Multiview-Based Cooperative Tracking of Multiple Human Objects
- PDF / 4,873,985 Bytes
- 14 Pages / 600.05 x 792 pts Page_size
- 28 Downloads / 208 Views
Research Article Multiview-Based Cooperative Tracking of Multiple Human Objects Kuo-Chin Lien1 and Chung-Lin Huang1, 2 1 Institute
of Electrical Engineering, National Tsing Hua University (NTHU), Hsin-Chu 30013, Taiwan of Informatics, Fo-Guang University, I-Lan 26247, Taiwan
2 Department
Correspondence should be addressed to Chung-Lin Huang, [email protected] Received 31 January 2007; Revised 28 July 2007; Accepted 3 December 2007 Recommended by Nikos Nikolaidis Human tracking is a popular research topic in computer vision. However, occlusion problem often complicates the tracking process. This paper presents the so-called multiview-based cooperative tracking of multiple human objects based on the homographic relation between different views. This cooperative tracking applies two hidden Markov processes (tracking and occlusion processes) for each target in each view. The tracking process locates the moving target in each view, whereas the occlusion process represents the possible visibility of the specific target in that designated view. Based on the occlusion process, the cooperative tracking process may reallocate tracking resources for different trackers in different views. Experimental results show the efficiency of the proposed method. Copyright © 2008 K.-C. Lien and C.-L. Huang. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
1.
INTRODUCTION
Currently, multiple-view multiple-object tracking has become an essential technology for many applications such as video surveillance system. Modern video-based surveillance systems [1] employ real-time image analysis techniques for the target tracking and identification. The major issue of target tracking is to identify the multiple moving objects. However, occlusions among the objects will complicate the tracking process and make it difficult for the system to identify the object after occlusion. This paper proposes a novel method for multiple human tracking in multiple views. Recently, researchers have shown a great interest in using particle filters for visual tracking [2–5]. For analyzing the occlusion between targets, Wu et al. [6] propose modeling occlusion relations as an extra hidden process in a dynamic Bayesian network. A hidden variable was used to indicate the three possible relations between two moving objects. The transition process is described with a three-state finite state machine. Hu et al. [7] extend the framework to human tracking. Analyzing the depth order around occlusions is indeed helpful to maintain tracking. However, as the number of targets increases, occlusion relations among targets
get more complicated. Another problem in object tracking is that the appearance of object changes quite often. Zhou et al. [8] present an approach incorporating appearance-adaptive models into a particle filter to realize robust visual tracking. Kang et al. [9] use time weighted color informati
Data Loading...
