Research And Implementation Of Park Assist System Based On Vehicle Look-around Algorithm

With the rapid development of ADAS(Advanced Driver Assistance System),onboard computing resources are increasingly scarce.In order to reduce the occupation of on-board computing resources,this paper proposes a vehicle surround view generation algorithm based on heterogeneous computing.GPU(Graphic Processing Unit)is used to speed up the calculation of compute intensive algorithm,and some traditional algorithms are optimized to improve the robustness of the system.In order to meet the strong trend of Internet of vehicles and cloud computing,the paper establishes a vehicle surround view generation algorithm based on Internet of vehicles.In order to improve the integration process of ADAS,the parking space recognition based on look-around image is deeply studied,and the accuracy of parking space recognition is improved.The paper focuses on the generation algorithm of vehicle look-around based on heterogeneous computing,the generation algorithm of vehicle look-around based on Internet of vehicles,and the parking space recognition algorithm based on look-around image(1)Firstly,the paper improves the traditional vehicle surround view generation algorithm,and proposes a vehicle surround view generation algorithm based on heterogeneous architecture.In order to improve the overall efficiency of the algorithm,the paper uses heterogeneous architecture for parallel computing of some computationally intensive algorithms,and improves the steps of the partial vehicle surround view generation algorithm to improve the robustness of the overall algorithm.An improved top view transformation algorithm based on geometric constraints is proposed to improve the quality of the top view transformed image and reduce the complexity of the subsequent image alignment and fusion;In the image alignment and fusion algorithm,the mosaic template is improved to improve the utilization of the original image information.(2)The paper constructs a vehicle surround view generation algorithm system based on the Internet of vehicles,which uploads the complex logic to the server with high computing power for calculation,and the vehicle client only carries out the necessary information acquisition and display,so as to greatly reduce the occupancy rate of vehicle CPU(Central Processing Unit)and improve the response ability to emergency tasks.(3)On the basis of the formation of the look-around image,the parking space recognition is further studied.Firstly,the double constraints of line detection and corner detection are used to determine the inflection point of parking space,so as to improve the recognition probability of inflection point of parking space and reduce the probability of false recognition.Then,the improved parking space template matching algorithm is used to determine the type and location of parking space.The experimental results show that the heterogeneous architecture of the lookaround image generation algorithm proposed in this paper can greatly improve the utilization of the system;in the part of the look-around image generation algorithm based on the Internet of vehicles,the algorithm proposed in this paper can correctly transmit the video stream,and it can achieve the real-time requirements due to the use of multithreading technology;in the parking position recognition stage,line detection combined with corner detection The double constraints can accurately identify the corner of parking space and greatly reduce the probability of false identification.The proposed method is robust and feasible.