Research On Method Of Video Detecting And Tracking Of Pedestrians

In the Intelligent Transportation System computer vision technologies are now widely applied, among which video detection and tracking technologies are the key points. Video tracking is the process of detecting and locating the target and based on video detection. In our country the urban traffic is featured by mixed flow with vehicles and pedestrians, so this thesis aims to investigate video tracking with the object of pedestrians based on the Mean Shift algorithm.This thesis puts forward a method to detectpedestrian based on detecting and locating the head.This method makes uses of mixed color model, Canny edge detection algorithm and Hough transform algorithm to calculate the location and size of the head.Traditional Mean Shift algorithm is a nonparametric estimation method based on density gradient and modeling in a kernel histogram manner. Usually, the target that is going to be tracked is manually selected in the first frame of a video sequence and at the same time establishes the target histogram model. Next the algorithm makes uses of the Bhattacharyya coefficient as similarity measure and then search for the most similar areas with the model iteratively in subsequent frames. Mean Shift algorithm is widespread concerned and fit to track pedestrians due to its advantages of strict in theory, simple and easy to implementation, and better tracking performance among all the tracking algorithms. However there are some defects. Firstly, the tracking target cannot be selected automatically. Secondly, the search window is not able to adjust to the size of the target by itself. At last, the algorithm regularly lost its target under the circumstances the target moves very fast or is sheltered.This thesis puts forward an improved Mean Shift algorithm to achieve the effect of selecting target automatically based on the detection of pedestrian head. The head of pedestrian is detected based on which the search window is confirmed in the first frame. The search window which contains the target is then confirmed on the side of location, length and widththrough further transform.This thesis works out the Bhattacharyya Coefficient Judging Method to determine different situations the target may meet such as high speed. This method is achieved by analyzing the change of similarity between search window and background and aimed to solve the problems respectively. In the situation of high speed and this thesis adapts Kalman Filter to ameliorate Mean Shift algorithm.Kalman Filter is used as an auxiliary mean in the case of normal condition and to predicate the location of the target while the target is in high speed. Mean Shift algorithm uses the predicated location as the start point of iteration and then makes the iteration result as the modified value of Kalman Filter. By the improved algorithm the correct location is determined after which calculation in the next frame is started. The examinations prove that this method impoves the effct of tracking targets in high speed effectively.