CN-121999128-A - Pine tree three-dimensional model rapid reconstruction method based on unmanned aerial vehicle photography

CN121999128ACN 121999128 ACN121999128 ACN 121999128ACN-121999128-A

Abstract

The invention relates to the technical field of computer graphics and three-dimensional modeling, and discloses a pine tree three-dimensional model rapid reconstruction method based on unmanned aerial vehicle photography, which comprises the steps of obtaining a multi-viewpoint image sequence, extracting a single pine tree topology skeleton line, and calling a parameterized fractal generation program to construct a virtual three-dimensional geometrical body; the method comprises the steps of carrying out Euclidean distance transformation on an image entity contour to generate a distance field diagram, calculating the projection consistency deviation of a virtual geometric body in the distance field diagram, carrying out iterative correction on geometric growth control parameters according to deviation gradients until convergence, and finally carrying out instantiation loading of a preset leaf cluster model at the tail end of a framework to generate a three-dimensional grid.

Inventors

LIU YUGUO
WANG FENG
LI JIAHAO
LI MIN
CAI YIFEI
LI XIAOYA
YANG KAIJIE

Assignees

中国林业科学研究院生态保护与修复研究所

Dates

Publication Date: 20260508
Application Date: 20251226

Claims (10)

1. A pine tree three-dimensional model rapid reconstruction method based on unmanned aerial vehicle photography is characterized by comprising the following steps: Step 101, acquiring a multi-viewpoint two-dimensional image sequence aiming at a target pine forest region, calculating sparse space point clouds based on the multi-viewpoint two-dimensional image sequence, and clustering the sparse space point clouds aiming at a single pine tree target to extract topological skeleton lines representing the main trunk and main branch space trend of the single pine tree target; 102, calling a preset parameterized fractal generation program, constructing a virtual three-dimensional geometric body according to a group of geometric growth control parameters by using a topological skeleton line as a growth reference, wherein the geometric growth control parameters at least limit the iteration depth and the branch deflection angle of the branches; step 103, performing Euclidean distance transformation on the pine-tree entity outline in the multi-view two-dimensional image sequence, and generating a distance field diagram representing the distance gradient from the pixel point to the outline boundary, wherein the pixel value in the distance field diagram defines the Euclidean distance from the pixel point to the nearest outline boundary; 104, projecting the virtual three-dimensional geometric body to a distance field diagram, determining a distance sampling value of each projection pixel point in the distance field diagram, and calculating a projection consistency deviation according to a weighted sum of the distance sampling values; step 105, judging whether the projection consistency deviation meets a preset convergence condition, when the projection consistency deviation does not meet the preset convergence condition, iteratively correcting the geometric growth control parameter according to the gradient direction of the projection consistency deviation, and updating the virtual three-dimensional geometric body based on the corrected geometric growth control parameter until the projection consistency deviation meets the preset convergence condition; and 106, locking the geometric growth control parameters and the topological skeleton line when the preset convergence condition is met, loading a preset leaf cluster geometric model at the end node position of the topological skeleton line through instantiation citation, and generating a target three-dimensional grid model.
2. The fast reconstruction method of the pine three-dimensional model based on unmanned aerial vehicle photography is characterized by constructing a virtual three-dimensional geometric body according to a group of geometric growth control parameters, comprising the steps of calculating a normalized height value of each growth node relative to an initial root node on a topological skeleton line, constructing a parameter mapping function, wherein the parameter mapping function defines a monotonic function relation between a branch deflection angle and the normalized height value, and calculating and assigning the branch deflection angle of a current node through the parameter mapping function according to the normalized height value of the current growth node when constructing the virtual three-dimensional geometric body so as to form a non-uniform geometric topological structure which varies with height.
3. The fast reconstruction method of the pine three-dimensional model based on unmanned aerial vehicle photography is characterized by comprising the steps of determining a topology growth level corresponding to each projection pixel point in a virtual three-dimensional geometry according to a weighted sum of distance sampling values, enabling the topology growth level to represent iteration depth of a geometric unit of the projection pixel point in a fractal generation program, constructing a level weight function, enabling the level weight function to limit negative correlation between the weighting coefficient and the topology growth level, distributing the weighting coefficient according to the topology growth level corresponding to each projection pixel point, enabling the weighted coefficient distributed to the area with the lower topology growth level to be larger, and carrying out weighted sum calculation on the distance sampling values based on the weighting coefficient to obtain the projection consistency deviation.
4. The fast reconstruction method of the pine three-dimensional model based on unmanned aerial vehicle photography is characterized by comprising the steps of detecting whether a discrete canopy region of missing trunk point cloud data exists in sparse space point clouds or not, constructing a surface grid model for the top point clouds of the discrete canopy region if the discrete canopy region exists, calculating Gaussian curvature distribution of the surface grid model, identifying curvature extremum points and curvature ridge line characteristics, taking the curvature extremum points as initial endpoints, leading out virtual trunk paths along the gravity direction, deducing virtual branch paths according to a space mapping relation of the curvature ridge line characteristics, combining the virtual trunk paths and the virtual branch paths as topology skeleton lines, and complementing the topology structure of a vision blind area.
5. The fast reconstruction method of the pine three-dimensional model based on unmanned aerial vehicle photography is characterized by comprising a process of iteratively correcting geometric growth control parameters according to gradient directions of projection consistency deviation, and further comprising a non-rigid correction step based on residual space distribution, wherein a binary difference map representing a difference region between a projection region of a virtual three-dimensional geometric body and a pine entity contour is generated, a geometric centroid of a non-zero pixel region in the binary difference map is calculated, an offset vector of the geometric centroid relative to the projection center of the virtual three-dimensional geometric body is determined, a Bessel curve-based non-rigid deformation is applied to a topological skeleton line by using the offset vector as a deformation control parameter, and the steps of constructing and projecting the virtual three-dimensional geometric body are re-executed based on the deformed topological skeleton line.
6. The fast reconstruction method of the pine three-dimensional model based on unmanned aerial vehicle photography is characterized by further comprising the steps of constructing a spatial index structure based on all extracted topological skeleton lines before constructing a virtual three-dimensional geometric body according to a set of geometric growth control parameters, calculating a neighborhood density value of the topological skeleton lines of each single pine tree in a preset spatial radius, constructing an iteration depth control function, limiting a negative correlation between a branching iteration depth and the neighborhood density value by the iteration depth control function, and taking the branching iteration depth obtained by calculation according to the neighborhood density value as a fractal generation termination threshold value for constructing a corresponding virtual three-dimensional geometric body to perform geometric simplification processing on invisible internal structures in a high-density region.
7. A fast reconstruction method of a pine tree three-dimensional model based on unmanned aerial vehicle photography according to claim 3, wherein the weighted sum calculation is performed on the distance sampling values based on the weighting coefficients, and the following weighted deviation calculation formula is followed: , wherein, In order to project the uniformity deviation out of the way, To project the total number of pixel points, Is the first The distance samples of the individual projected pixels in the distance field map, Is the first The topology growth levels corresponding to the projection pixels, Corresponding to a hierarchy determined by a hierarchy weighting function Is used for the weighting coefficients of (a).
8. The fast reconstruction method of the pine three-dimensional model based on unmanned aerial vehicle photography is characterized by obtaining a multi-view two-dimensional image sequence aiming at a target pine area, and comprises the steps of performing semantic segmentation on the multi-view two-dimensional image sequence, extracting a binary semantic mask containing a pine target, removing discrete connected domains with areas smaller than a preset noise threshold in the binary semantic mask, and taking the binary semantic mask with the noise removed as a pine entity contour for subsequent Euclidean distance transformation.
9. The fast reconstruction method of the pine tree three-dimensional model based on unmanned aerial vehicle photography is characterized in that a preset leaf cluster geometric model is loaded at the position of an end node of a topological skeleton line through instantiation citation, and the fast reconstruction method comprises the steps of building a prefabricated body library containing the leaf cluster geometric models with various morphological specifications, determining model index values of all end nodes according to end distribution density parameters in geometric growth control parameters, calling the corresponding leaf cluster geometric model from the prefabricated body library according to the model index values, and binding a coordinate transformation matrix of the leaf cluster geometric model to an end node coordinate system of the topological skeleton line.
10. The fast reconstruction method of the pine three-dimensional model based on unmanned aerial vehicle photography according to claim 1 is characterized in that judging whether the projection consistency deviation meets a preset convergence condition or not comprises calculating the change rate of the projection consistency deviation of a current iteration round and the projection consistency deviation of a previous iteration round, judging that the preset convergence condition is met when the projection consistency deviation is smaller than a first absolute threshold or the change rate is smaller than a second relative threshold and continuously maintains for a preset number of times, and terminating iteration and outputting a current optimal parameter set if the preset maximum iteration number does not meet the preset convergence condition.

Description

Pine tree three-dimensional model rapid reconstruction method based on unmanned aerial vehicle photography Technical Field The invention relates to a pine tree three-dimensional model rapid reconstruction method based on unmanned aerial vehicle photography, and belongs to the technical field of computer graphics and three-dimensional modeling. Background In the current large-scale outdoor scene three-dimensional digital operation, a visible light camera is carried by an unmanned plane to obtain a multi-view image sequence, a scene three-dimensional structure is restored based on a multi-view solid geometry algorithm, a general paradigm of geospatial information is obtained, the technology generally follows a link from motion restoration structure and dense point cloud matching to surface grid reconstruction processing, and mathematical model convergence depends on a priori assumption that manifold continuity and luminosity consistency are satisfied in a local neighborhood of the surface of a shot object. The method is characterized in that the method is based on digital surface model DSM and deep learning combination logic, and utilizes high-level information to assist species identification, technical kernels are limited to two-dimensional or 2.5-dimensional analysis of forest crown surface visual characteristics, the method focuses on species discrimination rather than structure reconstruction, and can not penetrate into the topological layer of the interior of a single forest, reconstructed objects are converted into pine and the like, and when non-rigid objects with high-frequency discrete needle features and high closure degree shielding are shielded, the modeling paths face principle failure risks based on DSM or pixel level dense matching; when a reconstructed object is converted into a discrete voxel set with high-frequency spatial frequency characteristics such as a pine forest from a building or a topography, a traditional dense matching modeling path based on a pixel level faces a principle failure risk, a pine canopy is formed by a large number of discontinuous needle leaf clusters, sub-pixel level geometric characteristics are easy to be lost in a conventional downsampling or characteristic matching link, so that point cloud sparsity or depth map calculation fails, unmanned aerial vehicle sequence image acquisition has time difference, pine branches and leaves generate non-rigid transient displacement in an outdoor uncontrolled wind field environment, and physical deformation and rigid geometric matching constraint have endophytic contradiction, so that an optimization function based on luminosity consistency cannot converge or sink into a local extremum, image resolution is simply improved or shooting overlapping degree is increased, calculation force load is exponentially increased, and the fundamental problems of view field shielding and dynamic deformation cannot be solved. Therefore, how to break through the reconstruction bottleneck of discrete non-rigid objects by only relying on the conventional two-dimensional image sequence and realize the construction of a high-fidelity three-dimensional model conforming to the topological rule of botanics under the conditions of low data density and dynamic interference becomes the technical problem to be solved by the invention. Disclosure of Invention In order to solve the problems in the background technology, the technical scheme of the invention is as follows, a pine tree three-dimensional model rapid reconstruction method based on unmanned aerial vehicle photography comprises the following steps: Step 101, acquiring a multi-viewpoint two-dimensional image sequence aiming at a target pine forest region, calculating sparse space point clouds based on the multi-viewpoint two-dimensional image sequence, and clustering the sparse space point clouds aiming at a single pine tree target to extract topological skeleton lines representing the main trunk and main branch space trend of the single pine tree target; 102, calling a preset parameterized fractal generation program, constructing a virtual three-dimensional geometric body according to a group of geometric growth control parameters by using a topological skeleton line as a growth reference, wherein the geometric growth control parameters at least limit the iteration depth and the branch deflection angle of the branches; step 103, performing Euclidean distance transformation on the pine-tree entity outline in the multi-view two-dimensional image sequence, and generating a distance field diagram representing the distance gradient from the pixel point to the outline boundary, wherein the pixel value in the distance field diagram defines the Euclidean distance from the pixel point to the nearest outline boundary; 104, projecting the virtual three-dimensional geometric body to a distance field diagram, determining a distance sampling value of each projection pixel point in the distance field diagram,