Bruce, I wanted to say I am living vicariously through your videos. There are things I could attempt to copy, and things way above my skill level. Keep up the great work making the videos and realize that you're probably the most techy renaissance man I've ever seen.
Can you access your node-red remotely ? If not, would node-red inside a cloud-server be able to work the same way. I am just trying to do first steps getting some esp32 camera modules to node-red for video if fast enough or snap shots at some interval. Really I am looking at node red as more a iot type of platform/website that I can access remotely. I am looking at cloud hosting because I don't want to be responsible for security that's not my bag.
Depends on your router. The camera snapshots are simple web pages, so would open them up for remote viewing the same way you would any web page, not really a node red issue. These days, rather than opening up a port (e.g. web server on port 80 or 81), the easiest/safest approach is to use VPN like TailScale.
I would really like it if you could share move of the details of the setup specifically around your detection and notifications and some of the Node Red functions that have the advanced false positive suppression.
I just exported that Cameras flow to here: github.com/brucewinter/myhouse/blob/master/nodered_cameras The main false positive suppression is done by only looking for a limited set of objects with a high confidence level (via the DOODs results). And ignore anything that happens when it is raining. Even with that, sometimes we get '90+% confidence it is a person' from shadows or branch movements. That's why I added the 'bbox_skip' function. It simply looks to see if we flagged the same object bounding box, within a fuzzy size range and position, too many times. Currently that 'skip if we saw that box too many times' count is reset daily. It would be more clever to decrement the count daily, rather than reset it to zero.
@@BruceWinter Thank you so much I'll review this and extract the parts I need. I'll probably have more questions later around your NodeRed dashboard of pictures if I can't figure out how you got all of that working. Are you just using Doods for the 90+ evaluation? The reason I ask is I have a PI3 and am concerned it won't have enough processing power to give a quick enough evaluation to be useful. Finally I have some 2k cameras that are only giving 50-60% accuracy for person. Any tips how to make this better?
@@jasoncrowe2663 I didn't know DOODs had a fee? When used via the HomeAssistant addon, it is free. My 4k cameras routinely give me 90+% confidence for person recognition using this configuration.yaml option: detector: inception
This is a great project, and I am trying to rebuild your project.May you provide information about the external Node you installed? This will help me a lot
Not sure what you mean by external Node. That node red code is pretty complex. Probably best to use it as a general idea on what can be done, rather than try to duplicate it exactly.
Hi Bruce, very good video but too complicated for me, I'm newby in this world. I have a couple of cameras integrated in Home assisntant with frigate and what i want to do basically is 2 functionallaties: - turn on lights when a camera tetects a person - send a telegram or home notification when camera detects a person and nobody is home. I have already setup this but as a Home assistant automatization, I would make it work with node-red could you show me simply example for this aims. Thanks in advance.
Hi Luis. If you are also using home assistant, here are a couple of examples that might be useful: github.com/blakeblackshear/frigate/discussions/940 th-cam.com/video/RWsT-x7yYXI/w-d-xo.html .
HA and NodeRed should run fine on a Pi 4. Plenty of people run it on Pi 3s. I use a 3 year old Intel NUC. The DOODs object recognition would probably not run very well on a Pi. I see about 0.5 seconds per image on my NUC.
Bruce, I wanted to say I am living vicariously through your videos. There are things I could attempt to copy, and things way above my skill level. Keep up the great work making the videos and realize that you're probably the most techy renaissance man I've ever seen.
Thanks for the encouragement Randy!
Wow!!! Impressive, I congratulate you this project is impressive 👏👏👏👏
That's very cool. I do javascript programing for work so its interesting seeing some uses of javascript outside of front-end web frameworks.
Yeah, Node Red is getting me to appreciate JavaScript more everyday.
Nice work, Bruce
Thanks Thomas!
Can you access your node-red remotely ? If not, would node-red inside a cloud-server be able to work the same way. I am just trying to do first steps getting some esp32 camera modules to node-red for video if fast enough or snap shots at some interval. Really I am looking at node red as more a iot type of platform/website that I can access remotely. I am looking at cloud hosting because I don't want to be responsible for security that's not my bag.
Depends on your router. The camera snapshots are simple web pages, so would open them up for remote viewing the same way you would any web page, not really a node red issue. These days, rather than opening up a port (e.g. web server on port 80 or 81), the easiest/safest approach is to use VPN like TailScale.
Thank you so much for the video, I don't know if it's possible to give it a shoot using a phone camera ?
Looks like it, but have not tried: www.home-assistant.io/integrations/android_ip_webcam/
I would really like it if you could share move of the details of the setup specifically around your detection and notifications and some of the Node Red functions that have the advanced false positive suppression.
I just exported that Cameras flow to here: github.com/brucewinter/myhouse/blob/master/nodered_cameras The main false positive suppression is done by only looking for a limited set of objects with a high confidence level (via the DOODs results). And ignore anything that happens when it is raining.
Even with that, sometimes we get '90+% confidence it is a person' from shadows or branch movements. That's why I added the 'bbox_skip' function. It simply looks to see if we flagged the same object bounding box, within a fuzzy size range and position, too many times. Currently that 'skip if we saw that box too many times' count is reset daily. It would be more clever to decrement the count daily, rather than reset it to zero.
@@BruceWinter Thank you so much I'll review this and extract the parts I need. I'll probably have more questions later around your NodeRed dashboard of pictures if I can't figure out how you got all of that working.
Are you just using Doods for the 90+ evaluation? The reason I ask is I have a PI3 and am concerned it won't have enough processing power to give a quick enough evaluation to be useful.
Finally I have some 2k cameras that are only giving 50-60% accuracy for person. Any tips how to make this better?
@@jasoncrowe2663 I didn't know DOODs had a fee? When used via the HomeAssistant addon, it is free. My 4k cameras routinely give me 90+% confidence for person recognition using this configuration.yaml option: detector: inception
This is a great project, and I am trying to rebuild your project.May you provide information about the external Node you installed?
This will help me a lot
Not sure what you mean by external Node. That node red code is pretty complex. Probably best to use it as a general idea on what can be done, rather than try to duplicate it exactly.
@@BruceWinter Alright, thanks a lot.it is really good project.Learn a lot Thank you.
Hi Bruce, very good video but too complicated for me, I'm newby in this world. I have a couple of cameras integrated in Home assisntant with frigate and what i want to do basically is 2 functionallaties:
- turn on lights when a camera tetects a person
- send a telegram or home notification when camera detects a person and nobody is home. I have already setup this but as a Home assistant automatization, I would make it work with node-red
could you show me simply example for this aims.
Thanks in advance.
Hi Luis. If you are also using home assistant, here are a couple of examples that might be useful: github.com/blakeblackshear/frigate/discussions/940 th-cam.com/video/RWsT-x7yYXI/w-d-xo.html .
how to node red rtsp camera to http camera converter?
Not sure how to do that in node red. go2rtc is a nice lightweight program that can convert between various formats.
Ohhh wow ...and I barely hook up some nodes in home assistant and when they work it takes forever to troubleshoot ;)
Love this! How much compute power are you using for OpenCV to run in real time? This looks way more than a Pi 4 could do.
HA and NodeRed should run fine on a Pi 4. Plenty of people run it on Pi 3s. I use a 3 year old Intel NUC. The DOODs object recognition would probably not run very well on a Pi. I see about 0.5 seconds per image on my NUC.
@@BruceWinter Thank you!
Thanks for the video =)
Awesome video...I want to do the same with my cctv cameras
Any help would be highly appreciated :)
Sure, let me know if you have any questions.