Gait biometrics: A vision based approach for cloths invariant walking pattern classification

An interesting and simple human gait can be applied for person´s identification problems by analyzing their walking style in an explicit manner. Different apparel worn by different subjects produces an immense impact in changing the behavior of an individual´s locomotion. In this paper, we have proposed a novel feature extraction technique for classifying human gait irrespective of different cloths. A sequence of silhouette frames has been obtained from OU-ISIR gait database which comprises of 15 different subjects worn by 16 different attire altogether. A computer vision based approach has been applied to derive significant gait feature information from the series of silhouette frames. The Gait Entropy image extracted from the sequence of human gait is considered as feature vector because it provides consistent information of the gait motion signal as much as possible. An attempt has been taken to develop a statistical based classifier using Naïve Baye´s condition probability function. The uncertainties involved in the gait signal due to different cloths have been attended using the properties of entropy feature which produces an encouraging classification result. The performance analysis of the Baye´s classifier has been evaluated using the statistical metric, Receiver Operating Characteristics (ROC) curve.