the visible spectrum that are used as cat toys, so I am curious to know what power rating is required?
Yeah, the Motorola (really Vigilant) ALPR solution really is the best in the industry for license plate capture. I've seen the mobile one installed on patrol vehicles as well as the stationary ones they use for the radar trailers. They all run Linux under the hood and have a passively-cooled "VLP" box the cameras connect to for video processing. You're right that they do pulse the IR floodlights in low light to capture reflective license plates between regular video frames. That's why they've got a regular RGB camera and an IR-bandpass one.
Coming from digital cinema cameras, the outboard processing computer may have more to do with the clarity than one might think. Modern digicine cameras offer raw footage recording capabilities to be processed later, generally this is preferred unless storage or processing time (budget) is an issue. These Linux boxes might not seem like much, but compared to a dinky processor (even a dedicated imaging chip) in a dashcam, they are likely far better at rendering plates in sharp detail. I think the dashcam market is going to be stuck in compression town inevitably, but perhaps companies can devise some software to separately capture or parallel process the plate camera footage with dedicated algorithms like your hfr multi exposure examples.
I've had some experience working with dual camera setups while researching for a project. Turns out you can use a Raspberry Pi computer and a set of specialty board that can allow it to use two cameras at once, either for stereo vision, or for extended fields of view; or in this case, for a much more detailed view off a section of the main (wide angle) image. These can be loaded with quite advanced IMX477 image sensors, and use separate lenses (say, 140 degree fov as wide angle, 20 degree fov for zoom in); and use the computer to process the data of both and deliver near-real-time enhanced imagery.
Been thinking about this problem myself. I'm an embedded software/hardware engineer and don't think it would be hard to have a 4 camera system. X2 facing front and x2 facing rear. The x2 cameras can be a wide angle stitched with a telephoto. Maybe a good idea for a kick starter? Could also incorporate some license plate reading and saving to a cloud where the data can be mined for when, for example, an amber alert is put out.
THANK YOU! I know optics and math and resolutions, and after trying and returning two high end dashcams, I figured these wide angle cameras just can't resolve plates. You've given me the first confirmation I was right on this. One novel solution might be to have two front facing cameras, each with a 60-70 degree field of view, each barely overlapping in the center, and letting software stitch them together. Effectively doubling the resolution.
Note: LE ought to need a reason to believe the person-in-question is *actually* driving the vehicle *before* initiating a stop. Otherwise, its tantamount to pulling vehicles over to examine their 'papers', which is expressly forbidden according the SCOTUS. At the very least, if an officer does initiate a stop, they must abandon it as soon as they determine its not the person they're interested in. As we all know, however, LEO's will do it anyway until they're penalized for doing so.
how can cops know before hand that the person they're interested in is indeed the one driving if they cannot stop to verify? just doesn't make any sense or am i missing something here?
@@lordjaashin You're close but you're missing his point. The police still cannot legally pull The vehicle over because they believe that the registered owner could be the one driving. However with how convoluted and overly complex driving laws have become over time they can just pull you over for any number of reasons if they follow you long enough... And that's what they usually do. If they do pull you over simply because you popped on the plate reader, then you could possibly get it thrown out but they would have to testify that's why they told you over and most of them are not going to do that.
Phone cameras can do face or eye detection. You could do something similar with License Plates which could auto adjust exposure for that region. That tech exists today. I agree, two lenses would be ideal. The second camera wouldn't even have to take video, just photos with timestamps, it could even integrate it into the main video with a little intelligent mixing. There are a lot of options.
Hi Nice video I own a company that has been manufacturing ANPR or as you call them ALPR cameras for car since 1997 we have supplied the police and the armed services in the U.K we also specialised in capturing farsi (arabic licence plates) we looked at using dash cams for in car use and found that the Chinese made dash cams with "starlight" features which would capture licence plates in the dark very well. You may need to look into lens filters as well. However we had to modify the firmware in the cameras a little to capture cars at speed. I look forward to your next video.
I also think 30fps in basically unacceptable and unusable for daytime footage, at night it should automatically switch to lower framerate for maximum light, unless headlights are on turn go back to 60fps…
@@johnwilliams1091 but then a wider aperture decreases the depth of field, meaning the image would be out of focus for most scenarios, defeating the object of capturing a legible plate image
I think manufacturers would sacrifice video quality (lower bit rate) to make it 60fps. I think that's kinda similar issue in those 4K dash cams. (One of many reasons why you get garbage 4k video quality )
Very nice work. As a lil tip, you coulda try using a HDMI flexible female to female extension cable, paired with a mcable and you'd have a better clarity, "native" 1080p upsampled with the possibility of having both cameras facing forward. Of course, theres quite a bunch cables to organize plus gotta also mount properly the camera, but you'd have pretty much what you were hoping to get from this video.
Good video. I did not know that the regular dash cams were so poor in capturing license plate numbers. The license number is critical information and I regard this as a key requirement of any dash cam system that I would consider buying and I would be willing to pay a premium for such a capability.
UltraDash Z3 actually has the dual cams feature for $150...one wide angle and the other telephoto. Please review for the channel and keep up the great work!
Excellent video(s) on this subject. A couple of things… You mentioned aperture… the trade-offs here are for lenses with wider-open “maximum” apertures (smaller f-numbers) the lenses must be larger, the cost increases, and depth-of-field becomes shallow. Another optical consideration is the lenses formula and the materials used in the lens elements. Edge-to-edge sharpness, or lack thereof, might be negatively impacting some cameras’ ability to capture legible license plate numbers towed the edges of the frame. Perhaps a RAW video format might be useful in one camera of a multi camera set-up to allow the adjustment of exposure to isolate license numbers. Again, I found and enjoyed your videos today and subbed!
Another option, 2nd cam would be a 240 fps (preferably) Black & White cam with telephoto, although a slower fps would allow for higher res pictures. LPR cams are usually B&W and are proven to work. The characteristic of B&W cameras are more suited to License Plate Readers.
*Simple solution* :- Go with the Nextbase dashcam at 5m45s and just connect the small Telephoto-Cam with a *USB EXTENSION CABLE* !!! ... and locate it where you want with a windshield suction-cup mount, with either an adjustable holder or magnetic type using a small metal disc/washer stuck to back of the telephoto cam.
If people are already willing to pay upwards of $850 for an Escort Redline 360C, I’m sure there’s a market for more expensive dash cams. Not to mention custom installed solutions like Radenso RCM that are already over $2K. However, if dash cam companies decide that they don’t think that these can become a mass market option, I could see them upping the price to compensate. So they may just end up doing the equivalent of radar detector companies of the single front camera being the regular(windshield mount detector) and the premium 2 front camera model(custom installed).
Not sure what the obsession is with capturing plates. I just want a dash cam that will capture the event so the other party can’t lie about what happened. If you need a plate and you have Audio recording on your dash cam, just read it out load so it’s included in the footage.
@@markbajek2541 a plate would be the last thing I would be worried about in a head on collision. Both vehicles would most likely be disabled and hopefully everyone survives.
Human Error. In the heat of the moment, you could mistake a Zero for an O or a D. Same could be said about I or 1, G or 6. When capturing a license plate you don’t need to worry about any of that and tend to injuries or getting off the main road.
Capturing plates is not most important thing.in most cases dashcam is using to find out how accident happened. I rather spend money on reliability . In hot climate, long parking outdoor, dashcam is too weak to survive.
I feel computer learning and better SOC systems are going to play a huge role in this in the next 10 years. Along with better camera hardware as it should be doable to have AI do on the fly exposure and post processing activities not on the Dash cam but on big iron servers that can process what was captured. The key is to have a a crap ton FPS, and multiple sources from multiple lenses. Then its a matter of uploading and let computer learning do its thing. That said, no light is no light. There is only so much AI can do to bring out details. But I will be honest some of the stupidly blurry images I've used on publicly available tools to bring out image details borders on voodoo magic. In 10 years? Sheet. This stuff be easily integrated with dash cams assuming manufacturers feel the need to push the edge on what can be done.
My smart security camera can recognize people cars and dogs, and then track them, adjusting the orientation and follow them. It wouldn't be a huge reach to track rectangles with the ratio of a license plate, then apply zoom and exposure changes. putting this into usable data is merely code. We want to export into a document , Car make model year and color, and of course the license plate number in the form of a .csv file with spreadsheet compatibility. the problem with my 8mp smart security cam though, it sucks at capturing motion clearly.
I think a lot could be done in post processing: there is plenty of ways to stack the multiple still frames from the video in order to get an enhanced image. Astrophotography leads the way in this field. Other esoteric solutions may involve free-form optics that give you a higher focal length in the centre of the sensor, and "squish" the rest of the wide field of view to the outside area of the sensor. Another solution would be using monochrome sensor, which would give you better resolution, better dynamic range and better low light performances; c-mos sensors could, in theory, shot with two different exposures at the same time, at the cost of having about 0.7 the native resolution.
One thing I haven't seen mentioned yet is using aspheric lenses. They're typically used to subtly correct optics, like focal distances for glasses or correcting chromatic aberration in compound camera / telescope systems, but there is nothing preventing their manufacture to give telephoto-like capabilities toward the central region with a smooth transition to a wide angle-like toward the edges of the the image circle. It would require additional optics or post-processing to deal with not just further chromatic aberration but also make it sensible to view in software since otherwise you would have a very fisheye-like (-like; they are very different optically) view of the image. All in all, the custom manufacture and R&D required is likely more expensive than simply placing a second camera module, but thought I'd throw it out there.
Modular dash cam? Single Dash Cam (main product), attachment cam on both sides of the main cam. One for Telephoto front cam and one for interior cam. And a cable port for rear cam. That way people can buy what they want and you mainly push the single front camera and/or front and back kit. Have the others as add ones for people who want them.
So, if we were willing to spend the money now, is there a recommendation for a second (dedicated) camera, available on the consumer market, to specifically capture license plates?
7:04 according what i heard about us plate chaos with different designs and plates the the same text but different design (even in the same state) being able to be given to different ppl the telelens cam is a lot better for specifically getting that.
Several things came to mind. 12:45 - You would have to have a separate, tiny IR light with a flexible angle-cut lens that conforms to the windshield glass. The light will shine out, but stop the light from reflecting in the interior. Probably too much work, but it would fix that glare problem. 19:19 - You could buy (if they make them) a 1/2 inch or 1 inch MALE TO FEMALE cord/connector and simple twist the inner facing cam to face out the same as the main cam. 14:40 to 15:06 - Since ALPR and speed cameras need a reflective surface, I can use some cover or chemical and remove that reflective coating, rendering the systems useless, lol. evil grin
Covering the license plate is illegal in places. Steve Lehto from the Lehto’s Law TH-cam channel recently had a video where the plate was only covered around the edges by a decorative frame, and IIRC there was a citation issued and it was upheld in court.
@Commercial Guy Last point you made... ' And get pulled over immediately for altered tag '. There's a law against doing anything that can hinder a cop's ability to do his job. Not worth it.
Sounds to me that getting the data is the most important. The rest is software. I think that one 2k camera can do the job. Run it in HDR mode at variable frame rate. We're not shooting blockbuster movies here. Store each level of exposure into its own video file rather than merge the HDR frames into a single frame. Basically, stop thinking of a dashcam as a video camera where you want spacial coherence but think of it as a continuous fast photo camera taking many photos at a wider range of settings. As a video it will be ugly to watch but post processing should produce a result good enough for court. If you want pretty pictures... Get a gopro and shut up.
You’ve just been describing a modern smartphone. Multi lens, IR/LiDAR, GPS, Motion detect, image stabilization, ML image enhancement , HDR, multi-stitching, WiFi/BT/5G, large storage, fast, processing, standby battery, large screen, ETC. If one came in a durable heat resistant body, $1000 sounds about right.
love your dashcam videos, they truly are in a league of their own, no other channel comes even remotely close always so impressed by the quality of every single one of your videos
Not only is it important to me since the current setups cannot deal well with plates, especially moving ones, the cost of the current ones isn't relevant to me as I wouldn't purchase any at the moment. If you get hit and the jerk takes off what good is it to have a dash cam if it cannot resolve the plate?
Well shit, it does not do Oakland California pd any good. to have the cameras .You know why ? because hell most of the cars don't have plates to read. It is beyond me why they don't get stopped . Yesterday, I saw a car with no front windshield and expired tag as far back as 1989. This car was in downtown Oakland 2 blocks from the Pd station it's comical.
I think this technology could be easily retrofitted to existing dual camera setups. If say Viofo modified their rear facing camera to function better as a front facing license plate focused one then it would be instantly viable. Anyone could just plug it in.
I would definitely like a second telephoto camera for license plates. Two cameras (wide & narrow) at the front, two at the rear (ditto), and one interior, for five channels in all.
One method that I do to improve night time by using a second cheaper dash cam with the exposure comp settings set to their lowest setting. This allows the dash cam to use a higher shutter speed and lower ISO which improved license plate visibility as well as the distance at which they are visible.
you could just use a male/female HDMI cable extension in between your two cameras, that would allow you to mount the "rear" camera in a better position on the window for license plate capture. you could use the same to mount your IR camera externally so you are not getting IR reflection off of your windshield.
Hi I recall a recent video that you published like 2 months ago with the external cameras mounted on the two front side doors to show blind spots, these cameras are outside the car and will most likely take care of the front windshield glare you mentioned for the night UV vision. Just and idea maybe this can be the solution to night glare from windshield, to have a small camer installed outside perhaps in the grill area for weather protection and better visibility
Great detailed talk. There are tons of dash cam reviewers on TH-cam but you are the most detailed reviewer. Kudos to you. Today I just decided to grab the Tinkware U1000 after much research. Needless to say I am not a fan of having to pay a premium price for a 4K dash cam. But I want a very reliable product while not draining and damaging my car's battery for the long term especially when the cam is operating under parking mode. Anyway, I digress a bit... I think nowadays since folks have absolutely zero qualms about shelling out a cool $1000 just to own an iPhone, we have to ask ourselves how much we are willing to pay for a dash cam that can capture license plates under most conditions especially when accidents occurred? What will the price be when we have to deal with insurance companies arguing back and forth about accident claims? Personally I think if it is at or near a $1000 and no more it will be a fair price to pay for the added important dash cam feature. I believe most dash cam manufacturers are lazy. If they are making tons of money then they have no incentive to really innovate especially when we are just now, kind of, trying to get back to normal after the pandemic. Lastly, I don't think this will be a huge technical challenge to put in a second telephoto len like the ones found in our smart phones for dash cams. What needs though is to spend some time and effort to implement AI on top of the hw so we can optimize various light conditions, object speeds, shutter & exposure settings for capturing the license plates. Of course nothing is perfect. AI most likely can help but it is not perfect. There is no fool proof magic bullet. Maybe with the new Sony sensors on the horizon we can easily pair it with a low end, low cost, Nvidia graphic chip having optimized AI just for image processing, image tracking and nothing else. Together with the improved HWs we probably can make vast improvements in this area.
@@seekfactsnotfiction9056 Hi Hi. Thanks for the info. Do you personally own the A119? Can you elaborate more why you feel this is the best option available for said task? Thanks much!!! Drive Safe. 🙂
As you were discussing the telephoto aspects, I wondered if attaching one of the many cellphone "telephoto" add on lenses that are available to a standard dash camera. A person would need to run two cameras then, but it might be a cheap and easy "mod" if you already have an older dash camera sitting around.
As an owner of a dr900x-2ch+, I think that $7-800 for the same system, with an additional license plate capture lense, would be my target price. Maybe 850 if they can prove to have really good low light performance too. A dashcam can prove that the accident was not my fault, but it's still my insurance premiums at risk unless we can determine who's insurance the claim should be submitted to
In that low light condition, the camera with HDR support can do better compared to the telephoto lens. I also have done some testing with the telephoto lens, the challenge is the same as the video showed, it is still hard to get a good result for night recording. Most of dashcam Infrared LEDs are not so powerful, I have tested night recording with more than 20pcs IR LEDs, it doesn't help a lot to improve the brightness, but the LEDs PCB is overheating.
What are your thoughts on having the second lens focused on freezing motion in the daytime and exposing for bright plates at night? I like the idea of HDR, but I haven’t had good luck with using it to capture plates at night, especially while moving.
I have always been curious why post processing software doesn't exists that does image processing on PC rather than camera and create enhanced pictures from video.
Curious if the ir from a laser jammer could be harnessed as external ir headlights for helping with license plate capture ( with firmware and maybe some form of pairing between camera and laser systems
You can make a better test using a HDMI Down Right Angled 90 Degree Adapter and mount the camera in the write place. hope you do a video with the proper installation.
Extremely high refresh rate would be best need like a 120 shutterspeed with a very large aperture and a sensor that has very low noise at 800 iso. Pretty much these repurposed cell phone image sensors aren’t going to cut it
10:28 During this slow-mo, look how one frame around 10:35 is much clearer. Video compression is making things worse, and the clear frame is a key frame where it sends the whole frame fresh. Memory cards are so cheap now, they need to offer the ability to really push up the video compression quality.
I'd love an lpr addition. Just wonder if an active filter that can move between the lenses to block most of the light and basically keep just the (near) IR at night. How well would that work?
It's not really made to be a dash cam first and foremost but the comma 3 from comma ai (which is designed to be a after market driver assistant system, hands free driving and all that) has a smaller FOV camera and a large FOV camera. It's a 2,000 solution though so it doesn't really work as a dash cam by itself unless you could use the other features It is getting HDR support so hopefully license plate reading gets even better.
On board with the dual facing cameras. It could be two separate video files just how VIOFO does with their dual f+r camera. Ideally I would prefer 4 cameras. Two front, two rear. An external IR array accessory could be mounted behind the grill inconspicuously to help with low light parking conditions.
you missed one very important thing. It's optical performance. if plate is made out of 6 pixels, with a diffraction limited lens you can read it without a problem. problem is that optics are sooooooo bad they can't use all sensor performance, so it does not matter 4k or 1080p, quality would be the same. Test Sony RX0 II, it has 6 aspheric elements in objective
Teslas have the same issues.. plenty of reports on forums about plates garbled or not viewable, particularly at night. And not to mention issues like blinker light bleed into repeater cameras in early 3s
i have always wondered if it’s possible to use an object detection algorithm to spot police cars. and if it would be more effective than traditional radar detection alarms
About the infrared, there's the reason license plate cameras are mounted outside, glass reflects infrared, to the point temperature cameras have a ton of trouble seeing through it.
I run a 3-channel front, interior, and rear set up currently, but I could see myself getting a 2nd camera for telephoto front and telephoto rear setup. Capturing plates is very important.
That moment when "your doing it wrong". Grab a cheap security camera, the ones that have night vision. then point that at a license plate. the background will reflect IR light like a lot but the letters will not. then you read the black space not reflected. This is why the cop ones have a bunch of leds in it. there IR leds. the computer software just gets a black and white image from the IR camera.
Stereo cameras possible combination. 48MP wide + 2MP telephoto. At night the 2MP telephoto can keep its higher shutter speed thanks to it's bigger pixels. The 48MP would keep the overall wide view and help the telephoto with HDR stacking. 48MP left + 48MP right. Like a 360 camera but in a 120° view. Much more resolution for a smaller FOV and quad Bayer filtering for lowlight.
That DR900X Plus uses h264, low bitrate, most of the non keyframes are just made up pixels...h265 and intra would do wonders for frame to frame clarity. That DR900X Plus also has HDR always on, and extreme noise reduction always applied. Those are the main reasons even high shutter speeds in the day can't save the newest Blackvues. Oh yeah, can't turn the gain/brightness down on the plus units, so blowout not controlable in any way now.
That's because all dash cams are legitimately the same using hardware from the same 2-3 companies which are using technology from nearly a decade ago. If a company wanted to actually do some r&d work and not use the same off the shelf components that every other dash camera company is using they could easily dominate the market!!
LOL. You did a better job than the manufacturers! I think people will pay more money if the option exists. I have plenty of friends, most making less money than me, that go out and spend $1200 on an iPhone or flagship Android every two years without even batting an eye. In fact, sometimes, the more expensive it is, the more demand there is. Not for me, but for many people. Funny thing is, I've been looking at borescopes lately and they aren't very expensive yet some of them have THREE cameras. Although, I'm guessing those cameras aren't nearly as expensive. Anyway, good job on the video.
Dashcams would be so much more useful if they could pick out the license plates and save them to a text searchable database alongside the images they grabbed them from. You might have passed that car that hit you a week ago, but would you remember it? No. That database file would though. Even if it mistakenly put a I for a 1 or a L it would be an obvious false positive.
What you need is a iPhone or mobile phone with multiple rear cameras (front dash cam) one that’s wide HDR good in low light, one telephoto HDR, and one front facing cam for in car or rear view. Plus it has all the sensors your want for g forces, has LTE connectivity, WiFi, Bluetooth, live streaming, 500GB of storage, 4K for all cameras, uses hardware to encode hvac / h265 codec. Has a large built in battery for park recording. If you took out the screen from a phone, you could make the dash cam pretty small. About the size of a black view if you left out the big battery. Ideally I’d take the camera bits out of a phone, make a 3D printed housing for them that’s tiny. Run a extension to the main board in the phone. Put the phone in a box in the glovebox. Make a custom app that could record from all cams at the same time and had all the features you’d want. It would be mostly plugging into apis that already exist. Could easily use it to instantly upload video clips to a server. Could use as a in car hot spot. Use it with CarPlay. That would be ideal. I bet you could make a camera module and plug it into CarPlay via a lightning y adapter and have your regular phone be the recorder. Have it input metadata from the car into recordings via CarPlay. Or just make a app that acts as a remote for the dashcam phone that works in CarPlay that manages start stop, lets you see what the cameras are seeing, configure settings. You could open the remote app using your regular phone that’s connected to CarPlay and see love views from the cameras including the plate reader, and allow you yo review old recordings. One step further, use apples image recognition in the app to identify objects (red car, lady holding monkey) and use their image to text functionality for the license plate. So you’d just have to program the app to use all these built in features. Take snapshot. Save raw image. Hdr. Send to apple api for OCR, load back into app add text to list that is in a plate format. You wouldn’t have to develop or adapt any existing programming libraries that did all that. You could sell the software the app and accessories like holders, camera pod for adding remote cameras so they don’t have to have their phone in the windshield. But that’d work. Making it just the app. Phone mounted on windshield. And use CarPlay for control and still maintain full use of the phone to play music answer calls. Use Siri. Phones from the last few years have the processing power tye gpu power to do all this. I’m just curious if apples sdk let’s developers access more than one camera at the same time, and if the lightning port would be capable of supporting external cameras. Or would you need a external box to encode them and then send to the phone. Or heck just have a small DVR brain be mounted somewhere hidden and have cameras mounted up including a 3 in one camera. I don’t know why there aren’t more dash cams that work like a DVR or NVR. I had one several years ago that did this. Was great. Just had a tiny camera in the front and back and the recording processing box went in my glovebox which also had a AC vent so it kept the thing cool as well. I hate having a big ugly camera on the windshield. Also having the DVR in a locked glove box makes it a lot harder to be stolen. Heck these days if people can see anything in your car that might be of value, like a expensive camera sitting in the windshield they’re gonnna smash the window and take it. Speaking of, also if your dashcam iphone got stolen, you could track it, and remotely wipe it and disable it from being used again. Even track it via BTLE when the battery dies or power gets shut off.
I mounted an IR light outside to eliminate the glare from the windshield and disabled the ones on the unit itself. I was just tinkering around with a cheapy dash-cam to see if I could improve it at all, while running my trusty Viofo to do the real work. It actually improved them both, and the Viofo doesn't even have any IR lights built in, so I think the sensor is just able to use the additional light even though the unit isn't designed for it. If someone is serious about solving this problem (consumer or manufacturer), I think it's an easy decision to make. We're all getting out of our "comfort zones" or at least moving beyond plug-n-play when we start hard-wiring these things in to clean it up, semi-permanently mounting them instead of using suction cups, setting up apps and VLC players to decode, transfer and remote save files. . . . so, popping the hood and bolting a little box behind the grill should still be considered a DIY-able install for most people, as long as they can follow wiring instructions. Sure, it might be tough for manufacturers to green light something that makes the casual owner quickly flip the page to an easier option, but if the right marketing was in place, highlighting a clearly superior product, I know it would sell. (at least one)
I have a license plate reader camera at my house and it only cost me about $250 bucks, no need to spend thousands of dollars. Most decent security's its possible, you just have to change the shutter speed. Also need to move the camera close to the street at a good angle or if you wanted it mounted to your house you need one that has a telephoto lens.
As for the reverse HDMI - just get an HDMI extender cable.
Exactly!
how about one thing you plug in between the cameras that just flips the hdmi port around?
Beat me to it.
I could find a mini HDMI male to female cable but did find a micro to mini cable and adaptor back the other way, like $10 on eBay for both so bargain.
Came here to say this. Or, if you are just even slightly handy, it's quite easy to open the camera case and just flip the port. Closing it back together can be snapping in place or just a few drops of adhesive.
dual cam with telephoto makes the most sense, wouldnt be too complex.. i'm surprised its not done already. smartphones have 2-3 camera setups for years now.
But they only use one lens at a time as far as I know
@@Tom-sg4iv Most of the high end ones already got multiple sensors video modes. Theres even one from Axon that lets you use all 5 simultaneously, including the selfie one.
They can Remove rear cam and put it inside of a front unit to make dual cam system. I guess people needed rear view camera.
@Phillip Banes - I remembered that I saw a vid showing that it's possible to record from two back lens on a phone (wide + telephoto) at a very good quality recently. I don't remember what phone it was, but I found this video made two years ago about an app doing it. ...tho, what I saw a few weeks ago been native in the phone (no need 3rd part app). I wish I could find what phone I saw exactly but here's what I found for now: th-cam.com/video/dsjRHhx9zb0/w-d-xo.html
Maybe the future is to ONLY have a telephoto camera, but have 4 units covering each corner of the car. Stitching the image together gives you an Ultrawide footage anyway. Except the image is sharper.
In fact, I'd be completely happy if Western Nations made it mandatory for all new cars to have this as default. It should make insurance claims, and really MANY issues very solvable thanks to unbiased footage. Old cars on the road can get them retro-fitted, but with time slowly they will be replaced with new cars that have these installed.
Cameras, Lenses, Processors, Storage, and Connectivity has become far cheaper, abundant, and better around 2017 and beyond... compared to a decade before that, and the decaded before that.
Dude nice work, this vid could very well change the industry. You did most of the leg work for the manufacturer's. If I was one come Tuesday this would be the new direction of the company.
To me, the real leg work is getting that solution to a price point that normal people would actually buy.
Not really legwork, ideas are easy, implementation is hard
Hmmm, providing prof of concepts testing off the shelf parts and putting it all in a well organized presentation such as a TH-cam vid seems like quite a bit of leg work. Implementation is of course gonna be difficult that's why the ball lands in the manufacturer court as it will require a team of people.
@@toddstratton9197 Proof of concept is usually the stage where someone actually integrates the parts together as a starting point for development of a real product.
Great idea. Could be an opportunity here for one of the dash cam manufacturers to put a dual front-facing camera out to test the market. As long as it's not too expensive it could potentially be a winner.
Linus Tech Tips covered a similar issue with dash cameras very recently. He showed even the more pricer cameras image quality was not much better than the sub $100 ones. The reasoning is they are all basically using the same System on a Chip, and those are mostly at least five year old technology. We are probably better off making our own with something like Raspberry Pis
Most current dashcams are still using processors that came out 5+ years ago. Typical story of racing to the bottom.
The tech is there, SoC with built-in high-res computational videography optimized for tough lighting conditions. There's just not enough market demand for higher end dashcams.
And that's why we need to get more people on board with it. Insurance companies should require them. Bigger market, higher demand, more units sold, higher budget and more competition equals better products
All dash cams should be required to capture license plates in several tests before they're allowed to be sold.
@@runed0s86 yes what we need is more regulation and bureaucracies because life isn't unaffordable enough already smh
Another great video. I left almost this exact idea in my comment on your last video 😀 If you want to do some additional testing most of the rear camera setups that I've tested are actually still utilizing standard analog security camera wiring (AHD/CVI/TVI). You could get an analog security camera with a telephoto lens and with a little splicing you could plug it into the rear camera input of most dashcams (the ones with the TRRS connectors would be the easiest).
As far as exposure on the plates, the tech you are looking for is called "HLC" or highlight compensation. Basically the algorithm works so that there are no clusters of pixels on the image that have a value of 255 and the camera will dial down exposure until that block goes down to 254 or less. HLC would need to be built into the dashcam though, so that would need to be a firmware change.
Another trick used is 'stacker'.. you can take sequential images and process them to somewhat sharpen the image a lot.
Yep, Google popularized this in their smartphones, and Samsung and Apple have copied them. It's actually brilliant. First you capture high framerate, underexposed video. Then you combine ~4 frames into 1, compensating for anything that moved. Now every pixel has more precise color/brightness, so you can crank up the brightness and colors without making the image noisy. Obviously you need a lot of RAM and a powerful ISP/DSP.
You don't need a great framerate to capture dashcam footage. Can we use stacking on already-captured video from current dashcams?
@@runed0s86 Yea, that's something I've already thought about several times while watching dashcam footage on TH-cam... you can stack multiple frames from existing videos to enhance the license plate and make it readable. It would take both stacking and adjusting position, zoom, and skew to line each frame up to the reference frame, but it does work. I've worked on image processing systems that have done the same for identifying images printed on a printing press.
What about taking the dash cam that has the optional plug-in rear facing camera and using an HDMI cord between the main camera unit and the optional camera so that you can turn and face that rear facing camera towards the front?
Was thinking the same. An HDMI Extender (HDMI Female to HDMI Male) would probably work. There are some on Amazon that are as small as 15 cm (just under 6 inches).
@@beardo-baggins I was also thinking the same.
For a higher performance LPR capture while on the move, we need a higher frame rate camera. Many of the purpose built LPR cameras see in the infrared, so headlights won't affect 'seeing' the license plates, and because plates are reflective, the cameras have IR emitters and are constantly sampling at different exposures (similar to photography and exposure bracketing).
The color/visible image is simply just for context; maybe the Axon dashcam/LPR cam relies on the visible image and does OCR against it.
Edit: jumped the gun, commented while I was still watching :)
Also check out DIY-ALPR, you can run an ALPR on a Raspberry Pi with a camera.
Use a short male to female HDMI extension cable?
BRILLIANT!!! I would have suggested taking the tiny camera apart, turning the connector over and reassembling the camera. Your suggestion is so much simpler!!!!
@@bernardowens1648 well, thank you very much. God bless.
Great work! I would hope this gets to the manufacturers because this is exactly what we need. You need to buy an HDMI reverse adapter lol. To answer your question though, I'd pay $200 for a 2k dashcam, then $100 - $150 for the license plate add on. That's if I knew it worked very well. It'd be nice to have options to add a rear onto that especially since 3 channel dash cams are out there. So yeah, $350 for a dashcam + license plate reader.
Awesome review. I like this review and hopefully someone comes out a dash camera like this. Thanks Vortex!
Honestly, I have been in 4 accidents with a dashcam. Only one, was a hit an run, at night, and in the rain. So, anecdotally, this was only needed 25% of the time for myself. As expected the plate was unrecognizable between headlight blow out and the rain on the windshield. In the other cases, the only sufficiently needed information was the make, model, color, and the circumstances of the collision. While I wish there was a consumer level product that could assist in these incidents. I'm not sure if the additional expense is really needed. That premium would really be a luxury. Especially considering I have only ever run sub $100 dash cams. I still run two Yi Dash cams today. They are going on 7/8 years old and still perform the same.
I can't fathom spending $400 on a dashcam that gives me the same story as a $60 cam does, the Mid 2000s white Ford Escape failed to stop resulting in a collision and drove away. Of course my eyes told me the same story; but, it was nice to provide to the Highway Patrol and the insurance company to prove it wasn't an at fault collision. For what ever that was worth to my insurance rate.
FOUR accidents ? your problem isn't dash cam quality .
@@siriosstar4789 you're right, its driving in Florida.
Thanks been trying to record racers on my street, now I know what to buy.
These past couple videos have been some of the most helpful videos you’ve done. I think you are really on to something with multi-lens dash cams facing forward. I saw your video on the GoPro, are you going to review the new GoPro 11 Heroes that are coming out that are supposed to have better cooling & even higher res + frame rate? P.S. You have to remember heavy mark-up for government agencies like police departments because of paperwork & compliance costs, plus limited market. Mass marketing can drive price down by harnessing a larger audience. You should try to Zoom interviews manufacturers & engineers. I’d be curious to hear what they have to say. I don’t care about size of camera that much so long as I can mount it high on my windshield.
Apparently there needs to be someone that can produce a gimble mount IR laser to lock on to police license plate readers and saturate the CCD.
Get on this one now Uniden!!
How about a bright-AF IR wide angle LED mounted next to your plate. Just over expose the crap out of the image.
You can get covers for your plates that show your plate fine when looked at straight on but blurs your plate when viewed at an angle. Technically, they are illegal in many states because there are blanket laws about putting covers on your plates at all but since you can see the plate fine when viewed from head on...if you were pulled over, it looks normal from a cop behind you.
@@KG4JYS A police officer with a laser is doing that exact same thing as he or she sits on the highway detecting vehicle speed.
Ir lasers that would be powerful enough to do that would blind someone instantly. Pointing an invisible laser around Is not only illegal but immoral.
@@MrAgentEcho A) I never said anything about pointing it at anyone.
B) Since you seem to know what the power output from an IR laser to distort the capability of a CCD is, I would be curious to know what that would be, are we talking micro watt or straight watts of energy?
There are lasers in the visible spectrum that are used as cat toys, so I am curious to know what power rating is required?
Loved this video.. obviously it's not geared towards the masses but definitely is something that enthusiasts are interested in.
I’d definitely buy one!
Yeah, the Motorola (really Vigilant) ALPR solution really is the best in the industry for license plate capture. I've seen the mobile one installed on patrol vehicles as well as the stationary ones they use for the radar trailers. They all run Linux under the hood and have a passively-cooled "VLP" box the cameras connect to for video processing. You're right that they do pulse the IR floodlights in low light to capture reflective license plates between regular video frames. That's why they've got a regular RGB camera and an IR-bandpass one.
Coming from digital cinema cameras, the outboard processing computer may have more to do with the clarity than one might think. Modern digicine cameras offer raw footage recording capabilities to be processed later, generally this is preferred unless storage or processing time (budget) is an issue.
These Linux boxes might not seem like much, but compared to a dinky processor (even a dedicated imaging chip) in a dashcam, they are likely far better at rendering plates in sharp detail.
I think the dashcam market is going to be stuck in compression town inevitably, but perhaps companies can devise some software to separately capture or parallel process the plate camera footage with dedicated algorithms like your hfr multi exposure examples.
I've had some experience working with dual camera setups while researching for a project. Turns out you can use a Raspberry Pi computer and a set of specialty board that can allow it to use two cameras at once, either for stereo vision, or for extended fields of view; or in this case, for a much more detailed view off a section of the main (wide angle) image.
These can be loaded with quite advanced IMX477 image sensors, and use separate lenses (say, 140 degree fov as wide angle, 20 degree fov for zoom in); and use the computer to process the data of both and deliver near-real-time enhanced imagery.
A SBC and DIY camera setup sounds like it's going to be the only surefire way to get a no-compromises system at a reasonable price.
Is there a discord / website /whatever that has more details? Looking to mount ALPR for my street. Thanks!
I believe this solution proposal has not being developed just yet. It might be worth to start a github of some sort if people are interested enough.
Been thinking about this problem myself. I'm an embedded software/hardware engineer and don't think it would be hard to have a 4 camera system. X2 facing front and x2 facing rear. The x2 cameras can be a wide angle stitched with a telephoto. Maybe a good idea for a kick starter? Could also incorporate some license plate reading and saving to a cloud where the data can be mined for when, for example, an amber alert is put out.
An external IR blaster might help with the issues with the internal cabin cameras with IR modes.
THANK YOU! I know optics and math and resolutions, and after trying and returning two high end dashcams, I figured these wide angle cameras just can't resolve plates. You've given me the first confirmation I was right on this.
One novel solution might be to have two front facing cameras, each with a 60-70 degree field of view, each barely overlapping in the center, and letting software stitch them together. Effectively doubling the resolution.
This exactly. Instead of relying on a single lens solution we could get much better results by using multi lens designs.
Note: LE ought to need a reason to believe the person-in-question is *actually* driving the vehicle *before* initiating a stop. Otherwise, its tantamount to pulling vehicles over to examine their 'papers', which is expressly forbidden according the SCOTUS. At the very least, if an officer does initiate a stop, they must abandon it as soon as they determine its not the person they're interested in. As we all know, however, LEO's will do it anyway until they're penalized for doing so.
how can cops know before hand that the person they're interested in is indeed the one driving if they cannot stop to verify?
just doesn't make any sense or am i missing something here?
@@lordjaashin You're close but you're missing his point.
The police still cannot legally pull The vehicle over because they believe that the registered owner could be the one driving. However with how convoluted and overly complex driving laws have become over time they can just pull you over for any number of reasons if they follow you long enough... And that's what they usually do.
If they do pull you over simply because you popped on the plate reader, then you could possibly get it thrown out but they would have to testify that's why they told you over and most of them are not going to do that.
Phone cameras can do face or eye detection. You could do something similar with License Plates which could auto adjust exposure for that region. That tech exists today. I agree, two lenses would be ideal. The second camera wouldn't even have to take video, just photos with timestamps, it could even integrate it into the main video with a little intelligent mixing. There are a lot of options.
Hi
Nice video I own a company that has been manufacturing ANPR or as you call them ALPR cameras for car since 1997 we have supplied the police and the armed services in the U.K we also specialised in capturing farsi (arabic licence plates) we looked at using dash cams for in car use and found that the Chinese made dash cams with "starlight" features which would capture licence plates in the dark very well. You may need to look into lens filters as well. However we had to modify the firmware in the cameras a little to capture cars at speed. I look forward to your next video.
I also think 30fps in basically unacceptable and unusable for daytime footage, at night it should automatically switch to lower framerate for maximum light, unless headlights are on turn go back to 60fps…
The real thing that matters here is shutter time. Shorter time decreases motion blur at a cost of more noise (higher ISO).
@@gblargg good point, and it could be overcome by a wider aperture
@@johnwilliams1091 but then a wider aperture decreases the depth of field, meaning the image would be out of focus for most scenarios, defeating the object of capturing a legible plate image
I think manufacturers would sacrifice video quality (lower bit rate) to make it 60fps. I think that's kinda similar issue in those 4K dash cams. (One of many reasons why you get garbage 4k video quality )
Use a RASA-type telescope to gather more light. A solitary imaging chip would have to be at the front of this telescope. It is capable f#s of 2.0.
Very nice work. As a lil tip, you coulda try using a HDMI flexible female to female extension cable, paired with a mcable and you'd have a better clarity, "native" 1080p upsampled with the possibility of having both cameras facing forward. Of course, theres quite a bunch cables to organize plus gotta also mount properly the camera, but you'd have pretty much what you were hoping to get from this video.
Good video. I did not know that the regular dash cams were so poor in capturing license plate numbers. The license number is critical information and I regard this as a key requirement of any dash cam system that I would consider buying and I would be willing to pay a premium for such a capability.
UltraDash Z3 actually has the dual cams feature for $150...one wide angle and the other telephoto. Please review for the channel and keep up the great work!
It does! Someone else just mentioned it too. I got excited and bought one, lol. :) Time to check it out!
Dear ALP manufacturers, hook our guy up please
Excellent video(s) on this subject. A couple of things…
You mentioned aperture… the trade-offs here are for lenses with wider-open “maximum” apertures (smaller f-numbers) the lenses must be larger, the cost increases, and depth-of-field becomes shallow.
Another optical consideration is the lenses formula and the materials used in the lens elements. Edge-to-edge sharpness, or lack thereof, might be negatively impacting some cameras’ ability to capture legible license plate numbers towed the edges of the frame.
Perhaps a RAW video format might be useful in one camera of a multi camera set-up to allow the adjustment of exposure to isolate license numbers.
Again, I found and enjoyed your videos today and subbed!
Another option, 2nd cam would be a 240 fps (preferably) Black & White cam with telephoto, although a slower fps would allow for higher res pictures. LPR cams are usually B&W and are proven to work. The characteristic of B&W cameras are more suited to License Plate Readers.
I agree. For the sake of "Affordability", the LP reader shouldn't require Color. Only the image quality.
Commercial vehicles such as 18 wheelers would love to have this technology.
*Simple solution* :- Go with the Nextbase dashcam at 5m45s and just connect the small Telephoto-Cam with a *USB EXTENSION CABLE* !!! ... and locate it where you want with a windshield suction-cup mount, with either an adjustable holder or magnetic type using a small metal disc/washer stuck to back of the telephoto cam.
If people are already willing to pay upwards of $850 for an Escort Redline 360C, I’m sure there’s a market for more expensive dash cams. Not to mention custom installed solutions like Radenso RCM that are already over $2K. However, if dash cam companies decide that they don’t think that these can become a mass market option, I could see them upping the price to compensate. So they may just end up doing the equivalent of radar detector companies of the single front camera being the regular(windshield mount detector) and the premium 2 front camera model(custom installed).
Not sure what the obsession is with capturing plates. I just want a dash cam that will capture the event so the other party can’t lie about what happened. If you need a plate and you have Audio recording on your dash cam, just read it out load so it’s included in the footage.
Very hard to do if you are part of a collision caused by an oncoming car just crossing the center line into your path.
@@markbajek2541 a plate would be the last thing I would be worried about in a head on collision. Both vehicles would most likely be disabled and hopefully everyone survives.
Human Error. In the heat of the moment, you could mistake a Zero for an O or a D. Same could be said about I or 1, G or 6.
When capturing a license plate you don’t need to worry about any of that and tend to injuries or getting off the main road.
Not a good call if your not knocked or hit you and drive off before you or the cam could see the plates you'd be screwed
Capturing plates is not most important thing.in most cases dashcam is using to find out how accident happened. I rather spend money on reliability . In hot climate, long parking outdoor, dashcam is too weak to survive.
I feel computer learning and better SOC systems are going to play a huge role in this in the next 10 years. Along with better camera hardware as it should be doable to have AI do on the fly exposure and post processing activities not on the Dash cam but on big iron servers that can process what was captured. The key is to have a a crap ton FPS, and multiple sources from multiple lenses. Then its a matter of uploading and let computer learning do its thing. That said, no light is no light. There is only so much AI can do to bring out details. But I will be honest some of the stupidly blurry images I've used on publicly available tools to bring out image details borders on voodoo magic. In 10 years? Sheet. This stuff be easily integrated with dash cams assuming manufacturers feel the need to push the edge on what can be done.
My smart security camera can recognize people cars and dogs, and then track them, adjusting the orientation and follow them. It wouldn't be a huge reach to track rectangles with the ratio of a license plate, then apply zoom and exposure changes. putting this into usable data is merely code. We want to export into a document , Car make model year and color, and of course the license plate number in the form of a .csv file with spreadsheet compatibility. the problem with my 8mp smart security cam though, it sucks at capturing motion clearly.
I think a lot could be done in post processing: there is plenty of ways to stack the multiple still frames from the video in order to get an enhanced image. Astrophotography leads the way in this field.
Other esoteric solutions may involve free-form optics that give you a higher focal length in the centre of the sensor, and "squish" the rest of the wide field of view to the outside area of the sensor.
Another solution would be using monochrome sensor, which would give you better resolution, better dynamic range and better low light performances; c-mos sensors could, in theory, shot with two different exposures at the same time, at the cost of having about 0.7 the native resolution.
One thing I haven't seen mentioned yet is using aspheric lenses. They're typically used to subtly correct optics, like focal distances for glasses or correcting chromatic aberration in compound camera / telescope systems, but there is nothing preventing their manufacture to give telephoto-like capabilities toward the central region with a smooth transition to a wide angle-like toward the edges of the the image circle. It would require additional optics or post-processing to deal with not just further chromatic aberration but also make it sensible to view in software since otherwise you would have a very fisheye-like (-like; they are very different optically) view of the image. All in all, the custom manufacture and R&D required is likely more expensive than simply placing a second camera module, but thought I'd throw it out there.
Modular dash cam? Single Dash Cam (main product), attachment cam on both sides of the main cam. One for Telephoto front cam and one for interior cam. And a cable port for rear cam. That way people can buy what they want and you mainly push the single front camera and/or front and back kit. Have the others as add ones for people who want them.
Like the Nextbase design?
So, if we were willing to spend the money now, is there a recommendation for a second (dedicated) camera, available on the consumer market, to specifically capture license plates?
7:04 according what i heard about us plate chaos with different designs and plates the the same text but different design (even in the same state) being able to be given to different ppl the telelens cam is a lot better for specifically getting that.
Several things came to mind.
12:45 - You would have to have a separate, tiny IR light with a flexible angle-cut lens that conforms to the windshield glass. The light will shine out, but stop the light from reflecting in the interior. Probably too much work, but it would fix that glare problem.
19:19 - You could buy (if they make them) a 1/2 inch or 1 inch MALE TO FEMALE cord/connector and simple twist the inner facing cam to face out the same as the main cam.
14:40 to 15:06 - Since ALPR and speed cameras need a reflective surface, I can use some cover or chemical and remove that reflective coating, rendering the systems useless, lol. evil grin
Covering the license plate is illegal in places. Steve Lehto from the Lehto’s Law TH-cam channel recently had a video where the plate was only covered around the edges by a decorative frame, and IIRC there was a citation issued and it was upheld in court.
@Commercial Guy Last point you made... ' And get pulled over immediately for altered tag '. There's a law against doing anything that can hinder a cop's ability to do his job. Not worth it.
Sounds to me that getting the data is the most important. The rest is software.
I think that one 2k camera can do the job. Run it in HDR mode at variable frame rate. We're not shooting blockbuster movies here.
Store each level of exposure into its own video file rather than merge the HDR frames into a single frame.
Basically, stop thinking of a dashcam as a video camera where you want spacial coherence but think of it as a continuous fast photo camera taking many photos at a wider range of settings.
As a video it will be ugly to watch but post processing should produce a result good enough for court.
If you want pretty pictures... Get a gopro and shut up.
You’ve just been describing a modern smartphone. Multi lens, IR/LiDAR, GPS, Motion detect, image stabilization, ML image enhancement , HDR, multi-stitching, WiFi/BT/5G, large storage, fast, processing, standby battery, large screen, ETC. If one came in a durable heat resistant body, $1000 sounds about right.
Oh my... I hope some of the big dash cam manufacturers sees this and start make some new design. Great work
love your dashcam videos, they truly are in a league of their own, no other channel comes even remotely close
always so impressed by the quality of every single one of your videos
I would totally be onboard with a dual front facing setup wide and telephoto. Blackvue with 2 cameras on opposite sides facing forward.
Not only is it important to me since the current setups cannot deal well with plates, especially moving ones, the cost of the current ones isn't relevant to me as I wouldn't purchase any at the moment. If you get hit and the jerk takes off what good is it to have a dash cam if it cannot resolve the plate?
Well shit, it does not do Oakland California pd any good. to have the cameras .You know why ? because hell most of the cars don't have plates to read. It is beyond me why they don't get stopped . Yesterday, I saw a car with no front windshield and expired tag as far back as 1989. This car was in downtown Oakland 2 blocks from the Pd station it's comical.
uh ... test it with an HDMI cable? That will let you flip it around for $20 for testing purposes.
Great video man thank you you’re helping push things forward
I think this technology could be easily retrofitted to existing dual camera setups. If say Viofo modified their rear facing camera to function better as a front facing license plate focused one then it would be instantly viable. Anyone could just plug it in.
I would definitely like a second telephoto camera for license plates. Two cameras (wide & narrow) at the front, two at the rear (ditto), and one interior, for five channels in all.
One method that I do to improve night time by using a second cheaper dash cam with the exposure comp settings set to their lowest setting. This allows the dash cam to use a higher shutter speed and lower ISO which improved license plate visibility as well as the distance at which they are visible.
Can someone give me the cameras they are buying after watching this video. Thank you
Dual lens dashcam with an LED IR module in the grill might work well...
UltraDash Z3 Standard Edition - Dual Lens Dash Cam HD 1080P, Telephoto Zoom-in Lens do that
Oh hey cool, thanks for the heads up! I just bought one and will check it out. 👍
Just get a short HDBMI extension cable and rotate the camera. make a simple mount and there you go.
you could just use a male/female HDMI cable extension in between your two cameras, that would allow you to mount the "rear" camera in a better position on the window for license plate capture. you could use the same to mount your IR camera externally so you are not getting IR reflection off of your windshield.
Hi I recall a recent video that you published like 2 months ago with the external cameras mounted on the two front side doors to show blind spots, these cameras are outside the car and will most likely take care of the front windshield glare you mentioned for the night UV vision. Just and idea maybe this can be the solution to night glare from windshield, to have a small camer installed outside perhaps in the grill area for weather protection and better visibility
I'm pretty sure you live close to me. I recognized the sheriff vehicles and most of the roads where the test footage was recorded. I like the video.
Well done sir. Always your insights are useful. Watching from Uganda 🇺🇬 East Africa
I want a camera to capture license plates. The market is already flooded with general dash cameras.
In too many states, there is no front licrnse plate. Do you can't read a plate behind you.
Great detailed talk.
There are tons of dash cam reviewers on TH-cam but you are the most detailed reviewer. Kudos to you.
Today I just decided to grab the Tinkware U1000 after much research. Needless to say I am not a fan of having to pay a premium price for a 4K dash cam. But I want a very reliable product while not draining and damaging my car's battery for the long term especially when the cam is operating under parking mode. Anyway, I digress a bit...
I think nowadays since folks have absolutely zero qualms about shelling out a cool $1000 just to own an iPhone, we have to ask ourselves how much we are willing to pay for a dash cam that can capture license plates under most conditions especially when accidents occurred?
What will the price be when we have to deal with insurance companies arguing back and forth about accident claims? Personally I think if it is at or near a $1000 and no more it will be a fair price to pay for the added important dash cam feature.
I believe most dash cam manufacturers are lazy. If they are making tons of money then they have no incentive to really innovate especially when we are just now, kind of, trying to get back to normal after the pandemic.
Lastly, I don't think this will be a huge technical challenge to put in a second telephoto len like the ones found in our smart phones for dash cams. What needs though is to spend some time and effort to implement AI on top of the hw so we can optimize various light conditions, object speeds, shutter & exposure settings for capturing the license plates. Of course nothing is perfect. AI most likely can help but it is not perfect. There is no fool proof magic bullet.
Maybe with the new Sony sensors on the horizon we can easily pair it with a low end, low cost, Nvidia graphic chip having optimized AI just for image processing, image tracking and nothing else. Together with the improved HWs we probably can make vast improvements in this area.
Have you tried the Viofo A119 V3? I think it is the best dashcam in the market as of now when it comes to capturing license plate zoomed in.
@@seekfactsnotfiction9056 Hi Hi. Thanks for the info.
Do you personally own the A119? Can you elaborate more why you feel this is the best option available for said task?
Thanks much!!!
Drive Safe. 🙂
So if you’re looking for police at night, get a strong IR sensor 😂
As you were discussing the telephoto aspects, I wondered if attaching one of the many cellphone "telephoto" add on lenses that are available to a standard dash camera.
A person would need to run two cameras then, but it might be a cheap and easy "mod" if you already have an older dash camera sitting around.
very good video compared to most dashcam videos i put your link on linus dash cam video comments
As an owner of a dr900x-2ch+, I think that $7-800 for the same system, with an additional license plate capture lense, would be my target price. Maybe 850 if they can prove to have really good low light performance too.
A dashcam can prove that the accident was not my fault, but it's still my insurance premiums at risk unless we can determine who's insurance the claim should be submitted to
In that low light condition, the camera with HDR support can do better compared to the telephoto lens. I also have done some testing with the telephoto lens, the challenge is the same as the video showed, it is still hard to get a good result for night recording. Most of dashcam Infrared LEDs are not so powerful, I have tested night recording with more than 20pcs IR LEDs, it doesn't help a lot to improve the brightness, but the LEDs PCB is overheating.
What are your thoughts on having the second lens focused on freezing motion in the daytime and exposing for bright plates at night? I like the idea of HDR, but I haven’t had good luck with using it to capture plates at night, especially while moving.
I have always been curious why post processing software doesn't exists that does image processing on PC rather than camera and create enhanced pictures from video.
Curious if the ir from a laser jammer could be harnessed as external ir headlights for helping with license plate capture ( with firmware and maybe some form of pairing between camera and laser systems
You can make a better test using a HDMI Down Right Angled 90 Degree Adapter and mount the camera in the write place.
hope you do a video with the proper installation.
Please reference the multiple followup videos linked in the description, particularly part 3. ;)
what about the UltraDash Z3? that has a telephoto lens that can rotated to either front or rear. they have been out for some time now
Just got one. 😉 instagram.com/p/CiNmoafrfsQ/?igshid=YmMyMTA2M2Y=
Working with Google and this subject has been tested by using AI with great results on regular lens camera
Ah cool. Got a link to the testing and results?
You’re not doing the ANPR in real-time are you? If so, how? If not, why do it later from the recording?
Extremely high refresh rate would be best need like a 120 shutterspeed with a very large aperture and a sensor that has very low noise at 800 iso. Pretty much these repurposed cell phone image sensors aren’t going to cut it
10:28 During this slow-mo, look how one frame around 10:35 is much clearer. Video compression is making things worse, and the clear frame is a key frame where it sends the whole frame fresh. Memory cards are so cheap now, they need to offer the ability to really push up the video compression quality.
I'd love an lpr addition.
Just wonder if an active filter that can move between the lenses to block most of the light and basically keep just the (near) IR at night. How well would that work?
It's not really made to be a dash cam first and foremost but the comma 3 from comma ai (which is designed to be a after market driver assistant system, hands free driving and all that) has a smaller FOV camera and a large FOV camera.
It's a 2,000 solution though so it doesn't really work as a dash cam by itself unless you could use the other features
It is getting HDR support so hopefully license plate reading gets even better.
@Vortex Radar, I wonder if you could open the the rear camera and reverse the connector so you can plug it in "facing forwards"
On board with the dual facing cameras. It could be two separate video files just how VIOFO does with their dual f+r camera.
Ideally I would prefer 4 cameras. Two front, two rear.
An external IR array accessory could be mounted behind the grill inconspicuously to help with low light parking conditions.
You can get hdmi extension cables that would work with that add on camera and you could flip it and just need a way to mount it.
you missed one very important thing. It's optical performance. if plate is made out of 6 pixels, with a diffraction limited lens you can read it without a problem. problem is that optics are sooooooo bad they can't use all sensor performance, so it does not matter 4k or 1080p, quality would be the same. Test Sony RX0 II, it has 6 aspheric elements in objective
Makes me appreciate the included cameras in the tesla. The plates and cars are easily identified.
Teslas have the same issues.. plenty of reports on forums about plates garbled or not viewable, particularly at night.
And not to mention issues like blinker light bleed into repeater cameras in early 3s
i have always wondered if it’s possible to use an object detection algorithm to spot police cars. and if it would be more effective than traditional radar detection alarms
About the infrared, there's the reason license plate cameras are mounted outside, glass reflects infrared, to the point temperature cameras have a ton of trouble seeing through it.
This is great research, I would not be surprised if next year you see some dual lens dash cams hit the market.
I run a 3-channel front, interior, and rear set up currently, but I could see myself getting a 2nd camera for telephoto front and telephoto rear setup. Capturing plates is very important.
Oh hey, that intersection at 20s mark in video, that's 2 miles from my house!
That moment when "your doing it wrong". Grab a cheap security camera, the ones that have night vision. then point that at a license plate. the background will reflect IR light like a lot but the letters will not. then you read the black space not reflected.
This is why the cop ones have a bunch of leds in it. there IR leds. the computer software just gets a black and white image from the IR camera.
Stereo cameras possible combination.
48MP wide + 2MP telephoto. At night the 2MP telephoto can keep its higher shutter speed thanks to it's bigger pixels. The 48MP would keep the overall wide view and help the telephoto with HDR stacking.
48MP left + 48MP right. Like a 360 camera but in a 120° view. Much more resolution for a smaller FOV and quad Bayer filtering for lowlight.
That DR900X Plus uses h264, low bitrate, most of the non keyframes are just made up pixels...h265 and intra would do wonders for frame to frame clarity. That DR900X Plus also has HDR always on, and extreme noise reduction always applied. Those are the main reasons even high shutter speeds in the day can't save the newest Blackvues. Oh yeah, can't turn the gain/brightness down on the plus units, so blowout not controlable in any way now.
That's because all dash cams are legitimately the same using hardware from the same 2-3 companies which are using technology from nearly a decade ago. If a company wanted to actually do some r&d work and not use the same off the shelf components that every other dash camera company is using they could easily dominate the market!!
LOL. You did a better job than the manufacturers!
I think people will pay more money if the option exists. I have plenty of friends, most making less money than me, that go out and spend $1200 on an iPhone or flagship Android every two years without even batting an eye. In fact, sometimes, the more expensive it is, the more demand there is. Not for me, but for many people. Funny thing is, I've been looking at borescopes lately and they aren't very expensive yet some of them have THREE cameras. Although, I'm guessing those cameras aren't nearly as expensive. Anyway, good job on the video.
Dashcams would be so much more useful if they could pick out the license plates and save them to a text searchable database alongside the images they grabbed them from. You might have passed that car that hit you a week ago, but would you remember it? No. That database file would though. Even if it mistakenly put a I for a 1 or a L it would be an obvious false positive.
What you need is a iPhone or mobile phone with multiple rear cameras (front dash cam) one that’s wide HDR good in low light, one telephoto HDR, and one front facing cam for in car or rear view. Plus it has all the sensors your want for g forces, has LTE connectivity, WiFi, Bluetooth, live streaming, 500GB of storage, 4K for all cameras, uses hardware to encode hvac / h265 codec. Has a large built in battery for park recording. If you took out the screen from a phone, you could make the dash cam pretty small. About the size of a black view if you left out the big battery. Ideally I’d take the camera bits out of a phone, make a 3D printed housing for them that’s tiny. Run a extension to the main board in the phone. Put the phone in a box in the glovebox. Make a custom app that could record from all cams at the same time and had all the features you’d want. It would be mostly plugging into apis that already exist. Could easily use it to instantly upload video clips to a server. Could use as a in car hot spot. Use it with CarPlay. That would be ideal. I bet you could make a camera module and plug it into CarPlay via a lightning y adapter and have your regular phone be the recorder. Have it input metadata from the car into recordings via CarPlay. Or just make a app that acts as a remote for the dashcam phone that works in CarPlay that manages start stop, lets you see what the cameras are seeing, configure settings. You could open the remote app using your regular phone that’s connected to CarPlay and see love views from the cameras including the plate reader, and allow you yo review old recordings.
One step further, use apples image recognition in the app to identify objects (red car, lady holding monkey) and use their image to text functionality for the license plate. So you’d just have to program the app to use all these built in features. Take snapshot. Save raw image. Hdr. Send to apple api for OCR, load back into app add text to list that is in a plate format. You wouldn’t have to develop or adapt any existing programming libraries that did all that. You could sell the software the app and accessories like holders, camera pod for adding remote cameras so they don’t have to have their phone in the windshield. But that’d work. Making it just the app. Phone mounted on windshield. And use CarPlay for control and still maintain full use of the phone to play music answer calls. Use Siri. Phones from the last few years have the processing power tye gpu power to do all this. I’m just curious if apples sdk let’s developers access more than one camera at the same time, and if the lightning port would be capable of supporting external cameras. Or would you need a external box to encode them and then send to the phone.
Or heck just have a small DVR brain be mounted somewhere hidden and have cameras mounted up including a 3 in one camera.
I don’t know why there aren’t more dash cams that work like a DVR or NVR. I had one several years ago that did this. Was great. Just had a tiny camera in the front and back and the recording processing box went in my glovebox which also had a AC vent so it kept the thing cool as well. I hate having a big ugly camera on the windshield. Also having the DVR in a locked glove box makes it a lot harder to be stolen. Heck these days if people can see anything in your car that might be of value, like a expensive camera sitting in the windshield they’re gonnna smash the window and take it.
Speaking of, also if your dashcam iphone got stolen, you could track it, and remotely wipe it and disable it from being used again. Even track it via BTLE when the battery dies or power gets shut off.
I mounted an IR light outside to eliminate the glare from the windshield and disabled the ones on the unit itself.
I was just tinkering around with a cheapy dash-cam to see if I could improve it at all, while running my trusty Viofo to do the real work.
It actually improved them both, and the Viofo doesn't even have any IR lights built in, so I think the sensor is just able to use the additional light even though the unit isn't designed for it.
If someone is serious about solving this problem (consumer or manufacturer), I think it's an easy decision to make. We're all getting out of our "comfort zones" or at least moving beyond plug-n-play when we start hard-wiring these things in to clean it up, semi-permanently mounting them instead of using suction cups, setting up apps and VLC players to decode, transfer and remote save files. . . . so, popping the hood and bolting a little box behind the grill should still be considered a DIY-able install for most people, as long as they can follow wiring instructions.
Sure, it might be tough for manufacturers to green light something that makes the casual owner quickly flip the page to an easier option, but if the right marketing was in place, highlighting a clearly superior product, I know it would sell. (at least one)
I have a license plate reader camera at my house and it only cost me about $250 bucks, no need to spend thousands of dollars. Most decent security's its possible, you just have to change the shutter speed. Also need to move the camera close to the street at a good angle or if you wanted it mounted to your house you need one that has a telephoto lens.