Research On Self-driving Cars Positioning System Based On VISLAM In Road Environment

In the field of self-driving cars,how to obtain high-precision positioning information of vehicles is the key to realize unmanned driving.Visual Inertial Simultaneous Localization and Mapping(VISLAM)technology is an important method to achieve precise positioning of robots.It can achieve centimeter-level positioning in an indoor static environment,but the positioning effect is poor in an outdoor road environment under the influence of dynamic targets.Therefore,in response to the low-cost and robust positioning requirements of self-driving cars in road dynamic environment,this paper designs a VISLAM positioning system that integrates dynamic feature filtering to improve the positioning accuracy of self-driving cars in road dynamic environment.The main research of this paper is as follows:Firstly,Vehicle kinematics model establishment and VISLAM system research.The observation model is established for the monocular camera and the Inertial Measurement Unit(IMU),the vehicle kinematics constraints is established for the motion characteristics of selfdriving cars,at the same time,the coordinate system required by the VISLAM system of this article is established.Finally,the algorithm framework and mathematical expression of VISLAM are analyzed.Secondly,Aiming at the shortcomings of the poor positioning accuracy of the VISLAM system affected by dynamic targets,YOLO Filtering(YF)and YOLO Mean Shift(YMS)dynamic feature point filtering algorithms are designed.YF dynamic feature point filtering algorithm is designed based on YOLOv4-Tiny target detection network,and experiments are designed to analyze the effect of the algorithm.Aiming at the shortcoming that YF algorithm cannot detect the target state,this paper proposes a YMS dynamic feature point filtering algorithm based on YOLOv4-Tiny target detection network and Meanshift algorithm.This method clusters feature points according to the speed consistency characteristics of static feature points in the same frame of image,and fully filters out the dynamic feature points with fuzzy states.Finally,the design experiment verifies the effect of the YMS algorithm.Thirdly,Aiming at the outdoor road dynamic environment,a YMS-SLAM positioning algorithm is proposed.Based on the VINS-Fusion algorithm framework,the YMS-SLAM positioning algorithm is designed,and the YMS dynamic feature point filtering algorithm is introduced into the YMS-SLAM system to reduce the impact of dynamic targets in the environment on the system.At the same time,GNSS information and IMU information are introduced into the system as constraints in the loop detection module to improve the speed and accuracy of the loop detection of the system.Fourth,the positioning accuracy of the YMS-SLAM algorithm and the VINS-Fusion algorithm are compared using the autonomous driving KITTI data set.In order to further analyze the actual operating performance of the YMS-SLAM algorithm,this article does real vehicle experiments on the road environment of self-driving cars.The results demonstrate that the proposed YMS-SLAM algorithm in this paper has higher positioning accuracy and stronger robustness than VINS-Fusion.