Research On Visual SLAM Technology For Automatic Parking Environment

As an important part of the Advanced Driver Assistance Systems,the Automatic Parking System is also one of the most critical technologies for achieving autonomous driving.The sensing part of the automatic parking system,as the front end of control and planning,needs to solve the classic problem of "Where am I?".Simultaneous Location and Mapping(SLAM)can solve this problem.SLAM can be divided into visual SLAM(v SLAM)and radar SLAM according to different sensors.v SLAM has been widely researched and developed due to its low cost,diverse installation methods,and rich image information.This paper proposes a method based on v SLAM to solve the perception problem in an automatic parking environment,and it also has better robustness in complex scenes with low light or alternating light and dark.The research work in this paper mainly includes:First of all,because of the complex scene of low light or alternating light and dark may appear in the automatic parking environment,the v SLAM system in this article needs to be able to handle this complex scene.Based on the theory of Retinex,this article analyzes from the principle of object imaging and the LIME algorithm is used to solve this problem,and the weighting coefficients in the algorithm are designed in combination with the automatic parking scene,and the time complexity of the algorithm is considered to accelerate the optimization.Finally,some low-light pictures were optimized to verify the feasibility of the LIME algorithm.Then,the common problems of v SLAM,that is,easy to lose targets and easy to appear scale drift during tracking,are analyzed and studied.A method to reduce the probability of tracking failure by using a predictive model is proposed.A key frame selection strategy suitable for the parking environment is proposed.For stereo matching,firstly,a point of good quality is selected based on multiple errors,and then use rough to fine Method,and then,the correlation of the images is calculated to match and update the inverse depth of the pixels on the map,finally,in the map optimization module,error analysis is performed through two-way Sim3 detection to solve the problem of scale drift.Finally,the camera model RER_USB1080P01-LS43 developed by Volkswagen Sagitar and Shenzhen Ruiershi Technology Co.,Ltd.was equipped with an experimental platform.In the evening,multiple parking lot data were collected around the Automotive Engineering College of Nanling Campus of Jilin University and verified The feasibility of the LIME algorithm in a low-light parking environment enables v SLAM to be performed in the complex environment of low-light parking,and experiments show that the proposed algorithm is superior to the traditional algorithm in a parking environment.