Authors:
Ya-Li Hou
and
Grantham K. H. Pang
Affiliation:
The University of Hong Kong, Hong Kong
Keyword(s):
Crowd segmentation, Human detection, Occlusions, Coherent motion.
Related
Ontology
Subjects/Areas/Topics:
Informatics in Control, Automation and Robotics
;
Robotics and Automation
;
Surveillance
;
Vision, Recognition and Reconstruction
Abstract:
With a rough foreground region, crowd segmentation is an efficient way for human detection in dense scenarios. However, most previous work on crowd segmentation considers shape and motion cues independently. In this paper, a method to use both shape and motion cues simultaneously for crowd segmentation in dense scenarios is introduced. Some results have been shown to illustrate the improvements when multi-cue is considered. The contribution of the paper is two-fold. First, coherent motion in each individual is combined with shape cues to help segment the foreground area into individuals. Secondly, the rigid body motion in human upper-parts is observed and also used for more accurate human detection.