To search, Click below search items.


All Published Papers Search Service


Ball Automatic Detection and Tracking in Long Shot Views


Mohaddeseh Kia


Vol. 16  No. 6  pp. 1-8


This study presented a new method to recognize and track soccer ball as well as long shot view recognition from other views. Soccer ball recognition in the proposed method composed of two parts. Soccer balls initial recognition is done through comparative thresholding on grass, bubbles extraction and analysis. Recognition algorithm or cascade classifier and pseudo Har properties are used in the case of ball overlapping or obstruction with other bubbles occurred. Optical flow property was applied to decrease the search space, the ball is tracked by Locus Kanade method in the pyramid mode which not only identifies the ball more precisely in the candidate bubbles but also reduces the range of applying pattern recognition algorithm in the overlapped situation requiring less computational cost. The results of two films show that in long shot recognition the percentage of accurate recognition is higher than 99% and error percentage is less than 0.8%. Moreover, comparing ball recognition with two competing methods showed that the proposed method had more recognition number percent and the precision has been improved 9.43 and 0.46 % as compared to the two competing methods.


Tracking, Ball, Soccer, Locus Kanade, Cascade classifier, Pseudo Har properties