Cool project. I have some performance tips for you or others which may or may not be useful. Scaling a camera feed down to 640 is resource intensive, try to get a feed that's as close to 640x640 (or whatever you're using) as the source. Don't try to go as fast as possible with the detection, this will lock up your resources. Find what your resources are capable of and then detect on a schedule 30% less than capable (pick one frame every x frames). Instead of displaying all of the results to the screen, just render the original video and manually overlay the detected positive bounding boxes once every x frames (video running behind and a transparent overlay with bounding boxes in front) 2 html5 canvas's will make this easy. Once your resources are freed up a bit, move to a higher resolution for more accurate detections. I don't know how many plates you have but if it's like 10k or 100k, you can squeeze a little performance out by splitting those into ~36 separate lists/hash tables based on the first letter/number. At the end you'll have video at whatever speed you like and overlays like once per second - if you're driving and detecting the car in front the overlays won't move too far from the targets - well, unless you're getting chased but then detection is not so important anyway!!
Could we just talk about the plane coming right at you at 8:42. The A2's spaceframe should be able to go into orbit, a plane is no match for the mighty A2!
woaaahh as a computer science student that drives, i had the exact same idea!!!! it's really cool to see that someone has done that. Well done! suggestion: How about adding a option for voice recognition ? It would be easier/safer when driving to just hold a button and read out the plate maybe?
Idea is great. Probably not effective in actual usage, since if you break the rules, the plate scanner wont be able to scan from multiple kilometers :D but as a fun project this is great!
Hello friend, if they use tetra radios in your country i believe its possible to detect them with an rtl-sdr. They beacon their position. That could be a second line of detection? For countries where there isnt a public list i think undercover vehicles registeation database could be crowd sourced if the app allowed reporting suspicion.
What a good first project, well done 👍. I would like to add though that It's quite concerning that a commercial license plate reader (as opposed to a camera) could be used to identify police vehicles within your entire line of sight. This also raises the question of what other (sensitive) information could potentially be accessed and used maliciously by bad actors. In England, our Freedom of Information Act includes many exceptions, particularly when it comes to information related to law enforcement and national security, which we are not allowed to access.
You can see that it is a Police car, not who the police are, or where they live. It is not that super hard to spot undercover cop cars in Sweden cause everyone that I have seen have had those blue lights mounted under a black tinted rear window and also they usually have 2 extra mounted white lights at the bottom of the bumper.
Jag tänkte genuint på denna idén igår. Hade varit jättekul om det gick att ha en kamera fram och bak, som skickade information till en simpel dator som indikerade i bilen ifall det var en civilare nära. Det kanske är exakt vad du har gjort, har inte kollat videon ännu. Men sjukt att jag sa detta till min flickvän igår
That curve from Mantorp Park do I recognise lol. Iaf, eftersom texten (edit: texten = typsnittet) på nummerplåtar är standardiserad, så borde du kunna optimera din OCR ganska rejält antar jag?
![Путин и ПЕСКОВ ждут звонка от ТРАМПА 😁 [Пародия]](http://i.ytimg.com/vi/sm8iV0SxRHk/mqdefault.jpg)
8:41 - WHERE did that plane come from XD
That's Linköping, Swedish aviation capital/SAAB aviation capital, with monuments around the highway.
Cool project. I have some performance tips for you or others which may or may not be useful.
Scaling a camera feed down to 640 is resource intensive, try to get a feed that's as close to 640x640 (or whatever you're using) as the source. Don't try to go as fast as possible with the detection, this will lock up your resources. Find what your resources are capable of and then detect on a schedule 30% less than capable (pick one frame every x frames). Instead of displaying all of the results to the screen, just render the original video and manually overlay the detected positive bounding boxes once every x frames (video running behind and a transparent overlay with bounding boxes in front) 2 html5 canvas's will make this easy.
Once your resources are freed up a bit, move to a higher resolution for more accurate detections.
I don't know how many plates you have but if it's like 10k or 100k, you can squeeze a little performance out by splitting those into ~36 separate lists/hash tables based on the first letter/number.
At the end you'll have video at whatever speed you like and overlays like once per second - if you're driving and detecting the car in front the overlays won't move too far from the targets - well, unless you're getting chased but then detection is not so important anyway!!
Hej! Blev mycket intresserad av ditt projekt. Hur kommer jag i kontakt med dig?
@@REVMAD email antar jag, finns under "more" på kanalhemsidan
2:18 I love the fireworks in your hand :D
Could we just talk about the plane coming right at you at 8:42.
The A2's spaceframe should be able to go into orbit, a plane is no match for the mighty A2!
looks like 7:43? Guessing a static display of some sort.
@ehb403 Good eye!
Congrats on the new place!
I don't think you can do this in many other countries, if this existed in the US someone would be stalked and murdered
Hahaha, I think the dmv had public records before, but someone stalked their ex from their car and killed her
@@ShortShortsfromreddithahaha, very funny!!
@@nsglcck thats the actuall story though...
woaaahh as a computer science student that drives, i had the exact same idea!!!! it's really cool to see that someone has done that. Well done! suggestion: How about adding a option for voice recognition ? It would be easier/safer when driving to just hold a button and read out the plate maybe?
We have built a facial recognition database of police.
Awesome content, I love how much effort u put into it, this is a sick project
Very cool, I usually just know it's an undercover b/c 1. their tint is insane, and 2. b/c of all the extra gadgets they have.
Yo mate that desk is really clean, love that you can keep your workspace so clean man.
Nice video dude! I litterly had the same idea about a month ago, this shows me that it isn't impossible.
Idea is great. Probably not effective in actual usage, since if you break the rules, the plate scanner wont be able to scan from multiple kilometers :D but as a fun project this is great!
Great Vid. A little GPU offloading will go a long way.
Add Australian plates
I second this
Netherlands please ^^
this would work in in norway and finland as well!
Hello friend, if they use tetra radios in your country i believe its possible to detect them with an rtl-sdr. They beacon their position. That could be a second line of detection? For countries where there isnt a public list i think undercover vehicles registeation database could be crowd sourced if the app allowed reporting suspicion.
I don't care how, but you need to add US plates
maybe using a usb coral tpu by google could help speed it up without the need for new laptop
What a good first project, well done 👍. I would like to add though that It's quite concerning that a commercial license plate reader (as opposed to a camera) could be used to identify police vehicles within your entire line of sight. This also raises the question of what other (sensitive) information could potentially be accessed and used maliciously by bad actors. In England, our Freedom of Information Act includes many exceptions, particularly when it comes to information related to law enforcement and national security, which we are not allowed to access.
You can see that it is a Police car, not who the police are, or where they live. It is not that super hard to spot undercover cop cars in Sweden cause everyone that I have seen have had those blue lights mounted under a black tinted rear window and also they usually have 2 extra mounted white lights at the bottom of the bumper.
Jag tänkte genuint på denna idén igår. Hade varit jättekul om det gick att ha en kamera fram och bak, som skickade information till en simpel dator som indikerade i bilen ifall det var en civilare nära. Det kanske är exakt vad du har gjort, har inte kollat videon ännu. Men sjukt att jag sa detta till min flickvän igår
great vid! how did you obtain the list and from where? do you update it?
That curve from Mantorp Park do I recognise lol. Iaf, eftersom texten (edit: texten = typsnittet) på nummerplåtar är standardiserad, så borde du kunna optimera din OCR ganska rejält antar jag?
Could Norway be easy (Sorry for the 3 different comments lol)
Great for engagement, I'll take a look at Norway
Heja dej! Android app, skulle det vara möjligt att skicka datorn alls då kanske?
Nice! Is the Netherlands included?
I commented this before watching😭 but this video is gonna blow up
great content
Can you do the uk please
Police Number Plates are not public information in the UK
By the time you get that close to scan the plate you should be able to see the lights inside the rear window.
Are you trying to feed all the frames from the camera directly, and not limit the sampling size? Edit: oh, you have a dropdown in the rewritten one.
im going to be honest, idk what i did
Where are you from?
Maybe Sweden?
Va detta ville jag göra. Men nu slipper jag ;)
Need some GPU accel
UK plates!!!!