Ball Route Estimation in Broadcast Soccer Video.pdf

Ball Route Estimation in Broadcast Soccer Video Takumi Shimawaki1, Jun Miura1, Takuro Sakiyama1, and Yoshiaki Shirai2 1 Department of Mechanical Engineering, Osaka University, Suita, Osaka 565-0871, Japan {simawaki, jun, saki}@cv.mech.eng.osaka-u.ac.jp http://www-cv.mech.eng.osaka-u.ac.jp/ 2 Department of Human and Computer Intelligence, Ristumeikan University, Kusatsu, Shiga 525-8577, Japan shirai@ci.ritsumei.ac.jp Abstract. This paper deals with the analysis of broadcast soccer video. To recognize interesting events such as a goal, estimation of ball move- ments is necessary. It is, however, sometimes difficult to detect a ball by a simple color and shape-based method when it overlaps with players and lines. We therefore develop a method of estimating a ball route during such overlaps by considering spatio-temporal relationships between play- ers, lines, and the ball. The method can deal with difficult cases such as the one where a ball disappears at a player and re-appears from another player. Experimental results show the effectiveness of the method. 1 Introduction There are increasing demands for summarizing a broadcast video of a soccer game (or other sports) to make a digest of interesting scenes such as goal scenes so that viewers can quickly survey the game. We are now developing a system for retrieving interesting and informative scenes based on scene understanding. To understand various scenes in soccer games, it is essential to know the movements of players and a ball. This paper focuses on ball tracking. There are several ball detection methods, which use, for example, SVM [1] or a generalized Hough transform [2]. Most of the previous methods are, however, applicable only when a ball is sufficiently large and not so fast in the image. In our case, a ball is usually small and sometimes moves fast. Moreover, it sometimes overlaps with players and lines. Several ball tracking methods deal with such problems by applying statistical filters such as Kalman fi