Research On Multi-view Lidar Point Cloud Registration Technology Based On Tuple Constraint

The rise and rapid development of 3d laser scanning technology makes people have more cognition and exploration of 3d spatial information.In order to obtain a complete 3d Lidar point cloud data model of a building,efficient and accurate registration of point cloud data from different perspectives is required.Due to the characteristics of high resolution,discreteness and large amount of data of Lidar point cloud data,the density,scale and data integrity of point cloud data acquired from different angles are often different,which brings great challenge and pressure to Lidar point cloud registration.Therefore,this paper takes the Lidar point cloud data of buildings as the research background,and studies the problem of unstable local feature matching and point cloud data registration with different scales in the registration process of multi-view Lidar point cloud data.The main research work is as follows:1.In view of the problem that a large number of incorrect matching relations are generated in the local feature matching of 3d Building Lidar point cloud registration process,which reduces the stability of feature matching and leads to insufficient registration accuracy,a multi-view Lidar point cloud registration algorithm based on quaternion constraints is proposed.First,you need to calculate the normal vector of the 3d point cloud,using each point and the relationship between the k neighbor points calculation of the normal vector of each point,and then based on the feature point and point in the field of the relationship between the normal vector calculation point cloud FPFH feature descriptor,it can be said that building the geometric feature of Lidar point cloud;secondly,by using the three-dimensional characteristics describe the characteristics of the local matching,looking for two groups of different Angle of view of the initial matching points between the Lidar point cloud;On the basis of using the nearest neighbor constraint conditions to eliminate duplicate and wrong matching point pairs,quaternion constraints are added to better eliminate the wrong matching point pairs.By combining the two constraints,the stability of local feature matching is guaranteed,and finally point cloud registration is realized.Aiming at the point cloud data of buildings with concave and convex walls,tree scenes and large scale,by comparing the experimental results of the algorithm in this chapter,ICP algorithm,GICP algorithm and Fpr algorithm,it can be seen that the algorithm of multi-view Lidar point cloud registration algorithm based on quaternion constraints can improve the registration accuracy of Lidar point cloud of buildings from multi-view.2.The Lidar point cloud registration algorithm based on normal vector constraint is proposed to solve the problem of insufficient registration accuracy of multi-view point cloud data with different scales of multi-view.The principle of the method is to first calculate each point of the point cloud of 3d feature descriptor,then based on the characteristics of the nearest neighbor matching relationship,build the relationship between the Angle of point cloud set,then using ternary distance constraint optimize relationship's set to weed out the wrong relationship,try to join the candidate matching points on the normal vector views of the constraint,can use the different scales of two point cloud data more overlapping information,improve the accuracy of matching point pairs,at the same time can ensure the accuracy of multi-view Lidar point cloud registration.For four sets of point cloud data of buildings including windows,roofs and complex large scenes,the experimental results of the algorithm in this chapter and the other three algorithms are compared to verify that this algorithm can improve the registration accuracy of Lidar point clouds with different scales.To sum up,for the stability of local features in Building Lidar point cloud registration,a multi-view Lidar point cloud registration algorithm based on quaternion constraints was proposed.Then,for the problem of insufficient registration accuracy of point cloud with different scales,a multi-view Lidar point cloud registration algorithm based on normal vector constraints was proposed.Through multiple experiments and comparison of experimental results,it can be seen that the proposed method can effectively improve the registration accuracy of point cloud.