This project is realised for the course of ML4CV at UniBo.
The aim is to build an AI system able to track the volleyball players inside videos; the videos are in reference of my volleyball team and the camera is by standard configuration in the same spot behind the players and the system will track the players in the close half of the field.
The videos are given in format .m4v that is not suitable for OpenCV and other CV operations, so it was necessary to convert those in .mp4 thanks to the convert_videos.sh inside utils' folder.
./convert_videos.shFirst of all, I need to perform object detection on the frames of the input video. Obviously, this operation can't be done directly on the video files so I have to extract some frames; this was achieved with some basic operations of OpenCV and Numpy libraries. From a single video, I extract 3000 random frames (I can fix this parameter later based on my results) and these are inserted inside a new folder.
I choose to perform the detection with Yolov8, a popular pretrained detector that can give good results. To use it, I had to add some installations.
pip install ultralytics
pip install tqdm --upgrade