Gait based Human Identification through Intra-Class Variations

摘要

The Gait based human identification is the recently critical and challenging research problem for computer vision, visual surveillance and access control security applications. Human gait is behavioral biometrics that involves both the appearance of the body and the dynamics of motion for human walking. Intra-class variation, which involves different carrying condition, variation in clothes and normal walking with different variations in view angles that interfere affect the efficiency in recognition. The extraction of discriminatory and insightful characteristics from video frame sequences is also important. This paper proposes statistical gait characteristics to identify the biometric gait feature for human identification by combining statistical values of SURF and Haralick features. These features are appropriate for human identification to decrease the effects of “covariate variables” so that the accuracy of human identification is efficient. The discriminatory performance of gait pattern classification on CASIA-B (Multi-view Gait Dataset) is assessed by the K- Nearest Neighbor (KNN) classifier.