Research On Feature Extraction And Behavior Recognition Of Moving Object

Based on the continuous development of intelligent information, the intelligent analysis of foreground objects in sports scene becomes particularly important. The feature extraction and recognition technology of moving target directly affects the practicality of the intelligent analysis system, and at the same time, it is also the core components of the intelligent monitoring system. Thus the feature extraction and behavior recognition of moving target has become a hot research nowadays. One complete intelligent analysis system generally contains three parts, including target detection, feature extraction and behavior recognition. Considering there are many methods to furnish each part, thus it is particularly important for us to investigate a universal and practical approach for this system.The methods that commonly used for detection process of moving object have been made a deep study nowadays. In view of the shortcomings of these methods, a new algorithm which combines the improved method of W4 with the method of the average value is presented. Considering the drawbacks of classical modeling algorithm of background, it is proposed that making the minimum and the maximum gray level of each pixel with linear weighting, through combining with the initial background which is got from average method, a more stable background model is obtained. It effectively overcomes the influence of shadow and light on the effect of the mutation detection. And the experimental results show that the improved algorithm demonstrated both a highest accuracy and a best effect of detection.As the behavior of moving objects has different characteristics, during the process of the feature extraction, extract the movement behavior of feature value which is less dimension and better for reflecting to describe the target. Extracting the motion features such as speed, the center of mass and the edge features such as perimeter, area, tightness, external rectangle aspect ratio from the moving objects to describe the operation behavior of the target. In order to verify the effect of each description of feature on human behavior, the corresponding experiment study can be carried out. And the experimental results show that the four characteristics including the speed, the center of mass, tightness, and external rectangle aspect ratio could effectively describe the behaviors of the human body.After to detect the moving object in video image sequences, extract the motion and edge features from the moving target to use for statistical analysis. By this way, the characteristic data can be calculated and the threshold value is obtained by experimental method, which is used to distinguish the human behavior. The experimental which studies on the effect of movement features, edge features and the integration of multiple features to recognize the algorithms. Experimental results show that the recognition method by integration of multiple features can accurately and effectively identify the human behavior such as walking, running, jumping, hopping on one leg, or feet jumping, giving an ideal recognition, and the average recognition rate could reach nearly 92.89%.