Research On Key Techniques For Gait Recognition

Biometric identification realizes human recognition by using unique biological features, which has high application values in the field of information security. Biometrics identification becomes more and more important, and it can be widely used in economy, law and national security. Lots of researches have been done, and some recognition techniques have achieved big success and have been in utility in many important places. But all those methods are with some flaws, such as real touch and so on. Therefore, we need such a recognition technique that can overcome disadvantages, while Gait recognition is the right one. It is a novel biometrics recognition technology, which can recognize a person by his or her walking style, and it is catching more and more attention in the field of security protection for its special features: no need of physical touch, can be captured in distance, without violation, and elusion.This thesis is aimed to do research on gait recognition, main work is mentioned below:(1) For reducing the complexity of the process and saving time, improvements are done on background subtraction method which is based on OSTU adaptive threshold method and regional background update.(2) To abtain accurate human edge, improvements are done on Canny algorithm. Insert latitude to every pixel to increase accuracy, and gradient histogram is used to increase adaptability.(3) A novel recognition algorithm is displayed on the basis of existing technology, and based on this algorithm, using CASIA database A, the joint angle and areas of samples are subscribed which are sperately analyzed by Discrete Cosine Transform for removing relativity of siginals and principal components analysis for decreasing feature dimension later, and the gait cycle and contour vector are also subscribed.(4) Some experiments are done in different conditions using single features or fuse features,to certificate that,fusion of gait features can improve the performance of gait recognition.What is more, the result can also provide that the number and feature of sample, along with classifier style, affect the performance of gait recognition.