Research On Skeleton Extraction From One-scan Tree Point Clouds

Tree point clouds are widely used in virtual reality,the protection of the ancient and famous trees,tree growth research and the like.In practical applications,such as the study of tree nutrient transfer,tree skeletons express plant characteristics more directly than tree point clouds.Therefore,extracting skeletons from point clouds has important application significance.However,maybe only one-scan point clouds can be scanned due to the limitation of the shooting environment when obtaining data.For trees with only one-scan point clouds,a large number of points are missing.In addition,due to the error of acquisition equipment and algorithm,the point cloud has the problem of non-uniform distribution and noise.It is challenging to accurately extract tree skeletons from one-scan tree point clouds with uneven distribution,noise and missing structural information.The binocular stereo camera has the advantages of cheap and easy to obtain.In this thesis,the binocular stereo camera is used to take pictures of trees.The stereo matching algorithm is used to extract the depth map.On this basis,the depth map is repaired,the one-scan point cloud is acquired,and the tree skeletons of one-scan tree point cloud are extracted.Specific work includes the following aspects:(1)The tree depth map has been obtained by stereo matching.In this thesis,the images of trees have been taken by binocular stereo camera;camera parameters are obtained by camera calibration;camera distortion correction and image polar correction are preprocessed;the parallax of each pixel is obtained from the left and right views by SGBM stereo matching algorithm,and the depth map is obtained according to the relationship between parallax and depth.(2)The depth map has been converted to a one-scan tree point cloud.The acquired depth map will be missing due to the influence of hardware equipment and shooting environment.In order to obtain a complete one-scan tree point cloud,a depth map repair is needed.The repaired depth map is based on the camera imaging principle to obtain a one-scan tree point cloud.(3)One-scan tree point cloud skeletons have been extracted.In order to solve the problem of uneven distribution,noise and serious lack of structural information of onescan trees obtained by binocular stereo camera,the point cloud is homogenized and denoised by using adaptive local projection algorithm in this thesis,and the point cloud is segmented by using normalized cutting algorithm.In the process of obtaining skeletons points by iterative contraction based on "-median,point cloud segments are merged.Finally,the minimum spanning tree algorithm is used to connect skeletons points.(4)The validity of the proposed algorithm in this thesis is verified by experiments.In this thesis,one-scan tree point cloud skeletons are obtained based on binocular stereo camera,and the algorithm of depth map repair is validated.Finally,the results of this algorithm are compared with those of "-median and ROSA algorithm.The experimental results show that the proposed algorithm in this thesis has fewer topology errors in the extraction of the skeletons of one-scan tree point cloud,has stronger robustness in the case of uneven distribution of point clouds and noise.