Detection and Tracking of Sport Players on Videodata Using Deep Learning Methods

This article presents the research results of JDE algorithm for detection and tracking athletes in video data. The developed convolutional neural network was trained and tested on the NVIDIA DGX-1 supercomputer. To analyze the quality of the model, MOTA metric was used, which is related to how the human eyes track targets in the video stream. The detection and tracking people on video data is relevant to many tasks of computer vision, in particular, in sport to collect statistics about players. The quality of JDE algorithm was assessed by the video fragments of basketball games filmed in the sports halls at P.G. Demidov Yaroslavl State University.