Font Size: a A A

Gait Recognition Based On Stereo Vision

Posted on:2011-11-22Degree:DoctorType:Dissertation
Country:ChinaCandidate:H T LiuFull Text:PDF
GTID:1118360305966689Subject:Pattern Recognition and Intelligent Systems
Abstract/Summary:PDF Full Text Request
With the abroad of the application of the intelligence surveillance systems, people are not satisfied with the simple function about monitoring of it. However how can know or understand the behavior of the human and further recognizing his identity is to be an important issue. Therefore, all kinds of non-contact biometrics based on biometrics have emerged. They can analyze and recognize the identity of human in the surveillance scene, and offer the proof for the decision-making. It is noticeable that a fascinating biometrics named gait recognition is appeared. It is able to discriminate the identity of the people only with analyzing gait motion under a distance and non-intrusion. Nowadays gait recognition is an attractive biometrics and also is the unique biometrics at a distance.This dissertation proposed a novel algorithm with gait recognition based on stereo vision method after studying a lot of the approaches of gait recognition at a detail. We proposed a series problems on the study of gait recognition based on stereo vision, and made many solutions corresponding with them. Finally, system solutions on that basis for the gait recognition are given.The innovation and the major work of this dissertation are as follows:(1) The stereo vision technique is originally introduced to gait recognition in the world. Further, the system framework of gait recognition is raised, and the related system solutions are also given.(2) Since there have not any a stereo gait database, we constructed the stereo gait database PRLAB I and PRLABâ…ˇwhich laying the foundation for further study. The two stereo gait databases are able to test the effectiveness of gait feature, the feasibility of gait algorithm, and also evaluate the robustness of the gait feature and the gait algorithm. The purpose of PRLAB I is to test the effectiveness of the gait feature and the feasibility of our gait recognition algorithm, while the purpose of PRLAB II is to verify the robustness of the gait feature and the gait recognition algorithm. Meanwhile the potential of gait recognition also can be estimated by PRLAB II.(3) In the place of gait feature extraction, we exploited a descriptor named 3D Body Contour Descriptor (3D-BCD) to represent the silhouette of human after analyzed the rule of the 3D contour sequences. Hereby the 1D Gait Feature Representation (1D-GFR) which is the L2 norm of every unit of the 3D-BCD is formed the original stereo gait feature (SGF). And then, we performed a series of procedures such as the noise filtering, normalizer with dimensionality and scaling about the original stereo gait feature, for obtaining the actual gait signature. In addition, we reconstructed the stature of people utilized the 3D reconstruction which is the technique of stereo vision approach.(4) In the place of dimensionality reduction, the Principal Component Analysis (PCA) and the Manifold Learning approach was used for dimensionality reduction of gait feature data for the sake of removing the redundancy of it.(5) In the place of recognition algorithm, both of the Mean of Sample Sequences (MSS) and the Mean of Sample Sequences about Template (MSST) are defined based on the measure of Euclidean Distance. Accordingly, both of the Nearest Neighbor Classifier (NN) and the Nearest Neighbor Classifier about Template (TNN) is constructed for classifying and recognizing.For verifying the effectiveness of gait algorithm, a lot of experiments have been performed in the stereo gait database PRLAB I and PRLABâ…ˇ, and the stereo gait dataset ExN which is an irregular stereo gait dataset. Then, we analyzed and discussed the experimental results in depth. In addition to that, we also investigated the internal structure of the gait feature data. The final experimental results revealed that our algorithm have the higher recognition rate and stronger robustness.
Keywords/Search Tags:Biometrics, Gait Recognition, Stereo Vision, 3D Body Contour Descriptor, Stereo Gait Feature, Principal Component Analysis, Manifold Learning, Laplacian Eigenmaps
PDF Full Text Request
Related items