Human gait recognition system for viewpoint variations in visual surveillance system

Abstract

Biometrics has gained a lot of attention among researchers in recent years due to its importance in security. Gait is one of the biometrics modalities, which is capable of capturing unique features from long distances using a Closed-Circuit Television (CCTV) camera without the subject s attention and cooperation. It improves recognition of distance-based identification, analyzes low-resolution images, illumination changes, viewpoint variations, occlusions when compared with other biometric modalities. It characterizes a person s way of walk by taking raw input as image or video sequences to identify the walking behavioral pattern of humans. This has helped the police departments as a source of e document (evidence) to protect the laws of citizens. It also helps banks, restaurants, supermarkets, parking lots, homes in detecting the clear viewpoint (orientation) of a person walking from the CCTV camera footages. There is a need for visual surveillance to monitor suspicious activities happening all over the world to prevent various criminal activities like trafficking, robbery, harassment, and so on. The covariate factors that will affect conventional gait recognition systems are changes in viewpoint variations, resolutions, variations in carrying objects, wearing different footwear or clothes, and measuring speed or walking directions. Among the number of factors, change in viewpoint variations is considered as the ultimate challenge as it frequently varies with the graphical structure of the human. Also, the gait template obtained from sequences of silhouette images is not clear to recognize the viewpoint variations. Hence, a new framework called Human Gait Recognition System (HGRS) is proposed to address the viewpoint variations in human silhouette images for CCTV surveillance cameras. This framework provides cost-effective measures and helps to overcome the drawback of high newline