Research And Implementation Of Real-time Vehicle Target Detection System Based On FPGA

Vehicle detection is a key technology in the Advanced Driver Assistance System,It is of great significance to reduce the incidence of traffic accidents.In the detection of vehicle,real-time performance is an important indicator.During the driving of the car,only timely detect vehicles in front.And feedback the information to the driver to effectively avoid traffic accidents.In all kinds of embedded system,FPGA has the advantages of low power consumption,fast processing speed,small volume and so on.In particular,its parallelism and pipeline processing make it possible to meet the requirements of real-time video processing.Vehicle detection system based on FPGA is the keystone of the study.The system uses the Xilinx Spartan6 series XC6SLX45 chip as the processor,Using the OV5640 as an Image Sensor.on this basis,By using FPGA parallelism and pipeline processing features,build a real-time vehicle detection system.The main research work of this paper is as follows:(1)This paper analyzes the embedded image processing system based on FPGA,According to FPGA pipeline processing features and parallel processing features,divided the modules of the whole system.Design the camera configuration module,data cache module,preprocessing module,and video display module in the system.The real time video display processing based on FPGA is realized,which builds a basic platform for the implementation of the subsequent vehicle detection algorithm.(2)This paper analyze the target detection algorithm,Using convolutional neural network and SVM classifier for target detection.And the ROI Pooling layer is introduced to solve the problem of fixed size of input image in traditional convolution neural network.the multi-scale window is slid on the extracted feature map,extracting features from multiple scales,normalize the features of multiple scales to the same scale,and send them to the classifier for classification.Analyze the logical relationship between the timing and function of each module,and analyze the hardware structure of each module.Completed the hardware design of convolutional neural network module,ROI Pooling layer module,and SVM classifier module,at the same time the resource optimization of the convolution neural network module is carried out.(3)This paper analyze the overall system resource usage,and each module is tested for timing,at the same time the efficiency and functionality of the entire system are analyzed and tested.Through testing,the system can achieve real-time detection of vehicle targets.