I was inspired by this TED talks about YOLO Object detection.
When I saw it I instantly thought about my security cameras that send me push notifications multiple times a day when a branch of a tree is caught by the wind or a fly/spider is too close to the camera. I started to look into trying YOLO.
Turns out it’s pretty easy getting started… More info on YOLO
The fork I ended up using: here
I use Unifi G3 cameras that films whenever motion is detected in the cameras. I wrote a Python script that tails the recording.log and whenever a new motion is finished the python script will run YOLO on the recording and if a person is detected it will send me a push notice with an image. It was a fun project to work on.
Nice job, I’ve got unifi g3 cameras too and would love this but I’ve got no idea how to setup something like this, do you think something like this can be integrated into home assistant through a component?
Hi,
The script I am using is something i coded for myself. It contains hardcoded paths etc. I will try to clean it up and publish it but meanwhile you can try to get yolo working by following instructions here: https://github.com/AlexeyAB/darknet
Thanks, no problem if you do manage to share the script that would be great to see how you accomplished it.
I spent a bit of time researching Darknet versus TensorFlow etc and Darkflow making the two talk, and got a bit confused but I think conceptually I got it now. However, I am a beginner and using Ubuntu 16.04 no GPU, CPU only and in order to to be able to get this up and running I need some help with basic list of instructions, I mean basic as in the code to copy and paste into terminal.
Is there any chance you could help with providing the instructions to follow the Alexey instructions?
Is YOLO able to be trained further? For instance, could it conceivably be trained to recognize and classify subgroups of people (eg., specific individuals, USPS/FedEx/UPS drivers by uniform, etc.) Could it also be extended and combined with OCR to allow reading license plates (would be lovely to get a text that my mother’s car was pulling up the drive, or that a deliveryman was coming up the walk)?
Fantastic work and great to see the progress you have made. It seems like there is lot’s of interest in seeing your scripts, can you share? maybe just anonymise paths/passwords/IPs etc.
Were you able to clean this code up and publish? Im curious about this at the moment I am using this setup. https://youtu.be/fwoonl5JKgo
I would like to use this yolo system cause seen more accurate and can detect live feed. I wonder if it’s possible to get yolo integrate with Bi?
