| Point cloud segmentation is a crucial link of 3D point cloud data processing.Point cloud segmentation is an important part of scenario analysis which can locate,object recognize,classify and feature extract,so segmenting object point cloud with high accurate and robust is a great significance to 3D point cloud data processing.Point cloud data can provide high-quality 3D information,but the defects of point cloud are density can be influenced by distance and 3D points distribution is irregular.Above these defects can make point cloud segmentation results have remarkable under segmentation and over segmentation.Under the above-mentioned background,this paper studies the point cloud segmentation algorithm.Region growing point cloud segmentation algorithm and clustering point cloud segmentation algorithm have low complexity,high usability and can be easy to implement.Image has high resolution and also can provide abundant color and vein information so that image and point cloud fusion can be complementary,so this paper focus on improving the two algorithms by image and point cloud fusion.The specific research contents of this paper are as follows:(1)Construction and calibration of image and point cloud data acquisition system.The data collected by RGB camera and 3D Li DAR are based on their respective sensor coordinate systems.In order to fuse RGB image and 3D point cloud,it is necessary to calibrate the two data and obtain the extrinsic parameters of the two sensors.Due to the resolution ratio of Li DAR needs to match the view of RGB camera when acquiring the two data,this paper designs a 3D point cloud imaging system for the data acquisition system,and then uses checkboard plane constraint to calculate the extrinsic matrix of the two sensors,and establish alignment relation with the extrinsic matrix of the two sensors and the intrinsic matrix of the camera.(2)Fusion of image edge in region growing point cloud segmentation algorithm.The traditional region growing point cloud segmentation algorithm is always hard to select the proper seed points and design the corresponding growth strategy.In order to solve these problems,this paper uses image to obtain the edge points of the object point cloud,designs the growth strategy according to the properties of the edge,and set the edge points of the object point cloud as the seed points to region growing segment.(3)Fusion of image centroid in clustering point cloud segmentation algorithm.The traditional clustering point cloud segmentation algorithm is always hard to estimate the total number of classifications,the proper clustering centers and design the corresponding clustering strategy.In order to solve these problems,this paper uses image to obtain the center point of the object point cloud,calculate the correlation degree between the points of original point cloud and the classification of object point cloud according to the center point of the object point cloud and the point of object point cloud classification,and classify the 3D points of point cloud by correlation degree threshold value.In order to prove the effectiveness of the proposed algorithms in this paper,this paper proves that the calibration method can fuse image and point cloud with high accuracy,and the offset error of calibration will not influence the subsequent experiments.Finally,the two proposed algorithms and the other comparison algorithms are tested on the same data set.Counting under-segmentation rate,over-segmentation rate and intersection-over-union ratio of the experimental results as evaluation criterion.The experimental results show that the two proposed algorithms can segment the object point cloud with high precision from the dense point cloud with a large number of 3D points,and can effectively avoid under segmentation and over segmentation. |
