Research on Object Tracking Based on Graph Model in Sports Video

Research on Object Tracking Based on Graph Model in Sports Video

Zhexiong Cui (Harbin Normal University, Harbin, China), Jun Zhang (Harbin Normal University, Harbin, China), XiaoFei Zhang (Harbin Institute of Physical Education, Harbin, China) and Lishu Xu (Harbin Normal University, Harbin, China)
Copyright: © 2018 |Pages: 14
DOI: 10.4018/JITR.2018070101

Abstract

This article is about object tracking based on graph modeling. Object tracking is usually initialized by object detection methods. The fundamental hypothesis is that the object's pattern can be separated from its surrounding background sufficiently. However, for some objects, e.g., the ball in broadcast soccer videos, it is hard to extract effective features to detect the ball in a single video frame. The strategy adopted here is to identify the object's candidate regions in several consecutive frames, and then use a graph to construct the relationship between candidate regions. Finally, a Viterbi algorithm is used to extract the optimal path of the graph as the object's trajectory. This process is called short-term tracking. Then, it is used to initialize a Kalman filter to perform long-term tracking. In the process of tracking, the tracked region is verified to determine whether tracking is a failure, and short-term tracking is restarted if a failure happens.
Article Preview
Top

1. Introduction

When researchers design object tracking algorithm, objects are often initialized in a manual manner. However, it requires automatic initialization for the tracking system in actual application. The process usually completes by object detection algorithm. For some objects, such as human face, there are accomplished detection methods and have been applied to actual practice. But, for some other objects like football in the broadcasting videos, it’s difficult to detect any single one frame video. The developed automatic football trajectory extraction method has important significance to the analysis of football videos.

In the past decades, sport video analysis especially football video analysis arouses wide-spread concerns of researchers. The analysis of sport video is valuable both scientifically and practically. From the perspective of discipline construction, sport video analysis covers computer vision, pattern recognition, machine learning and signal processing. The advancement of sport video analysis promotes the development in such areas. From the point of application prospect, sport videos are attractive to mass audience. Take football for instance. Each year, football match broadcast by television reaches up to hundred sessions appealing billions of audiences. The resultant broadcast football videos are enormous. One football game often lasts 90mins; but the wonderful events like hitting home to audience’s attention takes up only a few minutes. How to extract simple and short highlights from longer video data has become an issue to be solved urgently by researchers.

In football videos, the occurrence of highlights is closely associated with the rubber ball. So to track the ball as to fetch its trails and analyze them, it will provide worthy clues to the detection of highlights. On the other hand, the extraction of football trace can provide technical aid for the 3D construction of football video scenes.

The object tracking method in the paper is proposed with football as background. Hence, to begin, we introduce about the work relating to football detection and tracking. Gong (Y. Gong et al., 1995) stated to detect football in the use of chromaticity and morphological features. Firstly, for white region whose area is bigger than threshold, calculate its annular ratio; if it’s bigger than the threshold, the region belongs to ball’s candidate region and will be searched in the next one-frame partial region; for the matching white region, if its size and annular ratio are identical to the candidate region, it’s believed the candidate region is ball. Yow (Yow, Yeo, Yeung & Liu, 1995) suggested dividing videos into evenly-spaced fragments; in the beginning of such fragments, with template matching strategy, they detected football in the differential image after motion compensation; next, followed up the ball among fragments. Seo (Seo, Choi, Kim & Hong, 1997) initialized manually the location of football; then, traced the ball by Kalman filter and template matching. Ohno (Ohno, Miura & Shirai, 2012) detected and tracked football by the advantage of movement characteristics. They used videos taken by static camera. D'Orazio (D'Orazio, Ancona, Cicirelli & Nitti, 2012) employed improved directed circular Hough transformation to inspect ball. They used videos taken by themselves, instead of broadcast football videos. Yu (Yuet et al., 2013) attempted to get ball’s candidate through field detection in every frame video; then, Kalman filter was employed to generate ball’s candidate trajectory; finally, the ball’s trail was chosen from candidates and was extended. Tong (Tong, Lu and Liu, 2014) did playfield detection in separate one frame video; then detected the ball by from coarse-to-fine strategy; next, they realized tracking of the ball by means of particle filter algorithm.

Although certain solution was offered to detect and follow football from methods mentioned above. However, the ultimate solution is still a long way ahead. The detection and tracking of football in broadcast videos are a very difficult problem. It faces lots of challenges coming from the following aspects:

Complete Article List

Search this Journal:
Reset
Open Access Articles
Volume 13: 4 Issues (2020): 1 Released, 3 Forthcoming
Volume 12: 4 Issues (2019)
Volume 11: 4 Issues (2018)
Volume 10: 4 Issues (2017)
Volume 9: 4 Issues (2016)
Volume 8: 4 Issues (2015)
Volume 7: 4 Issues (2014)
Volume 6: 4 Issues (2013)
Volume 5: 4 Issues (2012)
Volume 4: 4 Issues (2011)
Volume 3: 4 Issues (2010)
Volume 2: 4 Issues (2009)
Volume 1: 4 Issues (2008)
View Complete Journal Contents Listing