A Sports Classification Algorithm Combining LBP Characteristics And Motion Characteristics

Take various kinds of human actions for classification and recognition effectively from the video sequence, is the basic research task in the field of intelligent video surveillance, virtual reality, human-computer interaction, ect. With the in-depth study, human action analysis and recognition has become a hot research area in computer vision. The main task of human action recognition is to use the computer to process, analysis and study the collected original image or image squence and understand the moving body’s actions in the video, which can be used in the intelligent securiy monitoring system, human-computer interaction and public security fields, prosessing extensive research value. On the basis of in-depth studying of the moving targets’ capture algorithm in the existing video image, moving target classification and recognition algorithm fused with LBP feature and motion features was proposed, algorithm mainly consists of three parts: moving target detection, feature extraction and behavior recognition.In the study of moving target detection, this paper carried out comparative analysis of optical flow method, frame difference method and background subtraction method. According to using frame difference method to extract moving targets could emerge incomplete detection target and are likely to bring out holes, fused with three frames subtraction mothod and background subtraction method in moving target detection was proposed. Studies had shown that fused with three frames subtraction mothod and background subtraction method could obtain the moving objects which cotaining clearer and more complete information.In the study of motion features extraction, extracting the LBP histogram feature of the block video sequence in the first place, then let the processed block histogram feature combine into the moving targets’ LBP feature by the block order, then get the behavior recognition vector by concatenating the moving humans’ LBP feature with the speed feature of the center of mass. After the recognition features of the moving objects were obtained, we could use the BP neural network classifier to begin with classify and recognize.The recognition experiment were carried out in the weizmann database and the KTH database, through extracting the feature vector of the image video sequences in the two databases respectively, selecting training samples and test samples, training and learning was done on the classfier. After the classifier could stay convergence, the experiment research of human behavior classification and recognition was done using test samples. The experiments verified the effectiveness of the proposed algorithm and analyzed the effect of image block number on recognition rate. The results had shown that the moving classification algorithm fused with LBP feature and motion features presented in this paper could detect moving targets and identify behavior accurately and rapidly, and when the block is 5 ? 5, the recognition rate reached 89.94%, which could satisfy the need of actual applications.