On Using Gait Biometrics for Re-Identification in Automated Visual Surveillance

Introduction

Although personal privacy has emerged as a major concern for the deployment of large scale surveillance systems, research into automated visual surveillance has received remarkable interest within the computer vision community with potential integration of biometric technologies and human activity recognition systems. This is mainly due to the proliferating number of crimes and terror attacks as well as the vital need to provide safer environment. In fact, the inability of human operators to monitor the increasingly growing numbers of CCTVs installed in highly sensitive and populated areas such as government buildings, airports or shopping malls, has rendered the usability of such systems to be useless. According to the British Security Industry Association, the number of surveillance cameras deployed in the UK was estimated to be more than 5 million in 2015; this figure is expected to increase rapidly particularly after the terrorist attacks that London witnessed in July 2005. Despite the huge increase of monitoring systems, the question whether current surveillance systems work as a deterrent to crime is still questionable. Security systems should not only be able to predict when a crime is about to happen but, more importantly, by early recognition of suspicious individuals who may pose security threats via the use of biometrics, the system would be able to deter future crimes as it is a significant requirement to identify the perpetrator of a crime as soon as possible in order to prevent further offences and to allow justice to be administered. The process of tracking people from one place to another place using surveillance networked cameras would be crucial for gathering valuable security intelligence. Moreover, queries can be made to search for possible locations of a given suspect that can indeed help security officers in their investigations and can lead to further evidence. Traditionally, it is impossible for human operators to work simultaneously on different video screens in order to track people of interest as well as analyze their behaviors across different places. Thus, it has become an essential requirement for research scientists from the computer vision community to investigate visual-based alternatives to automate the process for identity tracking over different views in addition to human activity analysis. Recently, various approaches were published in the literature to accomplish this task based on using basic features such as shape or color information. However, their practical deployment in real applications is very limited due to the complex nature of such problem. An alternative solution would be to employ biometric-based systems that can work at a distance and for low-resolution images such as gait and soft-based biometrics.