Research of Video Detection System for Traffic Flow Real-time Prediction Based on the FPGA

Under the sustained development of economy and society, people’s living quality isimproved. As the advancement of urbanization, large numbers of people move into citieswhich results in the growth of the traffic volume, timely and accurate prediction together withquick and efficient control of traffic flow is a significant way to improve traffic congestion.Being capable to obtain all kinds of traffic information and being able to analyze and handlethe information are the premises of traffic flow timely prediction. Traffic video detectionsystem starts from image processing and uses its characteristics to analyze traffic problems. Itis an important direction of traffic research, analysis and management and it is a keyapplication field of image processing technology.The purpose of this paper is to provide timely and efficient information for real-timetraffic flow prediction and control. Image processing and FPGA were used in this dissertationto study the vehicle video detection system. This dissertation first introduces and analyzes themethods and parameters in traffic flow control and prediction. The situation of vehicle queuein the crossing has been studied and prediction model has been made. Then, the length ofqueue is collected. Aiming at the current method, which is easily affected by weather andshadow, algorithm is improved so as to reduce the interference mentioned above. This paperexplains the principles of the realization of the algorithm in details and uses Matlab to dosimulation verification, the result shows that the improved algorithm can effectively extractthe vehicle queue length from the intersection. In order to avoid the problems such asdesigning difficulty, long development period and lack of hardware and softwarecollaborative simulation in traditional FPGA, this paper adopts a new design concept, whichbases on Model Based Design idea, using Altera corporation DSP Builder design tools,realizes vehicle queue length extraction algorithm in the FPGA. This paper compares theperformance of DSP builder experimental method with traditional methods and eliminates thedoubts of low efficiency of the DSP Builder. Finally, comparing with the results of thissystem and Matlab simulation results, the outcome shows that the two ways to extract vehiclequeue length are almost the same. It is proved that the video detection system, which is basedon FPGA, can accurately extract the vehicle queue length and has practical reference value.