Research On The Video-Based Real-Time System For Vehicle Detection And Classifiction

With modern industrial development and environmental impacts of automobile traffic in urban areas, Intelligent Transportation Systems (ITS) has gradually become an important field of modern city-building at home and abroad.This paper focuses on the study of vehicle detection system based on image processing. The theory and realization of the key technologies of this system is presented and the feature of real-time and veracity emphasized. Two improved algorithms of moving object segmentation and a detailed algorithm of vehicle classification have been proposed.The main work and innovation in this paper can be summarized as follows:1) We proposed a method on detection of moving object using symmetrical DFD, adaptive optical flow and snake algorithm. First get the motion area of the moving object by symmetrical DFD, and estimate the optical flow of this area; then employ the C-Mean cluster algorithm to cluster the result of the optical flow; finally, the initial position of Snake is set automatically from the cluster result. The converged contour of the final object is got by using greedy algorithm. Experimental results show that the algorithm can quickly and exactly detect the multiple moving objects from natural images with complex backgrounds.2) An extraction algorithm of the moving object based on background extraction against the complex environment. This algorithm effectively overcomes the difficulties brought by light and the similar gray level between vehicle and background utilizing the characteristics of the colorful images; Corresponding background update schemes against two different situations are put forwarded and a fast and efficient filling algorithm by using the binary image features is worked out in order to meet the requirement of real-time moving object extraction.3) A feasible vehicle classification system based on the profile contour features of the vehicle. In this paper, the concept of soft sensor based on image information is proposed to replace the traditional hardware device. Further more, by compressing of the vector information using Douglas-Peucker algorithm, the complexity of data manipulation is reduced. In addition, a classification method based on hierarchical precisions is adopted, realizing the fully utilization of the invariable moment, area proportion, length and other important information, which leads to a satisfactory outcome.In this paper, an integrated applicable AVI system is realized and an improved background update algorithm proposed. Besides, a classifier based on the vehicle profile contour feature is designed.The system can be run smoothly on the PC installed with Microsoft Windows XP Professional (SP2), P4-2.8GHz processor and DDR-512M memory. The average accuracy of vehicles classification reaches 90% in the test of two 25 minutes video sequences. The processing time is less than 40ms per frame and the occupancy rate of CPU is lower than 40%. Frame by frame operation is fully achieved.