Research On Algorithms About Vehicle Detection And Real-time Tracking

Nowadays, with the social improvement and higher flexibility, as well as the automobile popularity based on the development of the automobile technology, it is an inevitable result for Intelligent Transportation System (ITS) to appear and draw public attention. As one of the most important parts of ITS, Video Surveillance System is aimed to detect and track the targets in the scenes. This paper proposed a video-based algorithm of vehicle detecting and real-time tracking. It has been proved that this algorithm shows robust and real-time performance in vehicle detecting and tracking, with the elimination of disturbance such as the illumination vibration etc.The detecting algorithm in this paper used the concept of background subtraction combined with the feature of vehicle edges. It is the first step to establish the background model. Instead of those traditional background models, a model based on zone-distribution method was introduced to extract the initial background image without any prior knowledge, which has both the accuracy of probability statistics methods and the briefness of the mean and median models. In the second step, the feature of edge is introduced to detect vehicles, for those edges contain plenty of information of targets with non-sensitivity to illumination. Instead of traditional edge-detecting methods, this edge-detection algorithm in this paper is based on multi-structure element morphology, which can preserve more edge details with more efficiency for image denoising. A symmetrical differencing is used for vehicle segmentation with the weakening of noise and the enhancement of target edges.A Kalman-filter model was introduce in the vehicle-tracking step, with the eigenvalues of rectangle Centroid and area. In the Kalman model, the vehicle coordinate, displacement, velocity and acceleration were selected as parameters in order to reduce the potential regions. Besides, the eigenvalues of rectangle Centroid and area can compensate the loss of pixels in vehicle tracking.The experiment results show that this algorithm can detect and track vehicles accurately in the complex and noisy scenes, which may be put into practice in future.