Recognition Of Human Body Postures In A Fixed Scene

Intelligent surveillance is a new technology using information technology, with content analysis. Monitoring system with motion object recognition can ignore a lot of useless information, and extract key information then automatically make a judgment according to the rules, thus saving a lot of labor costs. Human gesture recognition technology can be applied to home monitoring, such as the recognition of fall conditions of elderly alone at home.This thesis studies the body posture detection on a fixed scene and related technology. Design a gesture recognition method based on support vector machine(SVM). This method put the process into foreground extraction, feature extraction, classification and gesture recognition three steps. The first part studies the theory and algorithm, improving the Codebook modeling method, making it closer to the actual use. When a certain number of pixels’ luminance changes in the image, we identify it as the light switch moment, and use the video frame at this time to update the background model. Through experience, set the background image update rate. In the second process, the characteristic angle method is presented as a description method: establish the coordinate system with setting the center of the circumscribed rectangle in the foreground image as the origin, calculating the tangent of the four quadrants that the angle between the origin of the centroid of the four image and the connection with the x-axis, with the four a tangent to said gesture characteristics. Meanwhile, the paper studies the feature extraction, by calculating the geometric characteristics, such as aspect ratio, contour length and area ratio, eccentricity, etc. Constitute the feature vectors using these parameters. The third part uses a support vector machine approach. Based on algorithms require, a variety of body posture image samples is collected to extract sample characterization data set. Using this data set to train classifier. The classifier is applied to the detection process, so as to achieve the purpose of gesture recognition.The recognition system designed by this thesis is more proximate to the actual application. A complete system with prospect of testing, post-processing, classification and recognition is integrated. The recognition of such standing and walking, bendwalking, crawling and some other gesture are achieved. Recognition rate is higher. This paper can be simply summarized as: complete a gesture recognition system, using the camera to identify the video moving object to determine the dynamic target posture by the classification algorithm. Meanwhile, feedback the gesture recognition result to the monitoring video stream and finally a good effect is obtained.