Ball Detection And Tracking In Soccer Video

Ball detection and tracking is a most crucial step for automatic analysis of soccer video, for which Kalman-based template matching algorithm is widely applied. But the problem still has not been fully resolved contributing to difficulties like its small area, fast and changing speed, and frequent occlusion.To gain the ball on the playfield in long shots, a Viterbi-algorithm based ball detection and tracking scheme is proposed employing motion continuity. Ball sequence is determined over successive frames. But not traditionally determine a console ball in a single image. Frame differentiating, Canny operator and Dilation Operation of Mathematical Morphology is utilized to get segmented objects, and a region growth algorithm is utilized to get corresponding region information for each object. Ball candidates are sieved by removing obviously non-ball objects, and a weighted graph is constructed on candidates over T consecutive frames. Viterbi algorithm is used iteratively to select the ball-path pieces on the weighted graph. Interpolation functions are gained employing least squares, based on which, false alarms are eliminated and missing balls are reasoned or back tracked.The experimental results have verified the effectiveness of the given scheme. The ball could be effectively detected when it's merged with playfield lines, or only occluded in few frames. Comparing with Kalman-based template matching algorithm, the proposed scheme is with higher precision when the ball changing its directions frequently.