Gait identification using component based gait energy image analysis

In the modern era of computer vision technology, gait biometric trait increases the proliferation of human identification in video surveillance situation. This paper intends to discuss the robustness of gait identification irrespective of small fluctuation in subject's walking pattern. The Gait Energy Image (GEI) is computed on silhouette gait sequences obtained from OU-ISIR standard gait database. The advantage of working with GEI is to preserve the shape and motion information into a single averaged gait image with fewer dimensions. The three independent components such as head node, body torso and leg region are separated from subject's GEI in accordance to body segment ratio. The local biometric feature has been computed from the shape centroid to the boundary points of each segment. The normality testing of feature for each region of GEI body frame ascertains the discriminative power of each segment. The similarity measurement between gallery and probe gait energy image has been computed by cosine distance, correlation distance and Jaccard distance. The performance efficiency of different distance based metrics is measured by several error metrics.

I. Introduction

Gait Identification has proved its effectiveness through an extensive research especially in the field of surveillance security and medical implications. The Human gait biometric is easy to retrieve and difficult to mimic among all the popularly known biometrics available such as iris, fingerprints, and face. Gait can be defined as an outcome of complex muscular activity which exhibits a periodic motion and contributes to a distinctive biometric. There are two basic methodologies used for retrieving gait data: Video Based and Sensor Based. Computer vision based technique has been followed for tracking, detecting and identifying human due to its unobtrusive and non-invasive properties. This biometric will produce promising results even though distance between the camera and person is adequately enough. The video based gait analysis is computationally intensive for handling segmentation, tracking and silhouette extraction algorithm on human gait video. A biometric Gait is also preferred in treatment of locomotive diseases. On the other hand, the non-vision based technique is used in gait analysis which requires physical contacts of the person for data acquisition. The sensor based technique might attract wrong recognition results due to several factors like displacement of sensors during walking, errors in measurement of sensor readings, and discomfort in wearing sensor based body suit while walking.