Human Gait Identification System Based on Average Silhouette

Human identification is a common research topic in computer vision and it is substantial for many applications such as criminal investigations, surveillance system medical diagnoses and human-computer interaction. Identification systems can also be used in many places like airports, banks, hospitals and railway stations. In general, identification approaches have shortcomings that can influence the accuracy of such approaches thus they are sensitive to environmental and physiological changes and require subject cooperation as well as low resolution optical sensors that can affect the results. In this work, the proposed method has been well implemented and tested using CASIA dataset which consists of gait silhouette data of 20 subjects, where each subject offers 12 sequences of gait silhouette images. The algorithm that is proposed has three different stages. The representation stage computes the general average silhouette images to capture and gather the most important information from gait sequence. Then dimensionality reduction stage where PCA technique is applied to extract the most important gait features and greatly reduce the dimension of gait data depending on low frequency information. Furthermore, this approach will increase a discriminating power in the feature space when dealing with low frequency information. Since Low dimensional feature distribution in the feature space is assumed to be Gaussian, thus Euclidean distance is applied for classification and matching process in the classification stage. The experimental results indicate that best identification rate is 90% by representing the images using 50 gait silhouette frames for each and every gait silhouette sequence and 100 × 70 frame size.