Human gait recognition under changes of walking conditions

Abstract

The study of human gait is innate to human interest and pervades many fields including biometrics, clinical analysis, computer animation, and robotics. From a surveillance perspective, gait recognition is capable of identifying humans at a distance by inspecting their walking manners. It is an attractive modality which can be performed surreptitiously in an unconstrained environment. Gait is one of the few biometric features that can be measured remotely without physical contact and proximal sensing, which makes it useful in surveillance applications. However, in the real world, there are various factors significantly affecting human gait including clothes, shoes, carrying objects, walking surfaces, observed views, and walking speeds. Among these factors, changes of views and speeds have been regarded as two of the most challenging problems for gait recognition. Particularly, view change will significantly impact on available visual features for matching, while speed change will alter walking patterns of each individual substantially. This thesis is mainly to develop novel methods for recognizing gaits under changes of walking conditions focusing on views and speeds, without a cooperative camera system. Five major methods are proposed from several various perspectives to address key aspects of these problems. Principally, a view-normalization of gaits is obtained through a new learning process by using mapping/projection relationships between correlated gait features across different views, while a novel speed-invariant gait feature is developed by using a statistical shape analysis based on a local-static gait information. Based on widely adopted gait databases, the comprehensive experiments are carried out to verify the proposed methods. It is concluded that the proposed methods can achieve state-of-the-art performances for gait recognitions under view change and/or speed change. In this thesis, the other relevant problems are also sorted out, including gait period analysis, view classification, and walking speed estimation. Moreover, in order to enhance the performance, multi-view gait information is utilised to achieve more stable and convincing outcomes.