Gait-Face Based Human Recognition From Distant Video

This paper presents a non-intrusive video-based identification system with a single camera view placed at 90° angle to the subject. The aim is to efficiently extract biometric features from a low-resolution video (acquired from a common CCTV camera) for recognizing individuals. The proposed method uses gait cues and face profile features for identification of individuals. A fusion rule is applied to these feature sets to obtain a new set of attributes. Thus, three different recognition models are developed using the gait, face, and fused feature sets. An ensemble technique is defined over the three classification models based on these sets of cues to identify an individual. This approach is experimentally validated on the CASIA-B dataset that achieves 99.33% identification accuracy.