Abstract.
In this paper we propose a scheme for ball detection and tracking in
broadcast soccer video. There are two alternate procedures in the scheme: ball
detection and ball tracking. In ball detection procedure, ball candidates are first
extracted from several consecutive frames using color, shape, and size cues.
Then a weighted graph is constructed, with each node representing a candidate
and each edge linking two candidates in adjacent frames. Finally, Viterbi algorithm
is employed to extract the optimal path as ball’s locations. In ball tracking
procedure, Kalman filter based template matching is utilized to track the ball in
subsequent frames. Kalman filter and the template are initialized using detection
results. In each tracking step, ball location is verified to update the template
and to guide possible ball re-detection. Experimental results demonstrate that
the proposed scheme is promising.
1 Introduction
In the past decade, sports video analysis from the standpoint of computer vision has
attracted much attention, especially in ball games such as soccer [1-7], American
football [8], tennis [9], snooker [10] etc. Through detection and/or tracking of the
moving objects (players, ball), several high level analysis can be done, e.g. highlight
extraction [2], event detection [10] and tactic analysis [6]. This paper focuses on ball
detection and tracking in broadcast soccer video, which is a challenging task.
There are some literatures stating the problem of soccer ball detection and tracking.
Chromatic and morphological features are utilized to detect ball in [1]. In [2] the authors
use template matching to detect ball in difference image after camera motion
compensation at regular intervals, and then ball tracking is carried out between such
intervals. In [3] ball’s location is initialized manually, after that Kalman filter and template
matching are applied to track it. In [4] the authors employ motion information in
ball detection and tracking, but in their case the cameras are fixed. A modified version
of the directional Circle Hough Transform is used to detect ball in real (not broadcast)
image sequences in [5]. In [6], ball candidates are first obtained in each video frame
based on playfield detection. Afterwards, Kalman filter is employed to generate candidate
trajectories from which ball trajectories are selected and extended. In [7], the
authors exploit a coarse-to-fine strategy to identify ball in a single frame after playfield
detection, and then CONDENSATION algorithm is used to track the ball

DOWNLOAD FULL REPORT
http://jdl.ac.cn/doc/2005/A%20Scheme%20f...0Video.pdf