| Tree is an indispensable part of nature.Using virtual reality technology to simulate the growth process and movement of tree in the natural environment,understanding the stress re-sponse of tree to the external environment during their growth,accelerating all-round learning and understanding of the progress of tree knowledge,is of great significance to study the rela-tionship between people and the environment.The 3D laser point cloud is the data foundation for simulating plant growth and tree modeling.Point cloud data contains spatial information such as relative positional relationships between points,local topologies,and overall geome-try.With the deepening of the application of virtual reality technology and computer graphics in agriculture and forestry,the 3D modeling of tree point clouds and the simulation of plant growth have become the research hotspots of various relevant institutions at home and abroad.Due to the large number of branches and criss-crossing of trees and complex topologies.Using traditional point cloud reconstruction algorithms to perform triangulation directly on the surface of a tree point cloud will produce a large number of deformed structures that do not conform to the geometry of the trees in the natural world.The current tree point cloud re-construction algorithm relies heavily on a large number of prior knowledge enlightenment and complex interactive operations.The operation mode is cumbersome,the algorithm structure is complex,the time cost is huge,and the ideal model effect is not obtained.In order to fully utilize the self-information of the tree point cloud and effectively preserve the topological relationship between the point clouds,this paper proposes a 3D reconstruction method of tree point cloud based on skeleton extraction.The mesh smoothing algorithm is directly applied to the surface of the tree point cloud to extract the basic topology of the tree point cloud,and the tree model are geometrically constructed based on the skeleton.The main contents of this research and its innovations are as follows:?1?A tree point cloud acquisition and preprocessing scheme was constructed.The binoc-ular laser 3D scanner is used to obtain the point cloud data of the tree;In order to remove the noise,outliers and singular values contained in the point cloud of the tree,a bilateral filter and a density clustering algorithm are respectively adopted,and the average error is 0.28 mm;The domain geometric feature similarity and consistency drift registration method are used to regis-ter the tree point cloud with the registration error within 0.03mm.The point cloud is simplified based on the ray principle on the complete tree point cloud,and the reduction rate reaches 72%.Finally,the point cloud is segmented by using the octree structure;the tree point cloud prepro-cessing provides a good foundation for the subsequent skeleton extraction.?2?Aiming at the problem that the geometrical reconstruction rate is low due to sparse point cloud at the canopy,a tree point cloud skeleton extraction algorithm based on curvature normal stream operator is proposed.First,the point cloud is partially meshed,and a ring field is established at each vertex;Then the contraction force provided by the curvature normal line mesh shrinkage algorithm is used to shrink the tree point cloud to almost zero volume shape.In the local range,the curvature normal flow operator moves the vertices along the local normal to the interior of the point cloud at the speed of the average curvature,maintaining the relative po-sitional relationship between adjacent points;On the whole,it shrinks the point cloud internally by iterating,retaining the geometry and topology of the original tree point cloud.The shrink-age ratio of the model accounts for 1%to 5%of the initial volume,the degree of coincidence is above 97%,and the shrinkage efficiency is 800 points/s.?3?An improved quadratic error metric grid simplification algorithm is proposed for the smooth slenderness of tree branches and the gentle bending of branches.The algorithm uses the quadratic error metric of the mesh edge as the shape cost,so that the extracted one-dimensional curve skeleton and the skeletal mesh maintain the same topology and geometry;To prevent the generation of excessively long skeleton edges,a side length cost is set to preserve the exact mapping between the skeleton point and the original point cloud.The smooth angle of the skeletal branch junction is between[10?,45?],and the mesh simplification rate is 5000 faces/s.?4?A skeleton correction and geometric reconstruction scheme is constructed.In the aspect of skeleton correction,each skeleton point is moved to the center of its corresponding original point cloud based on point cloud-skeleton mapping,and the deviation error is less than 0.03cm;The extracted skeleton is connected and processed by growth angle constraint and two-way detection method.The connection angle is less than 45?.In terms of geometric reconstruction,the skeleton point structure and the tree hierarchy are established.Combining the plant charac-teristics of the trees and the allometric growth theory,each skeleton point is assigned a radius proportional to its weight,and the skeleton is used as the central axis to establish a generalized cylindrical for trees.The point cloud is geometrically reconstructed with an average radius error of±0.34cm.?5?Different contrast experiments were designed to study and analyze the tree point cloud skeleton extraction and geometric reconstruction.In the aspect of extracting the accuracy of the skeleton,the contrast experiment between the curvature normal operator and the cotangent operator is designed.It is verified that the operator can better preserve the geometry of the orig-inal point cloud at the canopy without the skeleton protrusion and ring structure and other phe-nomena.In the aspect of skeleton extraction speed and reconstruction degree,the comparison experiments between the proposed algorithm and other classical algorithms?ROSA algorithm,L1-axis algorithm,voxel algorithm,space colonization algorithm?are designed.Compared with other algorithms,the algorithm does not require additional pre-processing and normal vector calculations,and it shows good robustness for proper noise and missing models.For the point cloud of trees with the amount between 10k and 150k,the skeleton extraction speed of this al-gorithm is 600 points/s,which is more than 3 times higher than other algorithms,and the branch reconstruction rate is increased by 25%. |
PDF Full Text Request