Video-based Vehicle Detection And Tracking Technology Research

In order to solve various problems caused by the rapid growing of surface traffic, the research of Intelligent Transportation System comes to an issue of consequence. As an important and core part of ITS, the detection of traffic flow parameters becomes a research focus all over the world. In this paper, according to the key techniques of ITS filed, some relative issues in city vehicle detection and tracking are studied and analyzed on the base of video images, and our research work focus on the following aspects:(1) A mixed Gaussian background model and updating method integrated inter-frame difference is proposed, which can reconstruct background from video sequences quickly and accurately. Mixed Gaussian background model updating algorithm is presented to deal with abrupt changes and gradual changes, which can update the background nicely, and can effectively overcome the influence of illumination changes, clutters of circumstance to the background model. At the same time, the integrated of inter-frame difference method can effectively solve the mixed-Gaussian model shortcomings of poor real-time and has strong robustness.(2) Foreground was extracted and segmented based on the model of background. Morphological filter is used to eliminate of noise and empty. Based on the principle of the shadow, the video frame image is converted to HSV color space. And an adaptive shadow detect algorithm is proposed, which has a good real-time performance, and can adapted to the light and shadows changes well.(3) According to the way of moving target appears in the scene, it's given the three status of the moving, and discussed the relationship between the locations of the vehicle in video, and based on that vehicle tracking strategy was given. In order to meet the system requirements for real-time, this pepar uses the region feature tracking method, And extendes kalman filter. The expensed filter not only take account in the current characteristics of the target, but also increase the predictive value, so it can track and predict the target more accurately. Viewing of the characteristics of vehicle tracking, it acquainted parameters both geometric characteristics and gray color parameters. Similarity operator was used for multi-target tracking, and eliminated the problems of occlusion partly.The experiment's result shows proposed in this paper can identify, separate and track vehicle effectively from the video sequence. The algorithm has strong real-time and robustness.