Google Coral Inference
This tool is designed with the purpose of aiming CCTV cameras at basket markers to extract the match points and update in real time a webpage. Once the camera is installed, you set up all the parameters and do selections of the areas that will be cropped to process later.
The tool allows performing OCR with tesseract or inference with a custom tensorflow model on the selected area to extract numbers.
OpenCV is used to perform all the image pre-processing, and then the numbers are extracted.
C++ allows processing every single frame efficiently and run inference in quite a little time. The aim is to skip a frame or two but be able to run several streams simultaneously without saturating the server.
The Google Coral is used to load a classifier and perform number classification to obtain the score.
Certain environments due to lightning and font require a custom model while others simple OCR perform good enough already.
Technologies used:
C++, PHP, Google Coral, Symfony, tesseract, RTSP, tensorflow, tensorflow lite, OpenCV