Towards Real-Time Detection and Tracking of Basketball Players using Deep Neural Networks

Online multi-player detection and tracking in broadcast basketball videos are significant challenging tasks. In this environments, the target distributions are highly non-linear, and the varying number of objects creates complex interactions with overlap and ambiguities. In this paper, we present a real-time multi-person detection and tracking framework that is able to perform detection and tracking of basketball players on sequences of videos. Our framework is based on YOLOv2, a state-of-the-art real-time object detection system, and SORT, an object tracking framework based on data association and state estimation techniques. For training and testing, we use a given subset of the NCAA Basketball Dataset. As part of the bonus, we trained a two-layer LSTM to do action recognition.