Method for constructing temperature field based on 3D model point cloud mesh data

Abstract

This invention discloses a method for constructing a temperature field based on 3D model point cloud mesh data. The method comprises the following steps: obtaining image data using drone oblique photography; preprocessing the raw data, including reducing resolution, converting grayscale images, and sharpening edges, to optimize the data volume and modeling effect required for modeling; establishing a basic 3D model, comparing it with other preprocessing methods, and evaluating and adjusting parameters such as model accuracy, time, and point cloud quantity until they meet specific project requirements; extracting point clouds based on the model and performing semantic segmentation; meshing the point cloud data; performing heat treatment using the point cloud mesh data; and establishing a simplified temperature field. This method simplifies the pre-modeling data through preprocessing, facilitating efficient and accurate model construction and extraction of the point cloud data required for temperature field calculation.

Description

Technical Field

[0001] The present invention relates to three-dimensional reconstruction technology, in particular to a method for constructing a temperature field based on three-dimensional model point cloud grid data. Background Art

[0002] 3D reconstruction involves creating mathematical models of 3D objects suitable for computer representation and processing. It serves as the foundation for processing, manipulating, and analyzing the properties of 3D objects in a computer environment. It is also a key technology for creating virtual reality representations of the objective world within computers. With the rapid development of 3D modeling technology, temperature field processing systems utilizing 3D reconstructed point cloud data have become a rapidly evolving thermal processing method widely used by surveying, mapping, and simulation professionals. The same set of raw data from drone oblique photography has led to a variety of modeling approaches with varying emphases, depending on the model's intended use.

[0003] Many fields that require scene modeling and calculation are actively trying to combine them with 3D reconstruction technology. For example, in the mining field, many mining companies use drone 3D oblique photography technology to measure the appearance of mines and important locations before mining. In the field of ancient building research, by collecting multi-perspective images of ancient buildings and using 3D models to reconstruct them, the ancient buildings are protected and digitally archived.

[0004] Related research has also been conducted in the field of heat treatment. For example, to verify the performance of temperature-indicating paint, a Kinect depth camera and color camera were calibrated. Deep image filtering, point cloud segmentation, registration, and fusion were then used to perform 3D reconstruction to verify the temperature of the temperature-indicating paint on each module. To identify and predict engine status and potential safety hazards, a ray tracing algorithm was used to reconstruct and analyze the temperature field of aircraft engine exhaust.

[0005] However, research on large-scale temperature field reconstruction remains limited. The limited available literature typically calculates the temperature field using existing point cloud data, with the primary focus on physical phenomena such as thermal radiation and convection in three-dimensional space. There has been no effort to improve the 3D reconstruction process itself to better address temperature field requirements. Summary of the Invention

[0006] The purpose of the present invention is to propose a method for constructing a temperature field based on three-dimensional model point cloud grid data, thereby effectively improving modeling efficiency.

[0007] The technical solution to achieve the purpose of the present invention is: a method for constructing a temperature field based on three-dimensional model point cloud grid data, comprising the following steps:

[0008] Step 1: Obtain image data using drone oblique photography;

[0009] Step 2: Preprocess the raw data, including reducing the resolution, converting the grayscale image, and sharpening the edges;

[0010] Step 3: Build a basic 3D model, evaluate and adjust the model parameters until they meet the preset requirements;

[0011] Step 4: Perform semantic segmentation on the model;

[0012] Step 5: Meshing of point cloud data;

[0013] Step 6: Create a simple temperature field.

[0014] An electronic device includes a memory, a processor, and a computer program stored in the memory and runnable on the processor. When the processor executes the program, the method for constructing a temperature field based on three-dimensional model point cloud grid data is implemented.

[0015] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the above-mentioned method for constructing a temperature field based on three-dimensional model point cloud grid data.

[0016] Compared with the existing technology, the present invention has the following significant advantages: (1) For the three-dimensional model used to construct the temperature field, a method for optimizing the processing speed is designed in a targeted manner; and a complete six-part technical route is provided; (2) A new flight path is designed by combining five-way and tic-tac-toe flight, which greatly shortens the data acquisition time while ensuring data accuracy; (3) An image preprocessing method is proposed that ignores non-essential data such as color and vegetation details and emphasizes low edge discreteness, which can effectively reduce edge discreteness and reprojection errors after modeling; (4) The operator is allowed to process the point cloud directly, which is more convenient and direct, without converting the point cloud into a more regular format such as voxels or images of a specific perspective; (5) Through the surface thermal model and the ray tracing model, a simple temperature field can be obtained by simulation calculation using the point cloud patch data of some building facades in the present invention.

[0017] The present invention is further described in detail below with reference to the accompanying drawings: BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a flow chart of the method for constructing a temperature field improvement speed based on three-dimensional model point cloud mesh data according to the present invention.

[0019] Figure 2 This is a graph of temperature changes at a typical point within 48 hours. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0021] Combine Figure 1 The present invention provides a method for constructing a temperature field based on three-dimensional model point cloud grid data, the method comprising the following steps:

[0022] Step 1: Use drone oblique photography to obtain image data, including flight parameter design and route design;

[0023] Step 2: Preprocess the raw data, including reducing the resolution, converting the grayscale image, and sharpening the edges, to optimize the data volume and modeling effect required for modeling;

[0024] Step 3: Build a basic 3D model, evaluate and adjust the model parameters until they meet the specific project requirements;

[0025] Step 4: Perform semantic segmentation on the model;

[0026] Step 5: Meshing of point cloud data;

[0027] Step 6: Create a simple temperature field.

[0028] Furthermore, in one embodiment, the step 1 of obtaining image data by using drone oblique photography specifically includes:

[0029] Step 1-1: Interpret the built-in parameter design method of the drone, and experimentally compare and select the flight parameters with the highest aerial survey efficiency. The built-in parameters include flight altitude and gimbal angle;

[0030] In steps 1-2, the drone route design is to use omap software to divide the shooting area into five block areas: east, west, south, north and center. The center area is the target core modeling area. The five divided areas are photographed from four oblique directions and one vertical direction using a 2D route. Data collection is completed after one flight.

[0031] The solution of this embodiment divides the shooting area into five zones, which are photographed at vertical and four oblique angles using 2D flight routes. Compared with traditional oblique photography flight methods: five-way flight and tic-tac-toe flight, this method shortens the shooting time and reduces the amount of shooting data, while more accurately sampling the facades of the houses in the central area of ​​the shooting.

[0032] Furthermore, in one embodiment, the pre-processing of the raw data in step 2 specifically includes:

[0033] Step 2-1, reduce the resolution and shrink the image to 1 / 4 of the original pixel accuracy;

[0034] Step 2-2, convert the color image into a grayscale image;

[0035] Step 2-3, use edge sharpening to reduce edge discreteness, and use operator template;

[0036] The solution of this embodiment is adopted, in which the resolution is reduced and the color image is converted into a grayscale image to effectively shorten the modeling time. The modeling time after image preprocessing is 1 / 3 of that without preprocessing; after edge sharpening is used to reduce edge discreteness, the edge of the object is clearly highlighted and the overall accuracy is improved.

[0037] Furthermore, in one embodiment, the step 3 of establishing a basic three-dimensional model specifically includes:

[0038] Step 3-1, search the image positions at all scales; identify potential scale- and rotation-invariant points of interest through the Gaussian differential function. The Gaussian differential function formula used is:

[0039]

[0040] Where: σ is the standard deviation of the normal distribution, r is the blur radius, the blur radius refers to the distance from the template element to the template center, and N is the number of spatial dimensions;

[0041] In step 3-2, key points are selected based on their stability, and at each candidate location, a fine-tuned model is fitted to determine the position and scale. The matrix model used is:

[0042] (6σ+1)×(6σ+1)

[0043] In step 3-3, the local gradient of the image is measured at the selected scale in the neighborhood around each key point. The gradient and directional distribution characteristics of the pixels in the 3-neighborhood window of the Gaussian pyramid image are collected. The modulus and direction formula used to measure the gradient are:

[0044] Modulus value: Directional formula θ(x,y)=tan -1 ((L(x,y+1)-L(x,y-1)) / L(x+1,y)-L(x-1,y)))

[0045] Where: L is the scale space value of the key point; x, y are the reference directions of the key point;

[0046] In steps 3-4, all subsequent operations on the image data are transformed relative to the direction, scale, and position of the keypoints, thereby providing invariance to these transformations. The rotation invariance formula used is:

[0047]

[0048] Where r is the radius of the sampling point. This step will ensure that the gradient direction is in the x direction. is the coordinate value after rotation;

[0049] Steps 3-5, match feature points between two images, that is, for each image pair I and J, consider each feature f∈F(I) and find the nearest neighbor feature vector f nn ∈F(J), where f nn Expressed as:

[0050]

[0051] Where: F(I) represents the feature points around image I, F(J) represents the feature points around image J, f d is the distance between the point and the nearest neighbor matching vector, f′ d The distance between the point and the second nearest neighbor matching vector

[0052] In steps 3-6, use SVD to find the null space of the matrix to the left of coordinate x, and then normalize the last element to 1 to obtain x, the size of the 3D reconstructed object.

[0053] The solution of this embodiment is highly robust and suitable for extracting various image feature point information of scale transformation and rotation angle; it is highly accurate and does not require time cost consideration in offline algorithms; it obtains the position information (x, y) of the feature points, and can also obtain size and direction information.

[0054] Furthermore, in one embodiment, the semantic segmentation in step 4 specifically includes:

[0055] Step 4-1, extract point cloud data using SFM algorithm;

[0056] Step 4-2, perform affine transformation on the input point cloud, i.e., the xyz three-dimensional coordinates of a fixed number of points;

[0057] Step 4-3, extract the feature layer through two layers of perceptual muscles and connect each point;

[0058] Step 4-4, a higher-dimensional affine transformation, makes the solved affine transformation matrix close to an orthogonal matrix that does not lose information. The radiation transformation matrix used is:

[0059] L reg =||I-AAT || F 2

[0060] Where: L is the matrix after affine transformation, I is the orthogonal matrix, and A is the radiation transformation matrix to be estimated;

[0061] Step 4-5: extract global features from the entire point cloud;

[0062] Steps 4-6: Concatenate the local features and global features of each point to obtain an output of dimension nxm, which corresponds to the prediction score of each class.

[0063] In steps 4-7, based on the classification prediction score of each point, adjacent points of the same type are integrated and points of different types are separated, finally obtaining an independent target point cloud after semantic segmentation.

[0064] By adopting the solution of this embodiment, since point clouds have the characteristics of disorder, sparsity, and limited information, there is no need to convert the point clouds into more regular formats such as voxels. The operator can directly process the point clouds, which is more convenient and direct.

[0065] Furthermore, in one embodiment, the gridding of the point cloud data in step 5 specifically includes:

[0066] Step 5-1, set a 3D sphere with a radius of 10 cm and place it on the point cloud data;

[0067] Step 5-2: If the ball hits three points (and will not return to these three points), create a triangle based on the three points. Then start rolling along the sides of the existing triangle and continue to create new triangles.

[0068] Compared with the traditional method, the solution of this embodiment can achieve the advantages of uniform accuracy after point cloud meshing, adjustable number of facets, and no envelope surface generated due to discrete points.

[0069] Furthermore, in one embodiment, the step 6 of establishing a simple temperature field specifically includes:

[0070] Step 6-1: Establish a surface thermal model. The surface energy heat balance equation used is:

[0071] Q sun +Q skyr +M g +H+LE+G=0

[0072] Where: Q sun The solar shortwave radiation absorbed by the Earth's surface; Q skyr M is the atmospheric long-wave radiation absorbed by the surface;g is the radiant emissivity of the surface; H is the sensible heat exchange between the surface and the environment; LE is the latent heat exchange caused by water loss from the surface; G is the heat flux obtained by the surface;

[0073] Step 6-2: Calculate the latent heat and sensible heat flux of the bare surface. The sensible heat flux formula used is:

[0074] H=ρ a C P C D μ a ′(T a -T g )

[0075] Where: ρ a is the air density (kg / m 3 );C p is the specific heat capacity of air at constant pressure (J / kg·K); T a is the atmospheric temperature at the reference altitude (K), T g is the surface temperature (K); μ a ` is wind speed (m / s); C D is the drag coefficient;

[0076] The latent heat flux formula used is:

[0077] LE=ρ a LC D μ a w s (q′ a -q c )

[0078] Where: ρ a is the air density (kg / m 3 ); L is the latent heat of vaporization of water (J / kg); C D is the drag coefficient; W s is the surface water content; q a ` is the specific humidity at the reference height; q c is the saturated specific humidity at the surface temperature;

[0079] Step 6-3: Latent heat exchange on artificial surface. When considering artificial surface, it can be assumed that there is no water evaporation. The latent heat flux is set to zero, and the sensible heat flux is calculated based on the bare surface.

[0080] Step 6-4, write a temperature field calculation program.

[0081] Combine Figure 2The temperature change of the sample point is consistent with the expected temperature change, and the images obtained by selecting other points are also basically consistent with the trend of the curve; therefore, the thermal calculation method can be applied to the point cloud grid data, and finally processed into simulated temperature field data.

[0082] This paper proposes a comprehensive optimization method for constructing a temperature field from point cloud meshes. It further refines the technical approach from drone data acquisition to temperature field construction, encompassing drone flight, image preprocessing, 3D point cloud reconstruction, point cloud segmentation, and point cloud meshing. Ultimately, the model is constructed efficiently and accurately, extracting the point cloud data required for temperature field calculation.

[0083] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the foregoing embodiments. The foregoing embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for constructing a temperature field based on three-dimensional model point cloud grid data, characterized in that: The following steps are involved: Step 1: Obtain image data using drone oblique photography; Step 2: Preprocess the original data, including reducing the resolution, converting the grayscale image, and sharpening the edges; Step 3: Build a basic 3D model, evaluate and adjust the model parameters until they meet the preset requirements; The establishment of the basic three-dimensional model specifically includes: Step 3-1, search the image positions at all scales; identify potential points of interest that are invariant to scale and rotation by using the Gaussian differential function. The Gaussian differential function formula used is: Where σ is the standard deviation of the normal distribution, r is the blur radius, and N is the number of spatial dimensions; In step 3-2, key points are selected based on their stability, and at each candidate location, a fine-tuned model is fitted to determine the position and scale. The matrix model used is: (6σ+1)×(6σ+1) In step 3-3, in the neighborhood around each key point, measure the local gradient of the image at the selected scale; collect the gradient and direction distribution characteristics of the pixels in the 3-neighborhood window of the Gaussian pyramid image; the modulus and direction formula used to measure the gradient is: Modulus value: Solution:θ(x,y)=tan -1 ((L(x,y+1)-L(x,y-1)) / L(x+1,y)-L(x-1,y))) Where L is the scale space value of the key point; x, y are the reference directions of the key point; Steps 3-4 transform the direction, scale, and position of the key points of the image data to provide invariance to these transformations; the rotation invariance formula is: In the formula, r is the radius of the sampling point. This step will ensure that the gradient direction is in the x direction. is the coordinate value after rotation; Steps 3-5, match feature points between two images, that is, for each image pair I and J, consider each feature f∈F(I) and find the nearest neighbor feature vector f nn ∈F(J), where f nn Expressed as: Where: F(I) represents the feature points around image I, F(J) represents the feature points around image J, f d is the distance between the point and the nearest neighbor matching vector, f' d is the distance between the point and the second nearest neighbor matching vector; Step 3-6: Use SVD to find the null space of the matrix to the left of coordinate x, and then normalize the last element to 1 to obtain x, the size of the 3D reconstructed object; Step 4: Perform semantic segmentation on the model; Step 5: Meshing the point cloud data; Step 6: Establish the temperature field.

2. The method for constructing a temperature field based on three-dimensional model point cloud mesh data according to claim 1, characterized in that: The drone oblique photography described in step 1 includes flight parameter design and route design, as follows: Step 1-1: Explain the design method of the drone's built-in parameters and conduct experiments to select the most efficient flight parameters for aerial surveys. The built-in parameters include flight altitude and gimbal angle. In steps 1-2, the drone route design is to use omap software to divide the shooting area into five block areas: east, west, south, north and center. The center area is the target core modeling area. The five divided areas are photographed from four oblique directions and one vertical direction using a 2D route. Data collection is completed after one flight.

3. The method for constructing a temperature field based on three-dimensional model point cloud mesh data according to claim 2, characterized in that: The pre-processing of the raw data described in step 2 specifically includes: Step 2-1, reduce the resolution and shrink the image to 1 / 4 of the original pixel accuracy; Step 2-2, convert the color image into a grayscale image; Step 2-3, use edge sharpening to reduce edge discreteness, and use operator template.

4. The method for constructing a temperature field based on three-dimensional model point cloud mesh data according to claim 1, characterized in that: The semantic segmentation described in step 4 specifically includes: Step 4-1, extract point cloud data using SFM algorithm; Step 4-2, perform affine transformation on the input point cloud, i.e., the xyz three-dimensional coordinates of a fixed number of points; Step 4-3, extract the feature layer through two-layer perceptron and connect each point; Step 4-4, a higher-dimensional affine transformation, makes the solved affine transformation matrix close to an orthogonal matrix that does not lose information; the radiation transformation matrix used is: L reg =||I-AA T || F 2 Where, L reg is the matrix after affine transformation, I is the orthogonal matrix, and A is the radiation transformation matrix to be estimated; Steps 4-5: extract all features from the entire point cloud; Steps 4-6: Concatenate the local features and global features of each point to obtain an output of dimension nxm, which corresponds to the prediction score of each class. In steps 4-7, based on the classification prediction score of each point, adjacent points of the same type are integrated and points of different types are separated, finally obtaining an independent target point cloud after semantic segmentation.

5. The method for constructing a temperature field based on three-dimensional model point cloud mesh data according to claim 4, characterized in that: The point cloud data gridding described in step 5 specifically includes: Step 5-1, set a 3D sphere with a radius of 10 cm and place it on the point cloud data; In step 5-2, if the ball hits three points, create a triangle based on the three points; and start rolling along the sides of the existing triangle to continue creating new triangles.

6. The method for constructing a temperature field based on three-dimensional model point cloud mesh data according to claim 5, characterized in that: Step 6 of establishing the temperature field specifically includes: Step 6-1: Establish a surface thermal model. The surface energy heat balance equation used is: Q sun +Q skyr +M g +H+LE+G=0 Where Q sun The solar shortwave radiation absorbed by the Earth's surface; Q skyr M is the atmospheric long-wave radiation absorbed by the surface; g is the radiant emissivity of the surface; H is the sensible heat exchange between the surface and the environment; LE is the latent heat exchange caused by water loss from the surface; G is the heat flux obtained by the surface; Step 6-2: Calculate the latent heat and sensible heat flux of the bare surface. The sensible heat flux formula used is: H=ρ a C P C D m a ′(T a -T g ) Where, ρ a is the air density, C p is the specific heat capacity of air at constant pressure, T a is the atmospheric temperature at the reference altitude, T g is the surface temperature; μ a ′ is wind speed, C D is the drag coefficient; The latent heat flux formula used is: LE=ρ a LC D m a w s (q a ′-q c ) Where L is the latent heat of vaporization of water, W s is the surface water content, q a ′ is the specific humidity at the reference height, q c is the saturated specific humidity at the surface temperature; Step 6-3: Latent heat exchange on artificial surface. When considering artificial surface, it is assumed that there is no water evaporation, so the latent heat flux is set to zero and the sensible heat flux is calculated as if it were a bare surface. Step 6-4, write a temperature field calculation program.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the method for constructing a temperature field based on three-dimensional model point cloud mesh data as described in any one of claims 1 to 6 is implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method for constructing a temperature field based on three-dimensional model point cloud grid data as described in any one of claims 1 to 6 is implemented.

Images