Research On High-precision Point Cloud Map Construction And Localization Algorithm Based On Information Fusion

With the development of computer technology and the development of the automobile industry,autonomous driving has gradually become a research hotspot and the construction of high-precision maps based on point clouds and the localization based on high-precision point cloud maps that have been constructed are important research areas in autonomous driving technologies.Existing high-precision point cloud map construction based on point clouds and positioning algorithms based on point cloud maps usually do not make full use of valid information such as location and point cloud data structure.In the face of obstacles obstructing,point cloud noise,and a large amount of point cloud data,the accuracy of point cloud map construction and positioning accuracy still can be improved.Based on the idea of information fusion,this paper studies high-precision point cloud map construction and point cloud localization based on high-precision point cloud maps respectively.In terms of highprecision point cloud map construction,this paper proposes a point cloud road trunk extraction algorithm based on location information fusion and an improved ICP(Iterative Closest Point)registration algorithm based on the road trunk point cloud extraction algorithm and the improvement of the accuracy and timeliness of high-precision point cloud map construction is realized.In terms of point cloud localization based on high-precision point cloud maps,a localization algorithm framework based on a priori information fusion is proposed.And according to the framework,an improved particle filter positioning algorithm based on a priori information fusion is proposed to improve the positioning accuracy.The specific research work is as follows:(1)Aiming at the problem of large amount of point cloud data and much noise when constructing a point cloud map,this paper takes the overall structural characteristics of the road point cloud into account and proposes a road trunk extraction algorithm based on the position information fusion.This algorithm uses vehicle position information as the reference point of the road trunk point cloud,extracts points less than a certain road width distance from the reference point,and then extracts the trunk part of the road point cloud,reducing registration points with large errors.Next,this paper combines the road backbone extraction algorithm based on the location information fusion with the ICP algorithm to improving the registration accuracy and efficiency of ICP algorithm.In addition,this paper analyzes the construction process of the point cloud map based on the inertial navigation system,and applies the improved ICP registration algorithm to construct a high-precision point cloud map.(2)Aiming at the problem of poor localization accuracy cause by noise interference and inconsistency between the collected data and the map in the point cloud based on particle filter algorithm,a localization algorithm framework based on prior information fusion is proposed.The framework takes into account the prior information that the observation data is most similar to the map data of the real location of the vehicle on the point cloud map,and combines the relationship with the intermediate parameters of the localization algorithm to optimize the intermediate parameters,thereby improving the positioning algorithm and making up for obstacles problems such as object occlusion and dynamic noise.At the same time,according to the proposed framework,an improved particle filter localization algorithm based on particle drift is proposed.Through the particle set,the algorithm merges the a priori information that the particle set gathers closer to the particles closer to the real position.The least squares theory and simulated annealing are applied to bring the particles closer to the real value and ensure the shift of the particles.Finally,improving the update step of the particle filter algorithm to make the localization accuracy and robustness.