Improvement Of 3D Laser Point Cloud Feature Extraction And Registration Algorithm

3D laser point cloud data is widely used in various fields of modern engineering due to its wide access,fast scanning speed and high precision.However,point cloud data also has shortcomings such as poor anti-interference,poor robustness,and small data range for single acquisition,which increases the difficulty for the application of 3D point cloud data.Therefore,for the characteristics of point cloud data,this paper adopts feature attribute quadrant segmentation method and deep learning automatic editor cover set algorithm to try to solve the problem of point cloud element extraction and registration accuracy.The main work of the thesis is as follows:(1)Aiming at the complexity of 3D point cloud data structure and the uneven distribution of point cloud density,and improving the retrieval rate of matching point pairs,a new feature attribute quadrant segmentation method is adopted in this paper.This method firstly extracts the feature attributes of point cloud data based on the normal vector to ensure the compression of the whole point cloud data and the local feature invariance.Then,the extracted features are segmented into the spatial quadrant,and the point cloud data is processed into blocks to improve the correspondence.The correct match,rate for a point pair.(2)For the problem of 3D point cloud data registration accuracy,this paper improves the deep learning automatic editor coverage set(LORAX)algorithm,firstly transforms the feature point cloud of quadrant segmentation into super based on random sphere cove.r set(RSCS).Point,using the hyper-points in each quadrant to describe the three-dimensional local feature information of the object,and then using the automatic encoder feature vector(SAF)to describe the geometric information of the super-point,and pairing the super-point groups with the same attributes,and finally based on the super The registration parameters obtained by the point complete the coarse registration.The improved algorithm enhances the similarity of the super-point group when describing the object attribute information,and improves the accuracy of the point cloud coarse registration.(3)For the limitation of nearest point iterative method(ICP),this paper adopts the discriminant optimization(DO)algorithm,which will search the point cloud data obtained in the coarse registration,and continuously search the space in the search space.The sequence is iteratively updated to obtain the static point of the point cloud registration optimal transformation value,and the final registration of the point cloud data is completed.The ICP algorithm is often interfered with by extra point clouds,and the DO algorithm can accurately track and estimate the pose of an object even under overlapping occlusion or structural outliers.