With the increasing requirements for security, intelligent visual surveillance gained more and more attentions. Gait can be used as a kind of biometric to identify human at a distance. Gait recognition is a key element for improving the intelligence of the surveillance systems. Compared with other biometric features, the shortcoming of gait is that it is not robust to variations, such as view, clothing, carrying condition, illumination. View variation is very common in visual surveillance. In the dissertation we investigate view variation in gait analysis and recognition. Some other variations, clothing and carrying condition, are also considered. Besides, some research work is also done on gait based gender classification. This dissertation mainly includes the following issues: 1. Gait recognition is in its immaturity, and there is still no standard to evaluate different algorithms. We propose a evaluation framework to do this work and advance gait recognition technology. The framework contains a large gait database (Data Set B in CASIA Gait Database), 3 sets of experiments and some metrics. The framework can evaluate an algorithm's robustness to view, clothing and carrying condition variations. 2. There are two remaining open problems in gait recognition. One is which view is the most suitable one for gait recognition and why it is. Another is how view angle variation affects the performance of gait recognition. We proposes two models, a geometrical one and a mathematical one, in an attempt to address these two questions, and investigate and analyze the effect of view angle on the performance of appearance-based gait recognition. 3. Currently most gait recognition algorithms are view dependent and not robust to view variation. We proposed a linear model and a non-linear model to synthesize the gait feature from one view to another view. The models can be used when the probe angle is not equal to the gallery angle. Experimental results show that the models can solve view variation problem. 4. Some previous research shows that gait feature can be used for gender classification. Gait feature can be divided into two groups by gender, male and female, so gender can help to speed up gait database retrieval, improve gait recognition accuracy and surveillance systems' perception. We give a comprehensive study on gait based gender classification, which includes experiments by human observers and computer algorithms, comparison of different gait features and classifiers, experiments from multi-view and view invariant gender classification. We also do cross-race gender classification experiments and gained inspiring results.
修改评论