Research And Application On Vehicle Detection And Tracking Algorithms Based On Video Sequence

Vehicle detection and tracking in traffic monitoring system is the most important part and critical step of intelligent traffic management. With the continuous development of computer technology and image processing technology, vehicle detection and tracking based on Video Sequence has attracted more and more attention,and become one of the research focuses currently.After the study and research on the classic algorithms of vehicle detection and vehicle tracking based on video sequence, some algorithms and techniques for vehicle detection and tracking have been analyzed and improved to deal with the existing problems. As a result, a new method for vehicle detection and tracking in municipal transportation was presented. The main contents of this paper include such aspects as follows:(1)A new method of background model building was presented. According to the method, background models are built by using background subtraction and Inter-frame subtraction. Firstly, only the frames according with the demands are chosen for background initialization frame so that the accuracy of background model could be enhanced, and the number of the chosen frames is variable. Secondly, only the pixels which are regarded as background pixel through background subtraction can be chosen for background model updating, and the coefficient used for updating the exist background model is dynamic. The variation of coefficient depends on the variation of actual background, so the background model can be updated in time even when the actual background changes suddenly. In addition, the image segmentations are done by dynamic thresholds to avoid serious mistake caused by sudden change in actual background. Experimental results demonstrate that this algorithm can detect the vehicles on video sequence more accurately.(2)A new method of Vehicle tracking was presented, which is multi-feature matching based on region algorithm. According to the past moving features, the center position of tracking objects can be located approximately in the next frame. Accordingly, the field of object tracking is narrowed down. In the process of vehicle tracking, both gray images and binary images are used for the tracking of the moving Vehicle. In the matching stage, the features of center position, size and gray-value can be used in the object matching. The experimental results show that this algorithm can track the moving vehicle efficiently.