12 January 2018 Structure preserving clustering-object tracking via subgroup motion pattern segmentation
Author Affiliations +
Tracking clustering objects with similar appearances simultaneously in collective scenes is a challenging task in the field of collective motion analysis. Recent work on clustering-object tracking often suffers from poor tracking accuracy and terrible real-time performance due to the neglect or the misjudgment of the motion differences among objects. To address this problem, we propose a subgroup motion pattern segmentation framework based on a multilayer clustering structure and establish spatial constraints only among objects in the same subgroup, which entails having consistent motion direction and close spatial position. In addition, the subgroup segmentation results are updated dynamically because crowd motion patterns are changeable and affected by objects’ destinations and scene structures. The spatial structure information combined with the appearance similarity information is used in the structure preserving object tracking framework to track objects. Extensive experiments conducted on several datasets containing multiple real-world crowd scenes validate the accuracy and the robustness of the presented algorithm for tracking objects in collective scenes.
© 2018 Society of Photo-Optical Instrumentation Engineers (SPIE)
Zheyi Fan, Zheyi Fan, Yixuan Zhu, Yixuan Zhu, Jiao Jiang, Jiao Jiang, Shuqin Weng, Shuqin Weng, Zhiwen Liu, Zhiwen Liu, } "Structure preserving clustering-object tracking via subgroup motion pattern segmentation," Optical Engineering 57(1), 013101 (12 January 2018). https://doi.org/10.1117/1.OE.57.1.013101 . Submission: Received: 21 August 2017; Accepted: 11 December 2017
Received: 21 August 2017; Accepted: 11 December 2017; Published: 12 January 2018

Back to Top