Design And Implementation Of An Automatic Parking System Based On A Panoramic Camera

The rapid development of the automotive industry has made automatic parking systems one of the research hot-spots.The popularity of automatic parking systems is constrained by factors such as cost and sensors.At present,the automatic parking system faces two challenges: First,traditional image processing methods are greatly affected by light,etc.The parking space detection and positioning of underground garages are often completed by laser radar,fixed-end cameras,GNSS,etc,and the high cost of sensors,Large amount of data,easily affected by signals and other issues,affecting the popularization and application of automatic parking function.Secondly,during the parking process,the constraints are strong.The vehicle is limited by the minimum steering radius and obstacles.The system requires spacious parking spaces.The spacious parking spaces are easy to park,but the narrow road width is difficult to succeed.Therefore,this paper proposes an automatic parking system based on a panoramic top view: first,the parking space detection algorithm based on the panoramic top view is used to detect parking space information,and then the grid-based image positioning algorithm is used to achieve the relative position of the parking space and the vehicle.Then determine whether the parking space is free through the dynamic threshold,and finally plan the parking path for the available parking space to realize the parking path planning under the narrow road width condition.The main research work of this article is as follows:Aiming at the problems that the positioning methods of vehicles such as GNSS in underground garages are limited by signals,the cost of lidar and other sensors is high,and ordinary rear cameras cannot obtain complete parking space information,this paper proposes a parking space detection based on panoramic top view algorithm.Use four fisheye cameras to obtain information around the vehicle,and establish a panoramic top view around the body.A multi-scale loss function convolutional neural network incorporating linear convolution layers is proposed to improve the detection accuracy of parking spaces and reduce the cost of sensors.For the detected parking spaces,this paper proposes a grid-based image positioning method to achieve the relative positioning between the parking spaces and the vehicles,eliminating the influence of the signal,obtaining the size information of the parking spaces,and combining the dynamic threshold of the parking space idle Judgment method to judge whether the parking space meets the parking conditions.The path planning of automatic parking often needs to satisfy various constraints.Aiming at the problem that it is difficult to realize automatic parking under the condition that the road width is narrow and the existing parking spaces are parked,this paper proposes a parking path planning method for vertical parking spaces under narrow road width conditions.This method learns how the driver can park the car by reversing the vehicle multiple times under narrow road width conditions.Starting from the reverse process of the parking process,deducing from the last parking state to the initial parking state,calculating the key point information in the parking process,and calculating the parking path required in each stage separately,and finally combining each key point Complete the path planning with the intermediate path to realize the parking path planning of the vehicle under the narrow road width.In terms of system,this paper designs a complete automatic parking system based on the laboratory mobile platform,so that the mobile platform can complete automatic parking under the conditions of proportionally reduced parking spaces and corresponding road widths.