Study Of Gender Classification Based On Gait

Gait recognition is as a new emerging biometric identification technology, and gait-based gender classification belongs to a sub-realm of the biometric identification. Because there are five features which are unique, long-distance, non-invasive, easy acquisition and difficult to hide and camouflage for gait recognition, such features are more helpful to make the real-time video surveillance system intelligent, gait recognition is paid more and more attention. Gait-based gender classification can play an early screening effect in the video monitoring system, and such rough classification makes target’s range to be in half, thereby improves the speed and accuracy of the subsequent identification. Gait-based gender classification can also be used in densely populated places, such as markets and airports. It can make passenger flow analysis effectively, make the statistics of male to female ratio, so as to shorten the processing time of the computer, and to improve the performance of the video monitoring system.The process of gait-based gender classification mainly includes three parts:the first is object detection and image preprocessing, the second is gait feature extraction, and the third is gender classification. In this thesis, the three-part presentation on gait-based gender classification is mainly stated below:(1) Gait database image preprocessingHere the IRIP gait database and CASIA gait database which is used to be compared are applied. For video information in the database, there algorithms, containing Gaussian mixture model algorithm, Codebook algorithm and Vibe algorithm, are used to make objective detection and to finish image preprocessing. The results show that the Vibe algorithm’s effect is best and fast.(2) Gender classification based on elliptic model parametersThis method only applies to90degree’s gait image. The image is divided into seven areas, and4feature parameters are gotten by ellipse fitting for these seven areas. Then parameters are composed of feature vectors of the image, SVM classifier is used to make gender classification. In IRIP gait database, gender recognition rate is97.65%, and the gender recognition rate in CASIA gait database is91.16%.(3) Spatial and Temporal Feature Matrix FusionThe method is applied to8angle views of gait image. First of all, LLE is used to get gait period of the image sequences. Then GPCI both in time and space are extracted, in time,8different angles’GPCIs are extracted; in space, the same frame GPCIs of one gait period are gotten, then spatial and temporal feature matrix is fused. Finally, KNN classifier is applied for gender classification, and the biggest recognition rate can be up to98.25%.