Tracking of basketball players from a moving pan-tilt camera

Abstract

A novel multi-modal method for player tracking is developed, with an initial stabilization step done through the calculation of displacement of reference points between two frames. Color segmentation and pixel counting of edge images are implemented to detect and verify the location of players in the image frame. For all video sequences, the pixel difference for video stabilization ranged from 2.0 to 6.7 pixels. The positions of these players are tracked on a 2D top view though homography.