The Research Of Moving Object Detection Based On The Traffic Video Of Road

In recent years, our country road traffic safety situation is very serious, especially for the urban traffic, and the road monitoring network as the main force of urban security surveillance, transfer the condition of road traffic to road test center through monitoring equipment, so the personnel of monitoring center can take measures to the driving vehicles according to the actual situation on the road. Intelligent surveillance system as an important part of intelligent transportation system can automatically detect and recognize moving object from the traffic video, the video scene is divided into the static scene and dynamic scene according to whether the background moves or not, the detection of moving target is usually conducted in these two scene. In this thesis, according to the characteristics of video sequence, we will use two different methods to study target extraction. main works are as follows:The first part, list several common algorithms of extracting moving objects, and make a detailed description, judging the satisfaction of result by the target integrity in the experiments.The second part, aiming at the two same characteristics that most of the background pixel values changed little and foreground pixel values have the same shift in the two adjacent frames, considering use motion vectors to describe the characteristics of the pixel; while use the method of image registration to calculate point motion vector have a huge calculation, and because of the different selection method or the error of calculation process will be led to false detection. thus proposed the method of using block motion vector to extract the moving object, this method can extract the overall outline of target, finally use dot function to remove the background around the target. This method have small calculating amount, and it’s also can reduce the background moved and target hollowed caused by random factors.The third part, aiming at the two same characteristics that the standard deviation is big to the moving target passed pixel block, and the standard deviation is small to the moving targets don’t passed pixel block in all video frames, we suggest using the standard deviation features to extract the moving object, it can overcome the influence of noise on the detectionresults; because the standard deviation have a huge quantity of calculation, the deviation from mean and standard deviation has the same features and need less computation, so we put forward to use the deviation from mean instead of standard deviation to remove background,and also use dot function to remove the background around the target. Theory of this method is simple but it’s extraction effect is remarkable.From the two methods of this thesis introduced, we can get a whole target, and less background pixels around it, so the testing result is good.