Image Processing and Computer Vision in Basketball Detection and Tracking

Abstract

This thesis uses a two-stage framework to detect and track a basketball in video footage. In the first stage, the ball is detected in each frame of the video sequence using different feature-based methods. Color-based, shape-based, and combined features methods are tested to determine the superior ball detection algorithm. In the second stage, the trajectory of the ball is tracked throughout the video sequence using a Kalman filter based candidate verification procedure.