Floor template accurate measurement method based on unmanned aerial vehicle photography
By combining drone photography with SfM and 3D Gaussian scene representation, Alpha Shape segmentation, and sparrow search optimization algorithms, the problems of modeling quality and segmentation accuracy in floor template measurement were solved, and high-precision automated measurement was achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-15
- Publication Date
- 2026-03-31
AI Technical Summary
Existing technologies for floor formwork measurement suffer from poor modeling quality, low segmentation accuracy, and large measurement errors, especially in environments with weak textures and complex backgrounds where high-precision measurement is difficult to achieve.
A method based on UAV photography is adopted, which combines the SfM algorithm to generate sparse point clouds and 3D Gaussian scene representation to generate dense point clouds. The Alpha Shape algorithm is used to extract building outline boundaries, and point cloud segmentation is performed by combining RGB color gradient and Euclidean distance. The Sparrow Search algorithm is introduced to optimize the RANSAC algorithm parameters, so as to achieve plane fitting and 3D size calculation.
It achieves high-precision and automated measurement of floor templates, overcomes the reconstruction defects of traditional methods in weak textures and complex backgrounds, and significantly improves the accuracy and automation of measurement.
Smart Images
Figure CN121767428A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent inspection technology in building engineering, and in particular to a method for accurate measurement of floor formwork based on drone photography. Background Technology
[0002] In the entire construction process, the dimensional accuracy of formwork directly determines the construction quality of the main structure. Currently, the dimensional inspection of floor formwork mainly relies on manual measurement with handheld measuring tapes and calipers, which is inefficient and cannot meet the requirements of high-precision construction. Although UAV-based 3D reconstruction technology has been applied to construction monitoring, existing technologies have obvious limitations: the traditional SfM-MVS method is prone to voids and noise in the point cloud generated when dealing with weak textures and repetitive structures such as floor formwork, and it is difficult to unify the modeling geometry and appearance; due to the complex construction site environment, the point cloud of beam bottom formwork is prone to voids and noise, and existing point cloud segmentation methods (such as Euclidean clustering) are sensitive to the complex background noise of the construction site, which can easily lead to oversegmentation or undersegmentation; in addition, when measuring the flatness or depth of the formwork, the traditional RANSAC plane fitting algorithm relies on a fixed threshold, which cannot adapt to point clouds with local sparse or uneven noise distribution, resulting in large measurement errors. Summary of the Invention
[0003] To address the problems of poor modeling quality, low segmentation accuracy, and large measurement errors in the aforementioned floor formwork measurement methods, this invention proposes a precise measurement method for floor formwork based on UAV photography, achieving high-precision and automated measurement of floor formwork.
[0004] The specific technical solution of the present invention is as follows: A method for accurate measurement of floor formwork based on drone photography includes the following steps: Step 1: Image acquisition and preprocessing of floor template scene (11): Use a drone equipped with a camera to take close-up photos of the floor formwork at the construction site from multiple angles to obtain an image sequence of the floor formwork scene; (12): Read the camera intrinsic parameters in the EXIF information of the image sequence, and perform distortion correction and image enhancement preprocessing on the acquired image sequence to construct an image dataset of the floor template scene; Step 2: Generate sparse and dense point clouds of the floor template scene. (21): The SfM algorithm is used to estimate the camera pose and generate a sparse point cloud of the floor template scene; (22): Using sparse point cloud as initial input, a 3D Gaussian scene representation is established. Through differentiable rasterization rendering and backpropagation optimization, the position, covariance matrix, opacity and spherical harmonic coefficient of each 3D Gaussian ellipsoid in the scene are learned, and finally the dense point cloud of the floor template scene is exported. Step 3: Extract the building outline boundary and segment and extract the point cloud of individual template blocks. (31): The dense point cloud is projected onto a two-dimensional plane, and the Alpha Shape algorithm is used to extract the outline boundary of the point cloud in the building construction area and remove background interference; (32): For point clouds within the contour boundary, calculate the RGB color gradient difference of the point clouds, combine spatial Euclidean distance constraints, construct a fusion distance metric formula, cluster point clouds belonging to the same template, and realize the segmentation and extraction of single template point clouds; Step 4: Calculate the 3D dimensions of the template based on plane fitting optimized by the sparrow search algorithm. (41): For the point clouds of the top and bottom surfaces of the obtained single template, construct a plane fitting parameter optimization model; (42): Introduce the sparrow search algorithm, construct the fitness function with the inlier rate and distance standard deviation of the fitting plane, and automatically optimize the optimal distance threshold and number of iterations of the RANSAC algorithm; (43): Based on the obtained optimal distance threshold and iteration number parameters, the point clouds of the top and bottom surfaces of the template are finally fitted, the normal vector and the inter-surface spacing are calculated, and the length, width and depth dimensions of the template are obtained.
[0005] Furthermore, for step (11), a grid-shaped or circular flight path is set, with a heading overlap rate of ≥80% and a lateral overlap rate of ≥80%, to ensure multi-angle full coverage of the floor template area.
[0006] Furthermore, for the intrinsic parameters such as focal length and principal point coordinates in the EXIF data of the image sequence read in step (12), the image processing algorithm is used to eliminate lens distortion, and the histogram equalization technique is used to enhance image contrast.
[0007] Furthermore, for step (21), the image dataset is imported into the COLMAP software, and the SfM algorithm in the software is used. Specifically, the SIFT algorithm in the SfM algorithm is used to extract image feature points. Through feature matching and bundle adjustment, the camera pose of each image and the sparse point cloud of the floor template scene are generated.
[0008] Furthermore, the camera pose includes a rotation matrix R and a translation vector T.
[0009] Furthermore, for step (22), the 3D Gaussian ellipsoid is projected onto the 2D image plane by differentiable sputtering, the photometric error between the rendered image and the real acquired image is calculated, end-to-end training is performed using stochastic gradient descent, the density of the 3D Gaussian ellipsoid is adaptively controlled, and after iterative optimization, a dense point cloud of the floor template scene containing geometric position and appearance attributes is exported.
[0010] Furthermore, for step (31), the obtained dense point cloud is projected onto the XY plane and meshed. The Alpha Shape algorithm is used to set the radius of the rounding. ,when When the size is small enough, the algorithm can penetrate the gaps in the point cloud, extract the outline boundary of the point cloud in the construction area, and filter out external environmental noise.
[0011] Furthermore, for step (32), a collaborative segmentation algorithm that integrates RGB color information and Euclidean space distance is proposed for the point cloud within the contour boundary, defining the point... With point Fusion distance between : ; in, For spatial Euclidean distance, This refers to the distance in the RGB color space. , The distance metric is used as the weighting coefficient. Region growing clustering is performed based on this distance metric. When two points are spatially adjacent and have similar color and texture, they are classified into the same class. This effectively distinguishes adjacent templates that are closely arranged but have seams, and realizes the segmentation and extraction of single template point clouds.
[0012] Furthermore, in step (42), a distance threshold is selected from the RANSAC algorithm. and maximum number of iterations As an optimization variable, a sparrow search algorithm is introduced, and the population size is set. Maximum number of iterations Define the fitness function : ; in, To fit the number of points in the plane, Total points Let be the standard deviation of the distance from the interior point to the plane. , Using weights, this function aims to find the parameter combination that maximizes the inlier rate and minimizes the fitting error. .
[0013] Furthermore, in step (43), the sparrow population is divided into discoverers, joiners, and vigilants, and iterative search is performed in the solution space. The discoverers are responsible for the global search, the joiners follow the discoverers to conduct local exploration, and the vigilants are responsible for escaping local optima. After the iteration is completed, the global optimal parameters are output. By using the optimal parameters to perform a final fitting of the point clouds on the top and bottom surfaces of the template, the equations of the two parallel planes are obtained: ; Calculate the perpendicular distance between the two planes, and use it as the template depth: ; Calculate the length and width of the minimum bounding rectangle of the planar point cloud to accurately measure the three-dimensional dimensions of the template.
[0014] The beneficial effects of this invention are as follows: It overcomes the reconstruction defects of traditional MVS in weak texture areas by using 3D GS technology, solves the problem of complex background interference by multi-feature collaborative segmentation, and realizes the adaptive adjustment of plane fitting parameters by using intelligent optimization algorithm, thereby achieving accurate measurement of template plane size and depth, and significantly improving the automation level of floor template measurement. Attached Figure Description
[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0016] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a rendering of the dense point cloud of the floor slab template after reconstruction based on 3D GS. Figure 3 This is a schematic diagram of the plane of the point cloud to be fitted. Figure 4 This is a flowchart for optimizing plane fitting parameters based on the sparrow search algorithm.
[0017] The figures are labeled as follows: 1-Point cloud of the template to be fitted on the plate surface, 2-Point cloud of the template at the bottom of the beam to be fitted, 3-Depth to be measured. Detailed Implementation
[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0019] The relevant technical terms and abbreviations are explained below: EXIF (Exchangeable Image File Format) is a metadata standard that records the technical parameters of digital photos. It can be analyzed by professional viewers to obtain detailed information such as the shooting device, time, and GPS.
[0020] Colmap is an open-source computer vision software specifically designed for reconstructing 3D scenes from multiple 2D photographs. It combines Structure-from-Motion (SfM) and Multi-View Stereo (MVS) techniques to automatically recover camera pose, generate point cloud models, and output textured 3D meshes. Developed by Stanford University, this tool supports both graphical and command-line interfaces and is suitable for reverse engineering in fields such as archaeological digitization, film and television special effects, and autonomous driving.
[0021] Structure from Motion (SfM) is a photogrammetric technique that reconstructs the structure of a 3D scene by analyzing image sequences. The SfM algorithm includes feature extraction (such as SIFT), feature matching, camera parameter estimation (including extrinsic and intrinsic parameter matrices), and bundle adjustment optimization. Its mathematical foundation is based on a pinhole camera model and projection matrix calculations, relying on geometric constraints (such as epipolar geometry) and rigid body motion assumptions. Applications include architectural model reconstruction and digital preservation of cultural relics, and it is often used in conjunction with multi-view stereo vision technology.
[0022] 3D Gaussian Splatting (3DGS) is a cutting-edge technology for 3D scene reconstruction and real-time rendering. It explicitly represents the scene using a large number of 3D Gaussian distribution points (i.e., "Gaussian spheres"), enabling high-quality view compositing and efficient rendering. Alpha shape algorithm 4, also known as the rolling ball method, is an algorithm for extracting boundary points.
[0023] RANSAC (Random Sample Consensus) is an efficient method for estimating mathematical model parameters from a sample set containing outliers.
[0024] The Sparrow Search Algorithm (SSA) is a swarm intelligence optimization algorithm proposed in 2020. It simulates the foraging behavior and anti-predation mechanism of a sparrow population to solve complex optimization problems. Example 1
[0025] Reference Figure 1-4 A method for accurate measurement of floor formwork based on drone photography includes the following steps: Step 1: Image acquisition and preprocessing of floor template scene (11): Use a drone equipped with a camera to take close-up photos of the floor formwork at the construction site from multiple angles to obtain an image sequence of the floor formwork scene; (12): Read the camera intrinsic parameters in the EXIF information of the image sequence, and perform distortion correction and image enhancement preprocessing on the acquired image sequence to construct an image dataset of the floor template scene; Step 2: Generate sparse and dense point clouds of the floor template scene. (21): The SfM algorithm is used to estimate the camera pose and generate a sparse point cloud of the floor template scene; (22): Using sparse point cloud as initial input, a 3D Gaussian scene representation is established. Through differentiable rasterization rendering and backpropagation optimization, the position, covariance matrix, opacity and spherical harmonic coefficient of each 3D Gaussian ellipsoid in the scene are learned, and finally the dense point cloud of the floor template scene is exported. Step 3: Extract the building outline boundary and segment and extract the point cloud of individual template blocks. (31): The dense point cloud is projected onto a two-dimensional plane, and the Alpha Shape algorithm is used to extract the outline boundary of the point cloud in the building construction area and remove background interference; (32): For point clouds within the contour boundary, calculate the RGB color gradient difference of the point clouds, combine spatial Euclidean distance constraints, construct a fusion distance metric formula, cluster point clouds belonging to the same template, and realize the segmentation and extraction of single template point clouds; Step 4: Calculate the 3D dimensions of the template based on plane fitting optimized by the sparrow search algorithm. (41): For the point clouds of the top and bottom surfaces of the obtained single template, construct a plane fitting parameter optimization model; (42): Introduce the sparrow search algorithm, construct the fitness function with the inlier rate and distance standard deviation of the fitting plane, and automatically optimize the optimal distance threshold and number of iterations of the RANSAC algorithm; (43): Based on the obtained optimal distance threshold and iteration number parameters, the point clouds of the top and bottom surfaces of the template are finally fitted, the normal vector and the inter-surface spacing are calculated, and the length, width and depth dimensions of the template are obtained. Example 2
[0026] Reference Figure 1-4 A method for accurate measurement of floor formwork based on drone photography includes the following steps: Step 1: Image acquisition and preprocessing of floor template scene (11): Use a drone equipped with a camera to take close-up photos of the floor formwork at the construction site from multiple angles to obtain an image sequence of the floor formwork scene.
[0027] If a DJI Matrice 4E drone is selected as the data acquisition platform, equipped with a high-resolution visible light camera, a tic-tac-toe or circular flight path can be set, with a forward overlap rate of ≥80% and a lateral overlap rate of ≥80%, to ensure multi-angle full coverage of the floor template area.
[0028] (12): Read the camera intrinsic parameters in the EXIF information of the image sequence, and perform distortion correction and image enhancement preprocessing on the acquired image sequence to construct the image dataset of the floor template scene.
[0029] Specifically, intrinsic parameters such as focal length and principal point coordinates are read from the EXIF data of the image sequence. Image processing algorithms are used to eliminate lens distortion, and histogram equalization technology is used to enhance image contrast. The result is a distortion-free, high-quality image dataset, providing a reliable data foundation for subsequent 3D reconstruction.
[0030] Step 2: Generate sparse and dense point clouds of the floor template scene. (21): The SfM algorithm is used to estimate the camera pose and generate a sparse point cloud of the floor template scene.
[0031] The obtained image dataset was imported into COLMAP software, and the SfM algorithm in the software was used, specifically the SIFT algorithm in the SfM algorithm, to extract image feature points. Through feature matching and bundle adjustment, the camera pose (including rotation matrix R and translation vector T) and sparse point cloud of the floor template scene of each image were generated.
[0032] (22): Using sparse point cloud as initial input, a 3D Gaussian scene representation is established. Through differentiable rasterization rendering and backpropagation optimization, the position, covariance matrix, opacity and spherical harmonic coefficient of each 3D Gaussian ellipsoid in the scene are learned, and finally the dense point cloud of the floor template scene is exported.
[0033] Each 3D Gaussian ellipsoid is defined by its position, covariance matrix, opacity, and spherical harmonic coefficients. The 3D Gaussian ellipsoid is projected onto a 2D image plane using differentiable sputtering. The photometric error between the rendered image and the real-world captured image is calculated. End-to-end training is performed using stochastic gradient descent, and the density of the 3D Gaussian ellipsoid is adaptively controlled (cloning under-reconstructed regions and splitting over-reconstructed regions). After iterative optimization, a dense point cloud of the floor template scene containing geometric position and appearance attributes is exported.
[0034] Step 3: Extract the building outline boundary and segment and extract the point cloud of individual template blocks. (31): The dense point cloud is projected onto a two-dimensional plane, and the Alpha Shape algorithm is used to extract the outline boundary of the point cloud in the construction area and remove background interference.
[0035] Specifically, the obtained dense point cloud is projected onto the XY plane and meshed. The Alpha Shape algorithm is used to set the radius of the circle. ,when When the size is small enough, the algorithm can penetrate the gaps in the point cloud, extract the outline boundary of the point cloud in the construction area, and filter out external environmental noise.
[0036] (32): For point clouds within the contour boundary, calculate the RGB color gradient difference of the point clouds, combine spatial Euclidean distance constraints, construct a fusion distance metric formula, cluster point clouds belonging to the same template, and realize the segmentation and extraction of single template point clouds.
[0037] Specifically, for point clouds within contour boundaries, a collaborative segmentation algorithm fusing RGB color information and Euclidean space distance is proposed, defining points... With point Fusion distance between : ; in, For spatial Euclidean distance, This refers to the distance in the RGB color space. , The distance metric is used as the weighting coefficient. Region growing clustering is performed based on this distance metric. When two points are spatially adjacent and have similar color and texture, they are classified into the same class. This effectively distinguishes adjacent templates that are closely arranged but have seams, and realizes the segmentation and extraction of single template point clouds.
[0038] Step 4: Calculate the 3D dimensions of the template based on plane fitting optimized by the sparrow search algorithm. (41): For the point clouds of the top and bottom surfaces of the obtained single template, construct a plane fitting parameter optimization model.
[0039] (42): Introduce the sparrow search algorithm, construct the fitness function with the inlier rate and distance standard deviation of the fitted plane, and automatically optimize the optimal distance threshold and number of iterations of the RANSAC algorithm.
[0040] Specifically, the distance threshold in the RANSAC algorithm is selected. and maximum number of iterations As an optimization variable, a sparrow search algorithm is introduced, and the population size is set. Maximum number of iterations Define the fitness function : ; in, To fit the number of points in the plane, Total points Let be the standard deviation of the distance from the interior point to the plane. , Using weights, this function aims to find the parameter combination that maximizes the inlier rate and minimizes the fitting error. The sparrow population is divided into discoverers, joiners, and vigilants. Iterative search is performed within the solution space. Discoverers are responsible for the global search, joiners follow the discoverers in local exploration, and vigilants are responsible for escaping local optima. After iteration, the globally optimal parameters are output. .
[0041] (43): Based on the obtained optimal distance threshold and iteration number parameters, the point clouds of the top and bottom surfaces of the template are finally fitted, the normal vector and the inter-surface spacing are calculated, and the length, width and depth dimensions of the template are obtained.
[0042] Specifically, using the most optimal parameters The final fitting of the point clouds on the top and bottom surfaces of the template yields the equations for two parallel planes: ; Calculate the perpendicular distance between the two planes, and use it as the template depth: ; Calculate the length and width of the minimum bounding rectangle of the planar point cloud to accurately measure the three-dimensional dimensions of the template.
[0043] Figure 3 Figure 1 shows the point cloud of the template on the plate surface to be fitted; Figure 2 shows the point cloud of the template on the bottom of the beam to be fitted; Figure 3 shows the depth to be measured.
[0044] This invention overcomes the reconstruction shortcomings of traditional MVS in weak texture areas by employing 3D GS technology. It solves the problem of complex background interference through multi-feature collaborative segmentation and utilizes intelligent optimization algorithms to achieve adaptive adjustment of planar fitting parameters, enabling accurate measurement of template planar dimensions and depth, and significantly improving the automation level of floor template measurement. The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A floor formwork precision measurement method based on unmanned aerial photography, characterized in that: The method comprises the following steps: Step 1: floor formwork scene image acquisition and image preprocessing (11): using a camera-equipped unmanned aerial vehicle to perform multi-view close-up photography on the construction site floor formwork to obtain an image sequence of the floor formwork scene; (12): reading the camera intrinsic parameters in the EXIF information of the image sequence, and performing distortion correction and image enhancement preprocessing on the collected image sequence to construct an image dataset of the floor formwork scene; Step 2: generating a sparse point cloud and a dense point cloud of the floor formwork scene (21): estimating the camera pose and generating a sparse point cloud of the floor formwork scene using the SfM algorithm; (22): taking the sparse point cloud as the initialization input, establishing a 3D Gaussian scene representation, learning the position, covariance matrix, opacity and spherical harmonic coefficients of each 3D Gaussian ellipsoid in the scene through differentiable rasterization rendering and back propagation optimization, and finally deriving a dense point cloud of the floor formwork scene; Step 3: extracting the building contour boundary and segmenting and extracting single-piece template point cloud (31): projecting the dense point cloud to a two-dimensional plane, and extracting the contour boundary of the building construction area point cloud using the Alpha Shape algorithm to remove background interference; (32): for the point cloud within the contour boundary, calculating the RGB color gradient difference of the point cloud, combining the spatial Euclidean distance constraint, constructing a fusion distance measurement formula, clustering the point cloud belonging to the same template, and realizing the segmentation and extraction of single-piece template point cloud; Step 4: plane fitting based on sparrow search algorithm optimization, and calculating the three-dimensional size of the template (41): for the top and bottom point clouds of the obtained single-piece template, a plane fitting parameter optimization model is constructed; (42): introducing the sparrow search algorithm to construct a fitness function with the inlier rate and distance standard deviation of the fitted plane, and automatically optimizing the best distance threshold and iteration number of the RANSAC algorithm; (43): based on the obtained best distance threshold and iteration number parameters, the final fitting of the template top and bottom point clouds is performed to calculate the normal vector and the distance between the planes, and the length, width and depth dimensions of the template are obtained. 2.The unmanned aerial vehicle photography-based floor formwork precise measurement method of claim 1, wherein: For step (11), set the cross-shaped or ring-shaped flight route, set the heading overlap rate ≥ 80%, and the lateral overlap rate ≥ 80%, to ensure that the floor formwork area is fully covered by multi-angle shooting. 3.The unmanned aerial vehicle photography based floor formwork precise measurement method of claim 1, wherein: For step (12), read the focal length, principal point coordinates and other intrinsic parameters in the EXIF data of the image sequence, eliminate lens distortion using image processing algorithms, and enhance the image contrast using histogram equalization technology.
4. The drone photography based floor formwork precision measurement method of claim 1, wherein: For step (21), import the image dataset into the COLMAP software, use the SfM algorithm in the software, specifically use the SIFT algorithm in the SfM algorithm, extract image feature points, and generate the camera pose of each image and the sparse point cloud of the floor formwork scene through feature matching and bundle adjustment.
5. The drone photography based floor formwork precision measurement method of claim 4, wherein: The camera pose includes a rotation matrix R and a translation vector T.
6. The drone photography based floor formwork precision measurement method of claim 1, wherein: For step (22), the 3D Gaussian ellipsoid is projected to the 2D image plane by differentiable sputtering, the photometric error between the rendered image and the real captured image is calculated, the end-to-end training is carried out by using the stochastic gradient descent method, the density of the 3D Gaussian ellipsoid is adaptively controlled, and after iterative optimization, the dense point cloud of the floor template scene containing the geometric position and appearance attribute is derived.
7. The drone photography based floor formwork precision measurement method of claim 1, wherein: For step (31), the obtained dense point cloud is projected to the X-Y plane and gridded, and the Alpha Shape algorithm is used with a rolling radius of 0.2 m When is small enough, the algorithm can pass through the gaps in the point cloud to extract the contour boundary of the construction area point cloud, and filter out the external environmental noise.
8. The drone photography based floor formwork precision measurement method of claim 1, wherein: For step (32), a collaborative segmentation algorithm is proposed to fuse RGB color information and Euclidean distance for the point cloud within the contour boundary, and the fused distance d between points p and q is defined as d = sqrt((p-q)2 + (R(p)-R(q))2) where R(p) and R(q) are the RGB color values of points p and q, respectively. ； wherein, is the spatial Euclidean distance, is the RGB color space distance, , is the weight coefficient, and the region growing clustering is performed based on the distance metric. When two points are spatially adjacent and similar in color and texture, they are classified into one category, thereby effectively distinguishing adjacent templates with close arrangement but with seams, and realizing the segmentation and extraction of single-piece template point cloud.
9. The drone photography based floor formwork precision measurement method of claim 1, wherein: In step (42), the distance threshold in the RANSAC algorithm is selected and the maximum number of iterations as optimization variables, the sparrow search algorithm is introduced, the population size , the maximum number of iterations , and the fitness function are set. ； where, is the number of inliers in the plane, is the total number of points, is the standard deviation of the distance of inliers to the plane, , is the weight, the function aims to find the combination of parameters with the highest inlier rate and the smallest fitting error .
10. The drone photography based floor formwork precision measurement method as claimed in claim 1, wherein: In step (43), the sparrow population is divided into discoverers, joiners and guards, and iterative search is performed in the solution space. The discoverers are responsible for global search, the joiners follow the discoverers for local development, and the guards are responsible for jumping out of local optimum. After iteration, the global optimal parameters are output The optimal parameters are used to finally fit the template top surface and bottom surface point cloud to obtain the equations of the two parallel planes. ； The vertical distance between the two planes is calculated as the template depth: ； The length and width of the minimum circumscribed rectangle of the plane point cloud are calculated, so as to complete the accurate measurement of the three-dimensional size of the template.