Research And Design Of Point Cloud Registration Algorithm Based On Local Curvature Feature

With the rapid development of three-dimensional point cloud technology,point cloud technology is widely used in the field of computer vision such as automobile automatic driving and UAV inspection.Due to the limitation of the perspective range of the acquisition equipment,it is necessary to collect the point cloud data of the measured object from multiple angles.The point cloud registration technology is used to seamlessly splice the collected multi-view point cloud to obtain the complete point cloud data of the object.How to complete point cloud registration fast and high quality is still an important research value.Aiming at the problems of slow speed and easy to fall into loca l optimal solution of 3D point cloud registration,this paper takes feature point detection and point cloud coarse registration as the core to study point cloud registration exhibition.The main contents are as follows :(1)Studying the existing point cloud pre-processing methods,analyzing the combination of point cloud filter and point cloud density feature in the point cloud pre-processing process,aiming at the point cloud data with uneven density distribution and noise,uniform sampling of point cloud and radius filtering algorithm are used for filtering.In this way,the noise can be eliminated without any holes and the point cloud feature can be preserved.(2)This paper focuses on the common methods of point cloud feature extraction,and proposes a feature extraction method based on curvature and its neighboring points quasi-plane,the curvature characteristic parameters are constructed by extracting the curvature of the detection point,calculating the distance from the detection point to the nearest point fitting plane,and the standard deviation of the distance between the probe point and the fitted plane of the near neighbor.The speed of extracting point cloud feature points by this method is faster than ISS,Harris,SIFT and other methods.When using point cloud feature combined with 4PCS algorithm for coarse registration,the proposed algorithm is superior to feature extraction algorithms such as ISS and Harris in speed and accuracy,which effectively improves the accuracy and speed of 4PCS coarse registration.When the algorithm in this paper combines SAC-IA algorithm for coarse registration,it can greatly improve the coarse registration speed of point cloud,but the accuracy of some multi-view point clouds is not good when FPFH feature descriptor is used.(3)According to the related algorithms studied in this paper,designed a point cloud processing and calibration system which using Clion development platform and QT realizes the graphical interface of the system.The input,output and work flow of each function module are described in detail.