Research On Moving Vehicle Detection And Tracking In Dynamical Scene Based On Optical Flow

Research on moving vehicle detection and tracking is the hot spot and takes very important place in computer vision system. With the development of intelligent vehicle system, People proposed higher request on vehicle detection. So far, there are many pressing problems need to solve. Moving vehicles in dynamical scene is more difficult because there are two independent motions; the issues of stability and accuracy on detection have large space to study. This paper proposed an effective method on vehicle detection and tracking.Firstly, paper analyzed the advantages of major methods of moving vehicle detection in dynamical scene, though taking the accuracy and computing speed into consideration to propose the method of feature point optical flow. This paper adopted Harris corner as feature point, and then Pyramid Lucas-Kanade Optical Flow was adopted to generate feature point optical flow. It can reach a compromise between accuracy and computing speed though choosing different parameters.The paper introduced Vector Quantization (VQ) to cluster the optical flow field. In order to there are great differences in direction and magnitude between vehicle feature point optical flows and background ones, we used the direction and magnitude of optical flow field as input variables, and adopted VQ to cluster the optical flows and combined Euclidean distance and similarity coefficient as similarity measure. This method greatly improved the accuracy of clustering. Eventually, this method eliminated all the mismatched optical flows though computing the variance of each class, and it can also extract the vehicles from background though variance.Finally, this paper studied the moving vehicle tracking algorithms, it generalized the major methods of vehicle tracking such as Kalman filter, Cam-shift algorithms firstly, and then adopted Kalman filter to act as the tracking method, it can avoid the problems of losing tracking and the situation of vehicles blocked. Eventually, the improved method was used to solve the problem of color disturb. Though that, the continuous tracking was reached.Though multiple tests, the method proposed can detect the moving vehicles and track accurately and steady with good robustness and can reach real-time request.