Research On 3D High Precision Map Generation Algorithm Based On Multi-lidar

In recent years,with the rapid development of science and technology,all walks of life set off the upsurge of artificial intelligence research.As one of the most important research fields of artificial intelligence,unmanned driving has attracted numerous enterprises to participate in it.Unmanned driving technology is divided into sensing,positioning,decision making,planning and control.Among them,positioning technology is an important prerequisite to realize autonomous navigation and decisionmaking planning of unmanned vehicles.With the acceleration of urbanization,the traditional positioning scheme is difficult to achieve high precision positioning effect due to environmental shielding and complex road conditions.The emergence of high-precision map provides a new idea for positioning technology.Unmanned vehicles can realize positioning through high-precision map,which not only ensures real-time performance but also achiecves high-precision positioning effect.Nowadays,most high-precision maps are realized based on point cloud registration algorithm,but the traditional point cloud registration algorithm still has many problems.For example,ICP(Iterative Closest Point)algorithm uses too much time and too many iterations to work out the local optimization problem.NDT(Normal TF Transform)algorithm has the problem of attitude solution due to unreliable sparse point clouds.Point clouds in high-precision maps are mostly obtained by 3D lidar.However,a single three-dimensional lidar has inevitable defects in performance.For example,the real-time data collected by a single lidar usually has fewer lines and sparse point clouds.However,sparse point clouds will lead to difficulties in feature point extraction during map generation.When key information needs to be extracted,the data value obtained from sparse point cloud will be greatly reduced,and meanwhile,the detection and identification ability of unmanned vehicles to obstacles will be greatly affected.Aiming at the above problems,the high-precision map generation technology based on multi-lidar registration is studied deeply in this paper,and the improved algorithm is applied to the actual scene.In view of the above problems,this paper mainly completes the following work contents:(1)Aiming at the sparse problem of single lidar point cloud,a new method of multi-lidar registration is proposed in this paper.The firefly optimization algorithm is introduced into the point cloud registration problem,which solves the problems of inaccurate attitude solution among multiple lidar,large registration error caused by sparse point cloud,etc.,and achieves the effect of accurate calculation and smaller registration error.(2)Aiming at the problems of too many point sets in plane fitting,too many iterations leading to too much workload and too low algorithm efficiency.This paper studies the classical RANSAC(Random Sample Consensus)algorithm,and designs a threshold setting method based on model adaptation.The corresponding parameterized model is designed according to the size of the threshold,so as to reduce the number of iterations and achieve the purpose of improving efficiency.(3)Aiming at the difficult problem of feature point extraction during map generation,this paper introduces the multi-lidar registration algorithm into the algorithm framework of LOAM-SLAM(Lidar Odometry and Mapping in Real-time)to solve the difficult problem of feature point extraction caused by sparse point cloud.Better complete the map generation.It is proved that the application of multi-lidar registration algorithm in LOAM-SLAM algorithm has certain research significance and practical value.In this paper,the firefly algorithm optimization is introduced into the multi-lidar registration problem,and the attitude solution is optimized.A threshold setting method based on model fitting was designed and applied to plane fitting to improve the fitting efficiency.By combining the multi-lidar registration algorithm with the framework of LOAM-SLAM algorithm,the difficult problem of feature point extraction in map generation is solved,and the extraction efficiency and accuracy are improved.