Parking Space Detection Method Based On Vehicle Around View Monitor

Along with the continuous growth of China’s national car ownership and usage rate,parking spaces have yet to meet the current traffic demand.The problems of narrow parking space and difficulty in finding parking spaces are becoming increasingly prominent,leading to the rising incidence of parking accidents,seriously threatening people’s life and property safety.Blind area is one of the main reasons for frequent parking accidents,and eliminating the driver’s blind area is necessary for improving parking safety.The Around View Monitor has realized the visualization of the area around the vehicle body,effectively solved the blind zone problem,and has been gradually assembled in market vehicles.Because of the visualization characteristics of the Around View Monitor,based on the construction of the Around View Monitor,this paper combines the long range of around view image with the depth convolution neural network to achieve parking space detection around the vehicle body.The main research contents of this paper are as follows:(1)The foundation for building the Around View Monitor.Firstly,the hardware composition and software configuration of the Around View Monitor are analyzed;Secondly,the imaging principle and model of the fish-eye camera,an environmental sensing device,are introduced,and the projection process of the shooting target in various coordinate systems is deduced in detail with the isometric projection model;Finally,Zhang Zhengyou calibration method is used to calculate the distortion coefficient and internal parameter matrix of the fish-eye camera,and the calibration area for experiment is designed to obtain the external parameter matrix of the four-way fish-eye camera,to realize distortion correction,perspective transformation and registration of the four channel fish-eye image.(2)The fusion method of around view image.According to different application purposes,the Around View Monitor is subdivided into Short Range-Around View Monitor System and Long Range-Around View Monitor System,each with different imaging area range and characteristics.In order to avoid the seam,different fusion strategies are developed,and "clock scan image fusion method" and "tan weighted image fusion method" are proposed to meet the different requirements of the two Around View Monitors.In order to verify the effectiveness of the image fusion algorithm,a combination of subjective and objective assessments is used to compare and evaluate the fusion effects of the various methods.(3)Parking space detection method.According to the inherent characteristics of the parking space and the display characteristics in the Long Range-Around View Image,combined with the actual parking demand,an efficient parking space detection strategy is developed.Based on the YOLO v5 s target detection network,the detection model of the parking space entrance area was trained.The type,availability,location of the entrance line and the corner area of interest of the parking space could be obtained by using the detection results.Furthermore,LSD straight line detection method and regional threshold direction identification method were used to identify the direction of the dividing line segment in the corner area of interest of the parking space corner.A comparative experiment is designed to verify the superiority of the algorithm.(4)System experiment verification.According to the experimental requirements,the hardware equipment included in the system is installed on the real vehicle,and six experimental scenes including parking spaces are selected,and the Around View Monitor and parking space detection system are verified simultaneously.The natural vehicle experiment verifies and analyses the feasibility of the Around View Monitor construction method and parking space detection method proposed in this paper.