This is a (Dockerised) demo for object detection and face recognition tasks in the browser. A Flask app serves some very simple pages and a JavaScript script, which takes snapshots from the camera stream and sends them back through a POST request. Then, the images are processed using OpenCV and face_recognition in python; bounding boxes, labels and scores are sent back in the request's response.
Important: This demo uses HTML5 to handle the camera, thus you'll need a browser that supports it.
First, of course, clone this repo ;).
For the object detection part, you'll need to download the Yolov3 model. To this end, go to https://github.com/pjreddie/darknet/ and download data/coco.names (rename it as yolov3.names) and cfg/yolov3.cfg. Also, download the pre-trained model weights from https://pjreddie.com/media/files/yolov3.weights. Save these 3 files under yolo_v3/. Alternatively, you could get these files using wget within the Dockerfile, if you don't mind downloading them every time you build the image.
For the face recognition part, put your face images under /local_data in jpg format with the name you want displayed as the image name. For example Alice.jpg and Bob.jpg would be the images of Alice's and Bob's faces, respectively.
To run it using Docker simply run
docker build -t webcam_cv .
inside the repo's directory to build the image (wait for a little while - maybe have a coffee), and then run
docker run -p 5000:5000 -v $(pwd):/app --name webcam_cv webcam_cv
if you're running it for the first time (add -d to run it detached from your shell). It will be available under http://localhost:5000 using the repo's directory as a shared volume. Later you can start/stop it by running
docker start/stop webcam_cv
You could also run it without Docker, but then you'll need to handle dependencies by yourself.
To run this on a server (i.e. other than localhost), you'll need to configure it to run over https, otherwise you won't have access to the camera - after all, you don't want your video stream to be sent unencrypted over the internet, do you?. How to do that is out of the scope of this document, but in case it's useful for you, I've successfully done this for a demo using a AWS EC2 instance, a free domain name, and Certbot. You'll have to either change the gunicorn.py config file or set up another web server, like apache or nginx (that's what I used).
As most of what we do (or human knowledge in general), this is based on others work.
HTML and JS scripts to handle the camera stream are based on: https://github.com/webrtcHacks/tfObjWebrtc
To know more about Yolo v3, go to: https://pjreddie.com and/or https://www.learnopencv.com/deep-learning-based-object-detection-using-yolov3-with-opencv-python-c/
Last but not least, the face_recognition github repo can be found here:
https://github.com/ageitgey/face_recognition