People Detection using Gait for Visual Surveillance

Abstract

Detecting, tracking and recognizing people using a single camera is a challenging problem due to occlusion, shadows, entry and exit of objects into the scene, and natural background clutter. Furthermore, the flexible structure of the human body, which encompasses a wide range of possible motion transformations, exacerbates difficulties for developing a vision-based surveillance system. We propose a multi-object tracking method based on feature correspondence between consecutive frames. Moving objects are assigned to different layers whereby blobs corresponding to the same object are assigned to the same layer. The criteria for allocating objects to layers is based on the Mahalanobis distance measure of shape-based features. Because of the dearth of visual surveillance systems that exploit human gait for object classification and their limited aim to detect people only using simple shape-based features extracted from silhouettes, we have explored an alternative technique for walking people detection based on their gait motion. The novelty of our approach is motivated by the latest research for people identification using gait

More information

Identifiers

ORCID for M. S. NIXON: orcid.org/0000-0002-9174-5934

Catalogue record

Export record