NaluNet

Nalu Detection System built in Python3 + Darknet + Yolo V3

Server (Darknet + Flask)

Details:

• Flask API + Darknet + Yolo V3 + Nalu Model

• It's purpose is when you call the endpoint with an image it will return the count of Nalu's it finds

• It has no data retention and no 3rd party calls

Installing:

• cd server

• docker build -t nalunet .

• docker run -p 3038:80 nalunet

• Or on production: docker run -p 3038:80 -p 80:80 -d nalunet

Running:

• curl -F "file=@/Users/user/Desktop/dog.png" localhost:3038

Client (Webcam + Detect)

Details:

• Python3 + OpenCV

• It's purpose is to take a picture from a webcam periodically, call the server's endpoint, store the count (locally or in firebase), and delete the picture

Installation:

• pip3 install firebase_admin

• Method 1 (Raspberry PI): fswebcam (instead of OpenCV)

• This method is to be used if you don't have as powered USB hub (webcam + opencv + v4l can be flaky)

• sudo apt-get install fswebcam

• When running the client add "fswebcam" to the args, e.g. python main.py fswebcam

• Method 2 (OSX): OpenCV installation

• brew install opencv

• echo /usr/local/opt/opencv/lib/python3.6/site-packages >> /usr/local/lib/python3.6/site-packages/opencv3.pth

• Method 3: OpenCV on Debian/Ubuntu (only if using the python script for the webcam instead of the script from "Method 1")

• https://milq.github.io/install-opencv-ubuntu-debian/

• Or try pip install opencv

• Crontab: @reboot cd /home/pi/nalu-net/client && python main.py &

Running locally:

• python main.py (or python3)

Crontab:

Start after booting up: @reboot cd /home/pi/apps/nalu-net/client && python main.py fswebcam &
Auto-pull changes: */5 * * * * cd /home/pi/apps/nalu-net && git pull origin master

To reboot every night use sudo crontab -e: 0 1 * * * /sbin/shutdown -r

Training the model

• Install "Yolo V3" (only necessary for training the model)

• https://pjreddie.com/darknet/yolo/

• git clone https://github.com/pjreddie/darknet

• cd darknet

• make

• wget https://pjreddie.com/media/files/yolov3.weights