A review of neural network-based gait recognition

This paper is a comprehensive and structured introduction to existing neural network-based gait recognition methods. It begins with an overall introduction to gait recognition in the first part, which describes the advantages of gait recognition. Application areas. In the second part, the development of gait recognition and the evolution of deep learning in gait recognition are introduced. The third part starts with the most popular types of feature representation class energy maps, and details the advantages, disadvantages and roles of different classes of energy maps; the fourth part starts with the most popular types of neural networks for gait recognition classification, and details the role of different types of neural networks in gait recognition. Finally, we discuss the current challenges and put forward some promising directions for the future research of gait recognition.

I. Introduction

Gait recognition is a new technology in biometric recognition, which aims to recognise movement by the posture of a person walking. In recent years, research related to gait recognition technology has attracted a great deal of scholarly interest because gait recognition technology has advantages that are unmatched by other biometric technologies. Firstly, human gait typically captures objects at a distance from the camera. In contrast, such as fingerprints, cannot be captured [1]. Secondly, gait recognition can be performed at low resolution[2]. Thirdly, gait recognition can be done with simple instruments [3]. Cameras, smartphones with their own accelerometers, floor sensors and even continuous wave radar can be used to capture human gait information. Fourthly, gait recognition does not require the cooperation of a human.Fifthly, gait characteristics are difficult to imitate. This property is important for crime analysis, as described in reference [4].