Research Of Real-time Video Traffic Accidents Detection Based On FPGA

With the rapid development of science and technology, human society has entered theinformation age. Any one research field closely linked with people’s life has become thescientific research focus topic. So, intelligent transportation has become a hot research topic.There are a lot of research theories about intelligent transportation, but most of them arebased on checking the video tape after the accident, which was recorded using electroniceye. This method has bad real-time capability and causes a great waste of human resourcesand financial resources. There are some studies on intelligent traffic accident detectionusing video image processing, but still few. The intelligent traffic accident detection canhelp us deal with the accident in time and save more human resources and financialresources. Therefore, real-time video traffic incident detection has not only theoreticalsignificance, but also great practical value.More and more scholars have favored the use of FPGA (Field Programmable GateArray) for video image processing, because it has parallel processing ability, flexibility andcan be configured. Our main work is building Nios II embedded system on FPGA hardwareplatform to realize the traffic accident intelligent real-time detection, including targetdetection, target tracking, collision detection and accident alarm. The main contents of thispaper are as follows:1、The paper introduces the significance of using Nios II to build a SOPC embeddedsystem on FPGA hardware platform, and it introduces the Nios II embedded system, itsdevelopment environment and development procedure. Meanwhile it focuses on altera’savalon bus that is used in the design of IP cores. And then the hardware resources of FPGAplatform are presented.2、The paper introduces the theoretical study of a vehicle collision detection using videoimage processing algorithms, including background modeling, target detection and collisiondetection. Background subtraction algorithm, binarization algorithm, dilation and erosionalgorithm and pixel elegant algorithm are used in target detection. The collision detectionalgorithm is implemented by analyzing the target’s physical characteristics such as size, shape, center of gravity and speed. In order to verify the feasibility of the algorithms, weimplement the above algorithms using Visual C++on personal computer; meanwhile weimplement functions such as video storage, collision results’ preservation and alarm.3、In the paper, the implementation of collision detection on FPGA platform isintroduced. The workflow of the system and the realization of video input and outputmodule, video pre-processing module, background subtraction algorithm, binarizationalgorithm, dilation and erosion algorithm are described in detail too. The paper describes themodules’ input and output interface and the implement of the core part of algorithms usingVerilog HDL language. Finally, it describes how to use the C language in the Nios II IDE toimplement the results’ preservation and alarm.This paper mainly studies intelligent traffic accident detection on FPGA hardwareplatform. In the paper the use of video image processing on traffic accident detection is stillat the starting stage, and it needs strict experimental environment. So in the experiment weuse road model and vehicle model to construct the experimental environment with relativelysimple background and fewer targets. The results of experiment showed that, the method ofusing video image processing to achieve traffic accident detection on FPGA platform isfeasible, flexible, and efficient and can be configured.