Research On Action-based Pedestrian Identification Via Hierarchical Matching Pursuit And Order Preserving Sparse Coding

In recent years,with the popularity of video monitoring equipment,there are more and more video data every moment. At the same time,there is no doubt that it will consume a lot of manpower and material resources for people to watch the monitor video. So people looking for all kinds of biometric identification technology,hoping that the computer can do the identification of pedestrians task independently. After observing the video data,we found that the video surveillance equipment often hang on high,it causes higher angle of elevation,and more far distance. Some original identification technologies,such as face recognition,are hard to get ideal result. Thus,in the basis of the research on the gait recognition,we proposed a pedestrian identification method based on a variety of actions.Like traditional pattern recognition methods,we mainly solve the problem of how to find a high expression of feature extraction method,and find a classifier which has stronger recognition ability. In this paper,we use the hierarchical matching pursuit algorithm for the extraction of raw video data characteristics,by building a pyramid model,to store multi-scale,spatial information of the video image into the characteristics. And select order-preserving sparse representation as our classifier. Through dumping the space and time information of the video sequence,the classifier can fully take the space information saved in the feature into account. At the same time,we also studied the spatiotemporal order-preserving pedestrian identification.We validate the effectiveness of our proposed algorithm through experiment on a public data set.During the research on the pedestrian identification in visible light,we found that if we have the depth information,it may probably result better recognition effect. In addition, since the launch of Microsoft equipment Kinect,there is more equipment which can produce accurate and cheap depth data in people’s daily life. With the development of science and technology,the depth camera is bound to replace now visible light camera,and become the mainstream video data acquisition device. So we adopt an improved hierarchical matching pursuit algorithm to extract features on RGBD data. And through the experiments on the depth video database MSR 3D Action,we verified the effectiveness of our algorithm. It lays a foundation for the further research in the future.