Introduction
Using gRPC (Google remote procedural calls) for transferring video over the internet and receiving the results back. The video is sent from the server to the client. The client processes the received video, generates some results and passes these results back to the server. This code uses bi-directinal streaming. Video is streamed from server to client. Results are streamed back from client to server.
Use
This is used to stream live footage from cctv cameras. Video from a CCTV camera is transferred in real-time to a distant proccessing facility. This is because computationally intensive hardware is not located near the camera. At the facility we can use some heavy computational procedure on it like deep learning and transfer the results back to the place at which the camera is located.
In this code, I've performed 'people counting' on the video. The video from the camera is transferred via gRPC. The client, running on another machine, receives this video via the internet. The client performs people counting on the video in real-time and continously transfers the results back i.e the number of people counted till now.
Main requirements:
Grpc openCV skvideo.io
Grpc: https://grpc.io/docs/quickstart/python.html
Install Grpc:
python -m pip install grpcio
Install Grpc tools:
$ python -m pip install grpcio-tools
Install openCV:
sudo apt-get install libopencv-dev python-opencv
Install skvideo:
sudo pip install scikit-video
Running the code
First we make the .proto file and place it in the grpcTest/protos folder. We execute the following command from the folder grpcTest.
python -m grpc_tools.protoc -I protos --python_out=. --grpc_python_out=. protos/imageTest.proto
This creates the files imageTest_pb2_grpc.py and imageTest_pb2.py in the folder grpcTest.
We then write the files imageTest_server.py and imageTest_client.py which use the above 2 generated files.
The server sends the video file via a stream of frames. The client receives these frames and continuously processes each frame and counts the number of people entering the video. This count is sent back to the server using streaming.
Note: Use internet through LAN for fastest transfer.
To start server,
python imageTest_server.py
To start client,
python imageTest_client.py
Links for reference
https://grpc.io/docs/tutorials/basic/python.html https://grpc.io/docs/quickstart/python.html https://grpc.io/docs/guides/concepts.html https://grpc.io/docs/guides/concepts.html#bidirectional-streaming-rpc https://stackoverflow.com/questions/47867440/cv2-imshow-giving-black-screen/47867865#47867865 https://stackoverflow.com/questions/47883841/grpc-not-working-when-i-use-2-computers-instead-of-1/47887825#47887825