Algorithms for Gait-Based Human Identification from a Monocular Video Sequence

Abstract

Human gait is a spatio-temporal phenomenon that characterizes the motion characteristics of an individual. It is possible to detect and measure gait even in low-resolution video. This makes it an attractive modality in surveillance applications, where it is often difficult to get face or iris information at high enough resolution for recognition applications. Psychophysical studies indicate that humans have the capability for recognizing people from even impoverished displays of gait, indicating the presence of identity information. From early medical studies it appears that there are twenty four different components to human gait, and that if all the measurements are considered, gait is unique. It is interesting, therefore, to study the utility of gait as a biometric. The goal of this thesis is to investigate the information contained in the video sequences of human gait and how to extract and exploit that information in ways that facilitate human identification. In our work, we present both deterministic and stochastic approaches for gait recognition. Human identification using gait, similar to text-based speaker identification, involves different individuals performing the same task and a template-matching approach is suitable for such problems. In situations where the amount of training data is limited, we show the utility of a simple feature viz. the width of the outer contour of the binarized silhouette of the subject and its derivatives for gait recognition in a dynamic time warping framework. By virtue of their deterministic nature, template matching methods have limited noise resilience. A careful analysis of gait would reveal that it has two important components. The first is a structural component that captures the physical build of a person while the second is the motion kinematics of the body during a gait cycle. We propose a systematic approach to gait recognition by building representations for the structural and dynamic components of gait using exemplars and hidden Markov models (HMMs). The stochastic nature of the HMM yields better noise resilience than the template matching technique. To recognize a person walking at a large distance, humans try to combine information such as posture, arm/leg swing, hip/upper body sway or some unique movements that are characteristic of that person. We demonstrate the same effect through fusion of different dynamic and static gait features in both determinisitic and stochastic frameworks. Most gait recognition algorithms rely on the availability of an exact side view in the probe. However, it is not realistic to expect that this assumption will be valid in most real-life scenarios. We present a view invariant gait recognition algorithm which is based on synthesizing a side view of a person from an arbitrary monocular view. The method is based on the planar approximation of a person that is valid when human identification at a distance is desired.