Research On Plane Feature Mapping Based On Lidar Point Cloud Data

Intelligent vehicles(IV)can efficiently provide people with safer and more convenient travel and improve people’s quality of life.At the same time,with the development of China’s socio-economic and engineering application science and technology,people have also put forward higher requirements for convenience,safety,environmental protection and other aspects of life travel,earnestly hope that travel will be more intelligent,and high-level IV can meet people Expectations.The composition of IV can be divided into three parts according to function:environmental perception,decision programming and vehicle control.Among them,the environmental perception of IV is the basis for decision programming and vehicle control,and is a prerequisite for assisted driving and high-level intelligent driving.The environmental perception of IV requires that the basic functions of vehicles are to map the surrounding environment and accurately identify the location of the vehicle in the environment.Lidar SLAM,namely,simultaneous localization and mapping based on lidar sensors is one of the core technologies in intelligent vehicle environment perception.It refers to the process that IV rely on their own lidar sensors to estimate the pose of the vehicle in the process of moving without any prior information of the environment and their own pose is uncertain.At the same time,it is a process of sensing the surrounding environment in an incremental form and building a specific form of map.Therefore,this paper makes an in-depth study on the key technologies of smart car laser slam from four aspects: the principle and model of slam,the plane feature extraction of single frame lidar data,the construction of global map and the experimental design of control group.The main contents are as follows:(1)For the intelligent vehicle SLAM system,a complete SLAM framework and mathematical expressions based on probability models are given,a mobile model of the intelligent vehicle and an observation model of the lidar sensor are established,and the map model commonly used in the construction of the map is introduced.(2)Regarding the problem of extracting geometric feature maps from point cloud data collected from a real three-dimensional environment,this paper introduces the plane extraction algorithm of 3D MSSE point cloud data.This method can extract planar features from the original 3D point cloud data.By removing the non-static plane and redundant plane from the extracted plane features,it provides a basis for the subsequent plane segmentation of single-frame lidar data.Regarding the extraction of plane features from a plane map,a method of establishing a search box and counting the number of point clouds in a single search box is proposed to realize the plane feature segmentation of the plane.(3)In order to solve the problem of intelligent vehicle positioning in the process of moving,this paper uses a method to transform the information of geodetic coordinates into global coordinates.In order to realize the transformation of the plane features in the lidar coordinate system to the global coordinate system,the rotation and translation of the plane features are carried out in this paper.In laser SLAM algorithm,data association is the core and key of building global map.In order to solve the problem of data association in laser slam,three conditions are proposed to judge the features of the same plane between two adjacent frames:(1)the angle between features;(2)the coincidence degree of feature circle;(3)the relative distance of features.Data association is carried out for features that meet these three conditions at the same time.In the stage of data association,"quasi characteristic line" is obtained by solving linear regression equation for feature points,and the feature points are projected on the quasi characteristic line.The longest line segment of the projection point line on this line is considered as the plane feature corresponding to the structured plane.(4)In order to verify the accuracy of the proposed algorithm,this paper adopts an offline method to build global map.In the stage of building off-line global map,through the statistics of the occurrence times of a certain feature in lidar data set,it can be distinguished that this feature corresponds to the structured building plane or the unstructured building plane;finally,through the integration of all the features of a structured corresponding,the off-line global map is obtained.Matlab simulation results show that the proposed algorithm can realize the construction of planar feature map in structured environment.