Research On The Image Detection And Action Understanding

The technique of moving pedestrian detection and tracking in urban traffic scenes is one ofthe key techniques for Intelligent Vehicle Navigation System. And the research on it will do greatcontribution to the development of Intelligent Vehicles and the safety of city traffic.To ensure the safety and smoothness of the pedestrian flow, this dissertation is focused on thestudy of moving pedestrian detection and tracking in urban traffic scenes by using the data from afixed CCD camera. Firstly, this dissertation describes the process of extract the pedestrians fromthe complex scenes and how to track the pedestrians. Then, the method of pedestrians' motionanalysis is discussed.The self-adaptive background subtraction method and the motion area segmentation methodwere adopted here to segment the moving pedestrians. On the basis of background differenceimage, the dynamic multi-thresholds method and the norphologic operation were used to extractthe pedestrians. So the noises brought by the difference images were restrained. And also, therupture in the images including pedestrians can be joined together. Then, the pedestrians can beextracted exactly.During the process of matching, the template for pedestrian matching and recognizing inimage sequence was presented by fusing several characters of targets. A new method based ongray model GM(1, 1) was proposed to predict the motion of pedestrians. So the detecting area isreduced greatly. The center of the tracking window was elected as the tracking point representingthe pedestrian. The influences of the pedestrians' self motion were restrained, so the robustness ofthe tracking method was improved. The experiment results of tracking are satisfying. What's more,the method of extraction the parameters of pedestrian traffic was discussed. This is necessary tocomprehend the pedestrians' motion. It is the theoretical basis to solve the existing problems innavigation system of Intelligent Vehicles and the monitor and control system of traffic.Experimental results of two real urban traffic scenes showed that this method could detectand track pedestrians robustly. The results indicate that this method is effective.