Research Of Preceding Moving Vehicles Detection And Tracking Based On Vehicular Camera

With the dramatic development of road traffic, it provides a great conveniencefor people’s living, but following the traffic jams, safety and pollution become moreand more outstanding, as a result of threating to people’s life and property security.Preceding moving vehicles detection and tracking is an important research area inthe field of intelligent transportation system. The real-time accurate drivingenvironment is obtained through a vehicular camera which is used to detection andtrack the preceding moving targets, which is popular in the domain of intelligentvehicle safety driving, autonomous navigation and traffic system research.In the preceding moving vehicles detection and tracking system, There aremany difficulties in the vehicles detection and tracking, such as the externalenvironment causing interference, blocked other vehicle, and how to optimizealgorithms based on DSP platform. For these problems, this paper improvescorresponding algorithms, and thus forms a real-time and accurate precedingmoving vehicles detection and tracking system. Major works are as follows:1.The paper presents an improved shadow vehicles detection approach. Basedon the shadow of vehicles close to the bottom of vehicles stably. The two adaptivethresholds segment the information of vehicles’ shadow, in order to create the areaof vehicles. For ruling out false areas exactly, according to Sobel vertical edges andthe method of vertical integral projection, it demarcates vehicles’ scale and positionaccurately. Then, combined with the vehicles’ vertical edge character and complextexture characteristics, vehicles are verified by calculating the entropy valuenormalized symmetry measure of the area.2.Vehicle tracking based on Kalman filtering. The size and position of thevehicle are treated as the vector of Kalman filtering, the filtering predicts the sizeand position of the next frame. In this paper, the method based on the number ofshadow mutations judge the number of vehicles, which solve the problem oftracking several targets effectively. According to the characteristics of edges andintegral projection, it can ensure the target area, and calculate the area of NMI andgray entropy value which are on two levels of judgment, so as to realize thevalidation for tracking vehicles. Our tracking method not only reduces the searchspace but also solves the phenomenon of shelter effectively.3.Consider the prospect that the application software implemented on DSP, the system is transplanted to DM6437successfully. In this paper, we use some methodsto optimize the algorithm, and make its running on DSP platform to be real time.Experimental results show that the method can detect and track based onvehicular camera fast and effectively.