DETECTION AND AUTOMATIC IDENTIFICATION OF HUMAN WALK

摘要

Use of video in robot vision and machine understanding is fundamental for number of high-level applications. These include identification of humans by "the way they walk" (biometrics), human-robot interaction, pedestrian safety, automated video surveillance etc. For high-level procedures to be performed, low-level operations have to be executed. These include target detection, tracking and labeling as well as understanding of target interaction. There are several commercially available programs for detection and analysis of human walk like W~4, Pfinder and Spfinder. Some of these programs use stereo imagery, which is not always suitable. Algorithm introduced in this paper doesn't require stereo imagery. It uses nonparametric background modeling for target extraction and star-skeletonization for identification of target class. Obtained results are promising and can be used in further development.