Three-dimensional point cloud reconstruction volume calculation method

Abstract

The invention discloses a three-dimensional point cloud reconstruction volume calculation method, and relates to the technical field of data processing, and the method comprises the steps: S1, obtaining point cloud data in a stock bin; s2, generating an initialized grid M according to the point cloud data; s3, generating an Alpha shape of the point cloud according to the point cloud data; s4, building the grid M, and deleting the point cloud data outside the Alpha shape; s5, performing Laplace smoothing processing on the point cloud data processed in the step S4; s6, performing hole filling processing based on a minimum curved surface on the point cloud data processed in the step S5 to obtain a three-dimensional reconstruction model; and S7, calculating the volume of the three-dimensional reconstruction model according to the calculus principle. The point cloud can be accurately calculated through a calculus method after the point cloud model is reconstructed.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a method for calculating volume of three-dimensional point cloud reconstruction. Background Art

[0002] In the field of modern animal husbandry, the management and monitoring of silos are crucial for real-time understanding of the amount of feed in the silo, scientific and reasonable allocation of feed, improving feeding efficiency and ensuring feed safety. Traditional silo monitoring methods usually rely on regular manual inspections or simple sensor detection. These methods are time-consuming and labor-intensive, and are also susceptible to human errors. With the development of 3D scanning technology, real-time scanning of the interior of the silo can be achieved, and the internal conditions of the silo can be accurately monitored to improve management efficiency.

[0003] During each feeding process, a lot of dust will be generated inside the silo, which seriously affects the correct value of the radar scan, thus generating some discrete points, called noise points. In the process of using the denoising method to remove these discrete points, some important points will be removed, thus affecting the integrity of the internal shape of the silo. In addition, problems such as dust, high temperature and humidity may cause the radar scan point cloud to be partially missing. Accurate reconstruction of the denoised 3D model is the basis for calculating the volume of the material.

[0004] However, in actual applications, due to the influence of environmental factors such as dust, humidity and light in the warehouse, the acquired point cloud data often have problems such as noise and partial loss, distortion or error of the scanned point cloud, which affects the accuracy of subsequent volume calculation based on the point cloud map and leads to misjudgment of feed storage and distribution. Summary of the invention

[0005] The present invention provides a method for calculating volume by reconstructing a three-dimensional point cloud, which obtains an incomplete point cloud image for surface reconstruction, reconstructs an accurate and complete three-dimensional point cloud model, and then can accurately calculate the volume through a volume calculation function.

[0006] According to one aspect of the present disclosure, a method for calculating volume by reconstructing a three-dimensional point cloud is provided, the method comprising: Step S1, obtaining point cloud data inside the silo; Step S2, generating an initialization grid M according to the point cloud data; Step S3, generating an Alpha shape of the point cloud according to the point cloud data; Step S4, trimming the grid M and deleting the point cloud data outside the Alpha shape; Step S5, performing Laplace smoothing on the point cloud data processed in step S4; Step S6, performing hole filling processing based on the minimum surface on the point cloud data processed in step S5 to obtain a three-dimensional reconstructed model; Step S7, calculating the volume of the three-dimensional reconstructed model according to the principle of calculus.

[0007] In one possible implementation, step S1, obtaining point cloud data inside the silo, includes: obtaining point cloud data through a laser radar, the laser radar is equipped with a gyroscope, the gyroscope performs Z-axis calibration on the point cloud image, and obtaining normal vector information n according to the local geometric characteristics of all points in the point cloud data, and the normal points to the direction inside and outside the object, wherein geometry specifically refers to the spatial distribution, shape characteristics and local structure information of points in the point cloud data.

[0008] In a possible implementation, step S2, generating an initialization grid M according to the point cloud data, includes: The normal vectors of all points in the point cloud data are regarded as a gradient field, and the vector field of the unknown object M is obtained. , there is an implicit function f whose gradient corresponds to the normal vector field: (1) in, is a point in space when the gradient With vector field When the modulus of the difference is minimum, the gradient of formula (1) is satisfied Approximation to a vector field ,Right now: (2) Perform divergence calculation on the above formula (2) Transformed into Poisson's equation: (3) According to the point cloud normal vector n, calculate the vector field: (4) Constructing Divergence : (5) The Poisson equation is discretized using the finite element method to form a linear system: (6) in The Laplace matrix is ​​constructed based on the neighborhood relationship of the point cloud. The conjugate gradient method is used to solve formula (6) to obtain the potential function , use the marching cube algorithm to extract the isosurface. The essence of the marching cube algorithm is to use three points calculated by difference in each unit cube to form a set of triangular facets to form its isosurface, that is, to initialize the mesh surface M.

[0009] In a possible implementation, step S3, generating an Alpha shape of the point cloud according to the point cloud data, includes: Alpha shape is a geometric structure based on Delaunay triangulation, which is used to describe the boundary of a point set. For a point set P and parameter α, the smaller the value of α, the finer the generated boundary; The Alpha algorithm is defined as follows: For each point and for the edge Calculate the radius of a circle between two points ,if , it is retained in the Alpha shape.

[0010] In a possible implementation, step S4, pruning the grid M to delete the point cloud data outside the Alpha shape, includes: Use the generated Alpha shape Alpha (P, α) to repair the Poisson reconstructed mesh M. For each triangle or vertex in the mesh M, check whether it is inside the Alpha shape; If the triangle is outside the Alpha shape, a fast Boolean operation is used to delete it. For the triangle that is partially outside the Alpha shape, the boundary of the Alpha shape is used for clipping. Clipping means: intersecting the triangle that is partially outside the Alpha shape with the boundary of the Alpha shape and cutting the triangle to keep the part that is inside the Alpha shape. The fast Boolean operation uses an AABB tree to quickly detect and process the intersection, union and difference operations of a complex triangular mesh, wherein the AABB tree is a binary tree based on the nodes of an axis-aligned bounding box.

[0011] In a possible implementation, step S5, performing Laplace smoothing on the point cloud data processed in step S4, and step S6, performing hole filling based on a minimum surface on the point cloud data processed in step S5 to obtain a three-dimensional reconstructed model, including: The Laplace smoothing is performed on the pruned Poisson reconstruction network, and the formula is as follows: (7) in is the updated position of the i-th vertex at the t+1th iteration, is the vertex position at iteration t, N i is the neighbor geometry of vertex i, that is, the vertices adjacent to vertex i, λ is the smoothing sparsity, which is used to control the strength of smoothing, is the position of neighbor vertex j; Minimum surface hole filling is to deal with holes in the mesh caused by data loss or reconstruction error in 3D reconstruction. The minimum surface satisfies the partial differential equation: (8) Where z is the height of the three-dimensional surface on the plane xy, and the curvature of the surface is 0, which means that the area of ​​any small region of the surface is the smallest.

[0012] In a possible implementation, step S7, calculating the volume of the three-dimensional reconstructed model according to the principle of calculus, includes: First, select the slicing direction: determine the direction in which the point cloud is divided, usually the Z axis (height) is selected for slicing After determining the slice direction, determine the slice thickness: set the slice thickness, usually with a small value. According to the principle of calculus, the smaller the value, the more accurate the calculated volume; Slicing operation: Reconstruct the model through the point cloud and divide the point cloud into multiple horizontal slices according to the set slice thickness; Compute the area of ​​a tile using the Delaunay triangulation algorithm: For a given three point coordinates, the area S can be calculated using two-dimensional geometry: (9) in, are the coordinates of the three vertices; For each slice, multiply the slice's area by its thickness , that is, the distance between adjacent slices on the z-axis, to approximate the volume of the slice, and the approximate volume of the entire 3D point cloud is obtained as follows: (10) where Y is the total volume, is the area of ​​the ith slice, is the slice thickness and m is the number of slices.

[0013] Compared with the prior art, the present invention has the following beneficial effects: According to a method for calculating volume of three-dimensional point cloud reconstruction of the present invention, the point cloud can be accurately calculated by calculus method after the point cloud model is reconstructed.

[0014] This method can complete the missing parts through three-dimensional point cloud reconstruction to obtain a closed three-dimensional model, further reducing the dependence on data integrity. It can be applied to more scenes with incomplete scanned point cloud maps. In addition, by calculating the volume through calculus methods, it can further not make requirements on the shape of the three-dimensional point cloud model and calculate the volume of any scanned scene.

[0015] It is mainly used in the livestock breeding field to monitor the material level in the silo. It can also be used to calculate various scenarios using radar scanning in the industrial field and reconstruct the scanned point cloud map. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1A flowchart of a method for calculating volume of three-dimensional point cloud reconstruction according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0017] Various exemplary embodiments, features and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0018] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0019] In addition, in order to better illustrate the present disclosure, numerous specific details are given in the following specific embodiments. It should be understood by those skilled in the art that the present disclosure can also be implemented without certain specific details. In some examples, methods, means, components and circuits well known to those skilled in the art are not described in detail in order to highlight the subject matter of the present disclosure.

[0020] A method for calculating volume of three-dimensional point cloud reconstruction, the method comprising: Step S1, obtaining point cloud data inside the silo; Step S2, generating an initialization grid M according to the point cloud data; Step S3, generating an Alpha shape of the point cloud according to the point cloud data; Step S4, trimming the grid M and deleting the point cloud data outside the Alpha shape; Step S5, performing Laplace smoothing on the point cloud data processed in step S4; Step S6, performing hole filling processing based on the minimum surface on the point cloud data processed in step S5 to obtain a three-dimensional reconstructed model; Step S7, calculating the volume of the three-dimensional reconstructed model according to the principle of calculus.

[0021] A method for calculating volume of 3D point cloud reconstruction involves a point cloud reconstruction model, which uses an implicit surface reconstruction method to generate an initial 3D point cloud model, and combines the Alpha shape generated by the initial point cloud to prune the initial 3D point cloud model, delete points outside the Alpha shape, and reconstruct an accurate, smooth and redundant 3D point cloud.

[0022] The implicit surface reconstruction and implicit surface function are flexible in construction and suitable for representing complex objects. The implicit surface reconstruction function adopts a Poisson reconstruction method, which utilizes the normal vector information in the point cloud and converts the surface problem into a solution formula of the Poisson equation.

[0023] When the laser radar scans and obtains three-dimensional point cloud information, the laser radar is equipped with a gyroscope, which can perform Z-axis calibration on the point cloud image and obtain normal vector information based on the local geometric characteristics of all points in the point cloud, with the normal pointing to the direction inside and outside the object.

[0024] In one possible implementation, step S1, obtaining point cloud data inside the silo, includes: obtaining point cloud data through a laser radar, the laser radar is equipped with a gyroscope, the gyroscope performs Z-axis calibration on the point cloud image, and obtaining normal vector information n according to the local geometric characteristics of all points in the point cloud data, and the normal points to the direction inside and outside the object, wherein geometry specifically refers to the spatial distribution, shape characteristics and local structure information of points in the point cloud data.

[0025] In a possible implementation, step S2, generating an initialization grid M according to the point cloud data, includes: The normal vectors of all points in the point cloud data are regarded as a gradient field, and the vector field of the unknown object M is obtained. , there is an implicit function f whose gradient corresponds to the normal vector field: (1) in, is a point in space when the gradient With vector field When the modulus of the difference is minimum, the gradient of formula (1) is satisfied Approximation to a vector field ,Right now: (2) Perform divergence calculation on the above formula (2) Transformed into Poisson's equation: (3) According to the point cloud normal vector n, calculate the vector field: (4) Constructing Divergence : (5) The Poisson equation is discretized using the finite element method to form a linear system: (6) in The Laplace matrix is ​​constructed based on the neighborhood relationship of the point cloud. The conjugate gradient method is used to solve formula (6) to obtain the potential function , use the marching cube algorithm to extract the isosurface. The essence of the marching cube algorithm is to use three points calculated by difference in each unit cube to form a set of triangular facets to form its isosurface, that is, to initialize the mesh surface M.

[0026] The three-dimensional point cloud initialization surface M is obtained by the above Poisson reconstruction method. However, the edge part of the generated three-dimensional reconstruction will be redundant, resulting in a larger subsequent volume calculation.

[0027] In a possible implementation, step S3, generating an Alpha shape of the point cloud according to the point cloud data, includes: Alpha shape is a geometric structure based on Delaunay triangulation, which is used to describe the boundary of a point set. For a point set P and parameter α, the smaller the value of α, the finer the generated boundary; The Alpha algorithm is defined as follows: For each point and for the edge Calculate the radius of a circle between two points ,if , it is retained in the Alpha shape.

[0028] In a possible implementation, step S4, pruning the grid M to delete the point cloud data outside the Alpha shape, includes: Use the generated Alpha shape Alpha (P, α) to repair the Poisson reconstructed mesh M. For each triangle or vertex in the mesh M, check whether it is inside the Alpha shape; If the triangle is outside the Alpha shape, a fast Boolean operation is used to delete it. For the triangle that is partially outside the Alpha shape, the boundary of the Alpha shape is used for clipping. Clipping means: intersecting the triangle that is partially outside the Alpha shape with the boundary of the Alpha shape and cutting the triangle to keep the part that is inside the Alpha shape. The fast Boolean operation uses an AABB tree to quickly detect and process the intersection, union and difference operations of a complex triangular mesh, wherein the AABB tree is a binary tree based on the nodes of an axis-aligned bounding box.

[0029] Step S5, performing Laplace smoothing processing on the point cloud data processed in step S4, and step S6, performing hole filling processing based on the minimum surface on the point cloud data processed in step S5 to obtain a three-dimensional reconstructed model, including: The Laplace smoothing is performed on the pruned Poisson reconstruction network, and the formula is as follows: (7) in is the updated position of the i-th vertex at the t+1th iteration, is the vertex position at iteration t, N i is the neighbor geometry of vertex i, that is, the vertices adjacent to vertex i, λ is the smoothing sparsity, which is used to control the strength of smoothing, is the position of neighbor vertex j; Minimum surface hole filling is to deal with holes in the mesh caused by data loss or reconstruction error in 3D reconstruction. The minimum surface satisfies the partial differential equation: (8) Where z is the height of the three-dimensional surface on the plane xy, and the curvature of the surface is 0, which means that the area of ​​any small region of the surface is the smallest.

[0030] Laplacian smoothing is performed on the pruned Poisson reconstruction network to reduce noise or irregularities in the Poisson reconstruction process, ensuring that the mesh is smoother and avoids sharp edges.

[0031] By dividing the space and building a bounding box hierarchy, the intersection detection between triangular meshes is accelerated, avoiding a large amount of unnecessary calculations.

[0032] The above implicit surface reconstruction method can reconstruct the three-dimensional point cloud model and complete the missing parts. In addition, the Alpha algorithm is used to correct the point cloud to restore the original shape of the three-dimensional point cloud to the greatest extent, providing an accurate point cloud shape for the subsequent calculation of the accurate volume.

[0033] In a possible implementation, step S7, calculating the volume of the three-dimensional reconstructed model according to the principle of calculus, includes: First, select the slicing direction: determine the direction in which the point cloud is divided, usually the Z axis (height) is selected for slicing After determining the slice direction, determine the slice thickness: set the slice thickness, usually with a small value. According to the principle of calculus, the smaller the value, the more accurate the calculated volume; Slicing operation: Reconstruct the model through the point cloud and divide the point cloud into multiple horizontal slices according to the set slice thickness; Compute the area of ​​a tile using the Delaunay triangulation algorithm: The triangulation is the process of dividing a two-dimensional point set into multiple non-overlapping triangles, so that no point falls within the circumcircle of any triangle, generating more stable triangles. For a given three point coordinates, the area S can be calculated by two-dimensional geometry: (9) in, are the coordinates of the three vertices; For each slice, multiply the slice's area by its thickness , that is, the distance between adjacent slices on the z-axis, to approximate the volume of the slice, and the approximate volume of the entire 3D point cloud is obtained as follows: (10) where Y is the total volume, is the area of ​​the ith slice, is the slice thickness and m is the number of slices.

[0034] By using the optimized 3D reconstruction algorithm, an accurate, smooth and non-redundant reconstruction model is generated based on the point cloud image scanned by the radar, so that the volume can be accurately calculated.

[0035] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to explain the principles of the embodiments, practical applications, or technical improvements in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A method for calculating volume by reconstructing 3D point clouds. It is characterized in that The method comprises: Step S1, obtaining point cloud data inside the silo; Step S2, generating an initialization grid M according to the point cloud data; Step S3, generating an Alpha shape of the point cloud according to the point cloud data; Step S4, trimming the grid M and deleting the point cloud data outside the Alpha shape; Step S5, performing Laplace smoothing on the point cloud data processed in step S4; Step S6, performing hole filling processing based on the minimum surface on the point cloud data processed in step S5 to obtain a three-dimensional reconstructed model; Step S7, calculating the volume of the three-dimensional reconstructed model according to the principle of calculus.

2. A method for calculating volume of three-dimensional point cloud reconstruction according to claim 1, It is characterized in that Step S1, obtaining point cloud data inside the silo, including: obtaining point cloud data through a laser radar, the laser radar is equipped with a gyroscope, the gyroscope performs Z-axis calibration on the point cloud image, and obtaining normal vector information n according to the local geometric characteristics of all points in the point cloud data, and the normal points to the direction inside and outside the object, wherein geometry specifically refers to the spatial distribution, shape characteristics and local structure information of points in the point cloud data.

3. The method for calculating volume of three-dimensional point cloud reconstruction according to claim 1, It is characterized in that Step S2, generating an initialization grid M according to the point cloud data, comprising: The normal vectors of all points in the point cloud data are regarded as a gradient field, and the vector field of the unknown object M is obtained. , there is an implicit function f whose gradient corresponds to the normal vector field: （1） in, is a point in space when the gradient With vector field When the modulus of the difference is minimum, the gradient of formula (1) is satisfied Approximation to a vector field ,Right now: (2) Perform divergence calculation on the above formula (2) Transformed into Poisson's equation: (3) According to the point cloud normal vector n, calculate the vector field: (4) Constructing Divergence : (5) The Poisson equation is discretized using the finite element method to form a linear system: (6) in The Laplace matrix is ​​constructed based on the neighborhood relationship of the point cloud. The conjugate gradient method is used to solve formula (6) to obtain the potential function , use the marching cube algorithm to extract the isosurface. The essence of the marching cube algorithm is to use three points calculated by difference in each unit cube to form a set of triangular facets to form its isosurface, that is, to initialize the mesh surface M.

4. A method for calculating volume of three-dimensional point cloud reconstruction according to claim 1, It is characterized in that Step S3, generating an Alpha shape of the point cloud according to the point cloud data, comprising: Alpha shape is a geometric structure based on Delaunay triangulation, which is used to describe the boundary of a point set. For a point set P and parameter α, the smaller the value of α, the finer the generated boundary; The definition of the Alpha algorithm is as follows: For each point and for the edge Calculate the radius of a circle between two points ,if , it is retained in the Alpha shape.

5. A method for calculating volume of three-dimensional point cloud reconstruction according to claim 4, It is characterized in that Step S4, trimming the grid M and deleting the point cloud data outside the Alpha shape, including: Use the generated Alpha shape Alpha (P, α) to repair the Poisson reconstructed mesh M. For each triangle or vertex in the mesh M, check whether it is inside the Alpha shape; If the triangle is outside the Alpha shape, a fast Boolean operation is used to delete it. For the triangle that is partially outside the Alpha shape, the boundary of the Alpha shape is used for clipping. Clipping means: intersecting the triangle that is partially outside the Alpha shape with the boundary of the Alpha shape and cutting the triangle to keep the part that is inside the Alpha shape. The fast Boolean operation uses an AABB tree to quickly detect and process the intersection, union and difference operations of a complex triangular mesh, wherein the AABB tree is a binary tree based on the nodes of an axis-aligned bounding box.

6. A method for calculating volume of three-dimensional point cloud reconstruction according to claim 5, It is characterized in that Step S5, performing Laplace smoothing processing on the point cloud data processed in step S4, and step S6, performing hole filling processing based on the minimum surface on the point cloud data processed in step S5 to obtain a three-dimensional reconstructed model, including: The Laplace smoothing is performed on the pruned Poisson reconstruction network, and the formula is as follows: （7） in is the updated position of the i-th vertex at the t+1th iteration, is the vertex position at iteration t, N i is the neighbor geometry of vertex i, that is, the vertices adjacent to vertex i, λ is the smoothing sparsity, which is used to control the strength of smoothing, is the position of neighbor vertex j; Minimum surface hole filling is to deal with holes in the mesh caused by data loss or reconstruction error in 3D reconstruction. The minimum surface satisfies the partial differential equation: （8） Where z is the height of the three-dimensional surface on the plane xy, and the curvature of the surface is 0, which means that the area of ​​any small region of the surface is the smallest.

7. The method for calculating volume of three-dimensional point cloud reconstruction according to claim 3, It is characterized in that Step S7, calculating the volume of the three-dimensional reconstruction model according to the principle of calculus, includes: First, select the slicing direction: determine the direction of dividing the point cloud, and select the Z axis for slicing. After determining the slice direction, determine the slice thickness: set the slice thickness with a small value. According to the principle of calculus, the smaller the value, the more accurate the calculated volume. Slicing operation: Reconstruct the model through the point cloud and divide the point cloud into multiple horizontal slices according to the set slice thickness; Compute the area of ​​a tile using the Delaunay triangulation algorithm: For a given three point coordinates, the area S can be calculated using two-dimensional geometry: （9） in, are the coordinates of the three vertices; For each slice, multiply the slice's area by its thickness , that is, the distance between adjacent slices on the z-axis, to approximate the volume of the slice, and the approximate volume of the entire 3D point cloud is obtained as follows: (10) where Y is the total volume, is the area of ​​the ith slice, is the slice thickness and m is the number of slices.

Images