Research And Implementation On Human Identification Algorithm Based On Gait

Gait recognition is a focused research direction in the field of machine vision currently. By analyzing the pedestrians of the surveillance video to extract the feature which existing in the current video and accomplish the gait recognition task eventually. As a new biological characteristic, gait is the only one which can be captured at a far distance without requiring physical information from subjects, so it has a broad application prospects in the field of intelligent monitoring. In this dissertation, in order to significantly improve the recognition rate of gait recognition and reduce the computational complexity, this dissertation proposed a gait recognition algorithm of generalized linear discriminant analysis based on euclidean norm. And then in order to make it more applicable to real life, this dissertation further put forward another manifold learning algorithm for view-invariant gait signal computing. So it can correctly identify different people when the angle is not fixed. The main work of this dissertation are summarized as follows:(1) Firstly, introducing the research background and significance, and the current research situation about gait recognition, point out the existing shortcomings. Further, detailed the related theoretical knowledge involved in this dissertation, it includes the basic framework of gait recognition, the contour extraction of gait, the period detection of gait, the feature extraction of gait, the feature processing and pattern classification. At last, giving an outline of the published gait database.(2) Due to the low recognition accuracy, high computational cost and low processing speed, the current gait recognition algorithm is difficult to meet the actual needs. So this dissertation proposed a gait recognition algorithm of generalized linear discriminant analysis based on euclidean norm. Firstly, computing the distance between boundary point and the centroid. Then by analyzing the defects of the traditional linear discriminant analysis algorithm which can not correctly distinguish the non-edge. So in order to reduce the impact of edge class to feature subspace, this dissertation added the distance as the weight to the definition of between-class scatter matrix to get the feature projection matrix. Finally one-against-ones Support Vector Machine (SVM) is applied to implement gait classification. Experimental results proved that our algorithm achieves higher recognition rate, lower computational cost, and faster processing time compared with others.(3) According to the existing gait recognition algorithms need the view fixed so that it cannot be applied to real life, this dissertation proposed a manifold learning algorithm for view-invariant gait signal computing. Firstly, extracting the gait energy image from different angles as the feature. And then this dissertation aims to learn a feature subspace to unfold the interclass samples collected from the same view and squeeze the intraclass samples collected from different views. Finally, completing the classification and recognition with classifier. The experimental results show that this algorithm can complete the recognition task of view-invariant independently, and its computational complexity is lower than other manifold learning method.