Vehicle Detection And Tracking Based On Mean Shift And Surf

This paper focuses on the detection and tracking of the vehicle and license plate based on characteristics. The detection and extraction of vehicle characteristics is mainly completed with the method of background difference, and the gaussian model is used to establish and update the background. Then the contours of moving objects can be framed out with a rectangular by the means of connected component labeling. The tracking of vehicle characteristics would be completed in two ways:One way is to use the mean shift tracking algorithm; another way is to use the target tracking method based on surf features. When the gray values of the target and the background are very similar or the overlapping region for two consecutive frames is small, mean shift algorithm for target tracking is prone to false matches or no marches, but the tracking method based on surf features is not affected by these factors. In addition, surf feature is invariant to rotation, scale zooming and brightness changes, and can remain stable on the transformation of the view-point, affine transformation and noise. We can get high accuracy for targets match using surf features. The detection and tracking of the license plate is designed to use the method with the extraction and match of the surf feature.In this paper, OpenCV and Visual C++6.0are used to build experimental platform. For the detection and tracking of the vehicle characteristics with the same video, The tracking using mean shift iterative algorithm or the surf features approach either has fast processing speed, and the processing speed difference is small. The tracking rate for the detected vehicles is up to90%. The processing speed of the license plate detection and tracking has the relationship with the video size. When the surf features of the license plate can be extracted accurately, the smaller video will have the faster processing speed. With this method the detection rate and the tracking rate are both up to90%.