Research On Vehicle Detection System With Monocular Vision Based On FPGA

With the number of vehicles increasing,traffic accidents and traffic jams are becoming more and more serious.Intelligent transportation system with better dispatching and safety capabilities has attracted much attention.Advanced driving assistant system,on which computer vision,path planning,automatic control and other technologies are applied to make driving safer and more efficient,is an important part of intelligent transportation system.Vehicle detection based on computer vision technology is the core issue,which affecting the development of intelligent driving assistance system.Recently,the breakthrough of deep convolution neural network has greatly improved the efficiency of computer vision algorithm.The algorithm based on convolution neural network has become the mainstream algorithm in the field of object detection.However,due to its huge computational and storage requirements,the application of convolution neural network for vehicle detection on vehicle platforms is limited.In this thesis,the application of deep convolution neural network for vehicle detection is studied,and the system-level optimization method based on hardware-software cooperation is adopted.At the level of algorithm architecture,aiming at building a vehicle detection model with high speed and high accuracy,a lightweight vehicle detection model which is suitable for embedded chips is built and trained,and then the algorithm is optimized and quantified.The test results of the model show that the proposed vehicle detection algorithm can greatly reduce the computing and storage requirements of the detection model while maintaining high detection accuracy.On the hardware level,a high performance hardware platform for vehicle detection model based on deep convolution neural network is developed according to the fixed-point computing optimization capability of programmable gate array.Based on ZCU102 development board,the hardware configuration and programmable logic,as well as the development of vehicle detection function are completed with SDSoC development environment.