Scene three-dimensional reconstruction method and system, electronic equipment and storage medium

Abstract

The invention relates to the technical field of three-dimensional reconstruction, in particular to a scene three-dimensional reconstruction method and system, electronic equipment and a storage medium, and the method comprises the steps: obtaining image data of multiple image types of a target region collected by multiple image sensors, and carrying out the operation processing of all image data, and obtaining an aerial triangulation result; wherein each image sensor collects image data of one image type; and according to the aerial triangulation measurement result, obtaining and utilizing a three-dimensional coordinate point set to construct an irregular triangular mesh TIN model, and performing texture mapping based on the irregular triangular mesh TIN model to generate a three-dimensional scene model of the target area. According to the method, the accuracy of aerial triangulation is improved, the precision of three-dimensional reconstruction is enhanced, and meanwhile, the sense of reality and the detail expressive force of the model are improved.

Description

Background Art

[0002] In the field of 3D reconstruction technology, traditional scene 3D reconstruction methods often rely on data collected by a single image sensor. These data are relatively limited in type and quantity, and it is difficult to fully and accurately reflect the actual terrain and features of the target area. Therefore, when constructing a 3D model, problems such as insufficient precision and loss of details are often encountered, resulting in the final 3D scene model lacking in realism and accuracy.

[0003] In addition, existing 3D reconstruction methods usually only perform simple preprocessing and feature extraction when processing image data, which makes it easy to be affected by noise and redundant data when extracting 3D coordinate point sets and constructing irregular triangular mesh TIN models, further reducing the accuracy and realism of the 3D model.

[0004] At the same time, traditional texture mapping technology also has certain limitations. Since the correspondence between image data and irregular triangular mesh TIN model is not accurate enough, the overall visual effect of the three-dimensional scene model is affected.

[0005] Therefore, it is urgent to provide a technical solution to solve the above problems. Summary of the invention

[0006] In order to solve the above technical problems, the present invention provides a scene three-dimensional reconstruction method, system, electronic device and storage medium.

[0007] In a first aspect, the present invention provides a method for three-dimensional reconstruction of a scene, and the technical solution of the method is as follows:

[0008] Using multiple image sensors, image data of multiple image types of the target area are collected, and all image data are processed to obtain aerial triangulation results; wherein each image sensor collects image data of one image type respectively;

[0009] According to the aerial triangulation result, an irregular triangular mesh TIN model is obtained and constructed using a three-dimensional coordinate point set, and texture mapping is performed based on the irregular triangular mesh TIN model to generate a three-dimensional scene model of the target area.

[0010] The beneficial effects of a scene 3D reconstruction method of the present invention are as follows:

[0011] The method of the present invention improves the accuracy of aerial triangulation, enhances the precision of three-dimensional reconstruction, and simultaneously improves the realism and detail expression of the model.

[0012] Based on the above solution, a scene 3D reconstruction method of the present invention can also be improved as follows.

[0013] In an optional manner, the step of performing computational processing on all image data to obtain an aerial triangulation result includes:

[0014] Preprocess each image data separately to obtain multiple preprocessed images;

[0015] Extracting multiple feature points from each preprocessed image respectively, matching multiple feature points of any preprocessed image with multiple feature points of a preprocessed image adjacent to or overlapping with the preprocessed image, and obtaining a set of similar feature point pairs corresponding to the preprocessed image;

[0016] Based on the set of similar feature points, a least squares operation or an iterative closest point operation is performed to obtain a relative position and posture relationship between a plurality of pre-processed images, and initial parameters of aerial triangulation of the relative position and posture relationship are determined;

[0017] Based on the initial parameters, the relative position and posture relationship between the multiple preprocessed images is optimized using the bundle adjustment optimization algorithm to obtain the optimization parameters corresponding to each preprocessed image and optimize the corresponding preprocessed images to obtain the optimized images corresponding to each preprocessed image. The optimized images corresponding to each preprocessed image are respectively used as the corresponding preprocessed images, and the step of respectively extracting multiple feature points of each preprocessed image is returned to execute until the preset iteration conditions are met, and the optimization parameters corresponding to each preprocessed image are determined as the aerial triangulation results.

[0018] In the above optional methods, the accuracy and stability of aerial triangulation are significantly improved by refining the process of obtaining aerial triangulation results, including preprocessing, feature point matching, least squares method or iterative closest point calculation, and application of bundle adjustment optimization algorithm. This refined processing flow helps to reduce errors and provide a more reliable data basis for subsequent 3D reconstruction.

[0019] In an optional manner, the preprocessing includes: orientation processing and image processing; the orientation processing includes: relative orientation and absolute orientation; the image processing includes: at least one of denoising, enhancement, correction, filtering, alignment and redundancy removal; the operation processing includes: at least one of the least squares method and iterative closest point algorithm.

[0020] In the above optional method, the specific steps of orientation processing and preprocessing, as well as the selection of operation processing, are clarified, providing clearer guidance for the entire 3D reconstruction process. The clarification and refinement of the above steps help ensure the accuracy and efficiency of 3D reconstruction.

[0021] In an optional manner, the step of obtaining and constructing an irregular triangulated mesh TIN model using a three-dimensional coordinate point set according to the aerial triangulation result includes:

[0022] Based on the aerial triangulation result, extract the three-dimensional coordinate point set including the terrain position information and the object position information of the target area, and filter the three-dimensional coordinate point set to remove noise points and redundant points to obtain a filtered three-dimensional coordinate point set;

[0023] The filtered three-dimensional coordinate point set is connected into a target triangular mesh by using a Delaunay triangulation algorithm to form the irregular triangular mesh TIN model; wherein the target triangular mesh is a continuous and non-overlapping triangular mesh.

[0024] In the above optional method, the accuracy and continuity of the irregular triangular mesh TIN model are ensured by extracting a set of three-dimensional coordinate points, filtering out noise points and redundant points, and using the Delaunay triangulation algorithm to construct the irregular triangular mesh TIN model. The above refined construction process helps to improve the realism and detail expression of the three-dimensional scene model.

[0025] In an optional manner, the step of performing texture mapping based on the irregular triangular mesh TIN model to generate a three-dimensional scene model of the target area includes:

[0026] According to the spatial position information and geometric features of the irregular triangular mesh TIN model, texture information corresponding to the surface of the irregular triangular mesh TIN model in the image data is extracted, and the texture information is mapped onto the surface of the irregular triangular mesh TIN model using a texture mapping algorithm to generate the three-dimensional scene model of the target area.

[0027] In the above optional manner, by extracting texture information based on the spatial position information and geometric features of the irregular triangular mesh TIN model, and mapping the texture information onto the surface of the irregular triangular mesh TIN model using a texture mapping algorithm, a three-dimensional scene model with realism and detail expression is generated. The above texture mapping method not only improves the visual effect of the model, but also enhances the realism and credibility of the model.

[0028] In a second aspect, the present invention provides a scene 3D reconstruction system, the technical solution of the system is as follows:

[0029] The scene 3D reconstruction system comprises: an acquisition module and an execution module;

[0030] The acquisition module is used to: use multiple image sensors to collect image data of multiple image types of the target area, and perform calculations on all the image data to obtain aerial triangulation results; wherein each image sensor collects image data of one image type respectively;

[0031] The execution module is used to obtain and construct an irregular triangular mesh TIN model using a three-dimensional coordinate point set according to the aerial triangulation result, and perform texture mapping based on the irregular triangular mesh TIN model to generate a three-dimensional scene model of the target area.

[0032] The beneficial effects of a scene 3D reconstruction system of the present invention are as follows:

[0033] The system of the present invention improves the accuracy of aerial triangulation, enhances the precision of three-dimensional reconstruction, and simultaneously improves the realism and detail expression of the model.

[0034] Based on the above solution, a scene 3D reconstruction system of the present invention can also be improved as follows.

[0035] In an optional manner, the acquisition module is specifically used to:

[0036] Preprocess each image data separately to obtain multiple preprocessed images;

[0037] Extracting multiple feature points from each preprocessed image respectively, matching multiple feature points of any preprocessed image with multiple feature points of a preprocessed image adjacent to or overlapping with the preprocessed image, and obtaining a set of similar feature point pairs corresponding to the preprocessed image;

[0038] Based on the set of similar feature points, a least squares operation or an iterative closest point operation is performed to obtain a relative position and posture relationship between a plurality of pre-processed images, and initial parameters of aerial triangulation of the relative position and posture relationship are determined;

[0039] Based on the initial parameters, the relative position and posture relationship between the multiple preprocessed images is optimized using the bundle adjustment optimization algorithm to obtain the optimization parameters corresponding to each preprocessed image and optimize the corresponding preprocessed images to obtain the optimized images corresponding to each preprocessed image. The optimized images corresponding to each preprocessed image are respectively used as the corresponding preprocessed images, and the step of respectively extracting multiple feature points of each preprocessed image is returned to execute until the preset iteration conditions are met, and the optimization parameters corresponding to each preprocessed image are determined as the aerial triangulation results.

[0040] In the above optional methods, the accuracy and stability of aerial triangulation are significantly improved by refining the process of obtaining aerial triangulation results, including preprocessing, feature point matching, least squares method or iterative closest point calculation, and application of bundle adjustment optimization algorithm. This refined processing flow helps to reduce errors and provide a more reliable data basis for subsequent 3D reconstruction.

[0041] In an optional manner, the preprocessing includes: orientation processing and image processing; the orientation processing includes: relative orientation and absolute orientation; the image processing includes: at least one of denoising, enhancement, correction, filtering, alignment and redundancy removal; the operation processing includes: at least one of the least squares method and iterative closest point algorithm.

[0042] In the above optional method, the specific steps of orientation processing and preprocessing, as well as the selection of operation processing, are clarified, providing clearer guidance for the entire 3D reconstruction process. The clarification and refinement of the above steps help ensure the accuracy and efficiency of 3D reconstruction.

[0043] In a third aspect, a technical solution of an electronic device of the present invention is as follows:

[0044] The invention comprises a memory, a processor and a program stored in the memory and running on the processor. When the processor executes the program, the steps of the scene three-dimensional reconstruction method of the present invention are implemented.

[0045] In a fourth aspect, the present invention provides a computer-readable storage medium having the following technical solution:

[0046] Instructions are stored in the computer-readable storage medium. When the computer-readable storage medium reads the instructions, the computer-readable storage medium executes the steps of the scene three-dimensional reconstruction method of the present invention.

[0047] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The accompanying drawings are only used to illustrate the embodiments and are not to be considered as limiting the present invention. In addition, the same reference symbols are used to represent the same components throughout the accompanying drawings. In the accompanying drawings:

[0049] Figure 1 A schematic diagram of a flow chart of an embodiment of a scene 3D reconstruction method of the present invention;

[0050] Figure 2 It is a schematic diagram of the fusion of two point cloud data;

[0051] Figure 3 To make a detailed indoor model diagram based on images;

[0052] Figure 4 To refine the schematic diagram of the indoor and outdoor integrated model effect;

[0053] Figure 5A schematic diagram of the structure of an embodiment of a scene 3D reconstruction system of the present invention;

[0054] Figure 6 The figure is a schematic structural diagram of an embodiment of an electronic device of the present invention. DETAILED DESCRIPTION

[0055] Exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein.

[0056] Figure 1 FIG. 1 is a flow chart showing an embodiment of a method for three-dimensional reconstruction of a scene provided by the present invention, wherein the method is executed by a controller. Figure 1 As shown, the following steps are included:

[0057] S1. Using multiple image sensors, collect image data of multiple image types of the target area, and perform calculations on all image data to obtain aerial triangulation results. In S1:

[0058] 1) Each image sensor collects image data of one image type.

[0059] 2) Multiple image sensors refer to different types of data acquisition equipment used when performing 3D reconstruction tasks. The above different types of data acquisition equipment can capture various image data of the target area, including but not limited to: photography equipment carried by drones, ground close-range photography equipment, and 3D laser scanners. Multiple image sensors can capture image information of the target area from different angles, distances, and accuracies, providing a rich and diverse data source for subsequent aerial triangulation, model reconstruction, and other steps.

[0060] 3) Image data refers to the image information of the target area captured by the image sensor. Image data includes but is not limited to: aerial photography data, drone image data, ground image data, vehicle image data and laser point cloud data. Image data is the basis of 3D reconstruction and also contains information such as the spatial structure, texture and color of the target area. By processing and analyzing the image data, a 3D model of the target area can be generated.

[0061] 4) The aerial triangulation result is obtained through aerial triangulation technology, using the characteristic points, same-name image pairs, relative orientation, matching connection points, regional network adjustment and other information in the image data, and after calculation and processing. The aerial triangulation result is one of the most important data sources in 3D reconstruction. The aerial triangulation result provides the precise location information of each point in the target area, and provides the necessary geometric constraints for subsequent model reconstruction and texture mapping.

[0062] 5) Image type refers to different types of image data distinguished by different acquisition methods, data accuracy and purposes, including but not limited to: drone images, ground close-up images, laser point cloud images. Different types of image data have different characteristics and advantages. By integrating multiple types of image data, it is possible to make up for the shortcomings of a single type of data and improve the accuracy and effect of 3D reconstruction.

[0063] Specifically, the controller collects image data of various image types of the target area by using various image sensors, and performs calculations on all the image data to obtain aerial triangulation results.

[0064] S2. According to the aerial triangulation results, an irregular triangular mesh TIN model is obtained and constructed using the three-dimensional coordinate point set, and texture mapping is performed based on the irregular triangular mesh TIN model to generate a three-dimensional scene model of the target area. In S2:

[0065] 1) The three-dimensional coordinate point set is obtained by processing and analyzing image data, and is a set of three-dimensional coordinates of each point in the target area. Each point in the target area is obtained through aerial triangulation or other three-dimensional measurement methods. The three-dimensional coordinate point set is the basis for building a three-dimensional model. The three-dimensional coordinate point set provides the precise location information of each point in the target area, so that the model can accurately reflect the spatial structure of the target area.

[0066] 2) The irregular triangular mesh TIN model is a spatial data structure composed of a group of irregular triangles. Each triangle represents a spatial plane. Multiple planes are connected together by sharing edges to form a continuous triangular mesh surface. The irregular triangular mesh TIN model can convert a three-dimensional coordinate point set into a continuous triangular mesh surface in three-dimensional reconstruction, providing a basis for subsequent texture mapping and model construction. At the same time, the irregular triangular mesh TIN model can also be smoothed and optimized to make the model more realistic and accurate.

[0067] 3) Texture mapping is the process of applying a texture image to the surface of a 3D model. The above process combines the 2D texture image with the geometric structure of the 3D model surface to give the model a realistic visual effect. Texture mapping can enhance the realism and realism of the 3D model, so that the model better reflects the actual situation of the target area. Through texture mapping, the model can be given more details and color information, improving the visual quality and performance of the model.

[0068] 4) The 3D scene model is a digital 3D representation obtained by integrating the 3D coordinate point set, irregular triangular mesh TIN model and texture image of the target area. The 3D scene model accurately reflects the spatial structure and texture information of the target area. Through the 3D scene model, the target area can be analyzed and planned in detail, providing accurate spatial information support.

[0069] Specifically, the controller further obtains and constructs an irregular triangular mesh TIN model based on the aerial triangulation result using the three-dimensional coordinate point set, and performs texture mapping based on the irregular triangular mesh TIN model to generate a three-dimensional scene model of the target area.

[0070] The technical solution of this embodiment improves the accuracy of aerial triangulation, enhances the precision of three-dimensional reconstruction, and at the same time improves the realism and detail expression of the model.

[0071] In an optional manner, the step of performing computational processing on all image data to obtain an aerial triangulation result includes:

[0072] Preprocess each image data separately to obtain multiple preprocessed images;

[0073] Extracting multiple feature points from each preprocessed image respectively, matching multiple feature points of any preprocessed image with multiple feature points of a preprocessed image adjacent to or overlapping with the preprocessed image, and obtaining a set of similar feature point pairs corresponding to the preprocessed image;

[0074] Based on the similar feature points, the least squares operation or iterative closest point operation is performed on the set to obtain the relative posture relationship between the multiple pre-processed images, and the initial parameters of the aerial triangulation of the relative posture relationship are determined;

[0075] Based on the initial parameters, the bundle adjustment optimization algorithm is used to optimize the relative posture relationship between multiple preprocessed images, and the optimization parameters corresponding to each preprocessed image are obtained. The corresponding preprocessed images are optimized to obtain the optimized images corresponding to each preprocessed image. The optimized images corresponding to each preprocessed image are used as the corresponding preprocessed images, and the steps of extracting multiple feature points of each preprocessed image are returned until the preset iteration conditions are met, and the optimization parameters corresponding to each preprocessed image are determined as the aerial triangulation results.

[0076] In this embodiment, aerial photography is used to acquire images, and software that supports calling mainstream aerial triangulation encryption engines and supports the integration of multiple image data such as aerial survey photography, drone images, ground images, vehicle-mounted images, and laser point clouds is used to perform high-precision aerial triangulation automated modeling and processing, thereby realizing aerial triangulation processing of indoor and outdoor ground close-range photogrammetry data.

[0077] After obtaining the initial parameters of aerial triangulation, collaborative constraints are performed based on POS data, camera files, image control measurement data, and image data, and the bundle method regional network is used for calculation (bundle method adjustment optimization algorithm) to optimize the initial parameters and obtain the optimized parameters that meet the preset iteration conditions, that is, the aerial triangulation results.

[0078] Furthermore, for feature point extraction and matching methods, a pre-trained deep convolutional neural network (such as VGG or ResNet) is introduced to extract high-level semantic features of the image, and at the same time, the traditional SIFT features are combined to improve the robustness of feature matching. The feature fusion expression is as follows:

[0079]

[0080] Among them, SIFT is the SIFT feature, fi CNN is the CNN feature, and α is the weight coefficient.

[0081] Furthermore, each image data is preprocessed to obtain multiple preprocessed images. By constructing image pyramids of different scales, Structure from Motion (SfM) and Multi-ViewStereo (MVS) processing are performed on each scale to extract multiple feature points of each preprocessed image. The multiple feature points of any preprocessed image are matched with the multiple feature points of the preprocessed image adjacent to or overlapping with the preprocessed image to obtain a set of similar feature point pairs corresponding to the preprocessed image.

[0082] For multi-scale fusion, we construct image pyramids of different scales, perform SfM and MVS processing respectively, and then fuse the reconstruction results of different scales. The fusion strategy is as follows:

[0083] P fused =βP coarse +(1-β)P fine

[0084] Among them, P coarse is the coarse-scale reconstruction result, P fine is the fine-scale reconstruction result, and β is the fusion weight.

[0085] By constructing image pyramids of different scales for SfM and MVS processing respectively, and fusing the reconstruction results of different scales, the accuracy, completeness and detail retention of 3D reconstruction can be improved.

[0086] It should be noted that after obtaining the aerial triangulation results, accuracy testing can also be performed. When the specified accuracy is not met, the accuracy can be improved by checking control points, adding control points, and reasonably dividing the blocks to ensure the accuracy of the aerial triangulation results. At the same time, a secondary aerial triangulation can be performed on the target area to further improve the accuracy of the aerial triangulation results and obtain more accurate aerial triangulation results of the target area.

[0087] Specifically, the controller further preprocesses each image data to obtain multiple images, extracts feature points of each image, and matches feature points of adjacent or overlapping images to form a set of similar feature point pairs. Based on the set of similar feature point pairs, the initial parameters of aerial triangulation are obtained by least squares method or iterative nearest point operation. The bundle adjustment optimization algorithm is used to optimize these parameters and images, and feature point extraction, matching, and optimization are iteratively performed until the preset conditions are met, and the aerial triangulation results are finally determined.

[0088] In an optional manner, preprocessing includes: orientation processing and image processing; orientation processing includes: relative orientation and absolute orientation; image processing includes: at least one of denoising, enhancement, correction, filtering, alignment and redundancy removal; operation processing includes: at least one of least squares method and iterative closest point algorithm.

[0089] It should be noted that relative orientation is to determine the initial posture and position of the image, providing a basis for subsequent processing. Relative orientation includes but is not limited to the initial rotation and translation of the image to align the feature points in the image with the preset reference point or reference coordinate system. Absolute orientation is to convert the relative orientation result of the image into an absolute coordinate system, that is, the earth coordinate system or the coordinate system specified by the project. Absolute orientation includes but is not limited to further rotation and translation of the image to eliminate errors in relative orientation.

[0090] Denoising refers to removing noise from image data, such as sensor noise, illumination changes, etc. Enhancement refers to improving the contrast and brightness of image data to make feature points more obvious. Correction refers to geometric correction of image data to eliminate geometric errors such as lens distortion. Filtering refers to smoothing image data to remove high-frequency noise while retaining edge and texture information. Registration refers to registering image data from different times, different perspectives, or different sensors for subsequent processing. De-redundancy refers to removing repeated or redundant information from image data to improve processing efficiency.

[0091] In this embodiment, relative orientation is performed by using POS data (i.e., the position and orientation information of the drone) and camera files in combination with feature points in the image data. The POS data provides the approximate position and orientation of the image, while feature point matching is used for fine-tuning.

[0092] Absolute orientation is achieved by using ground control points (GCPs) or known geographic location information combined with feature points in the image data. GCPs are known points on the ground whose positions and coordinates have been accurately measured and used to calibrate the geographic location of the image.

[0093] By constructing the error equation, the least square method is used to solve the posture parameters of the image data and the coordinates of the three-dimensional points. Or through the iterative closest point algorithm (ICP), that is, a set of initial points are selected as corresponding points, and the transformation (rotation and translation) that minimizes the average distance between corresponding points is calculated, and the transformation is applied to the entire point cloud data until the preset condition is met (that is, the change in the average distance is less than a certain threshold).

[0094] Furthermore, for the relative orientation and feature point matching optimization combined with POS data, first, the drone position and direction information provided by the POS data is combined with the feature point matching in the image data to optimize the relative orientation of the image. The POS data includes the position P of the drone at each shooting time. i and attitude quaternion q i .

[0095] When the image I i and Image I j The feature point X in k When already matched, the following relative orientation model is constructed:

[0096]

[0097] Among them, T i =[R i |ti] represents the camera pose provided by POS data, including the rotation matrix R i and the translation vector t i , and R i Can be achieved through quaternion q i Convert to:

[0098]

[0099] in, Yes i The conjugate of Represents quaternion multiplication. Display image I i and I j Feature point X k The reprojection error.

[0100] In order to introduce the confidence of feature point matching, the weighted least squares method is used to optimize the following objective function:

[0101]

[0102] in, is the feature point matching X k The weight of can be calculated based on the descriptor distance or reprojection error of feature matching.

[0103] Furthermore, for absolute orientation in conjunction with GCPs, ground control points (GCPs) provide known geographic location information for absolute orientation of image data. Assume that the known three-dimensional coordinates of GCPs are G m , the observation value in the image data is The following absolute orientation error equation is constructed:

[0104]

[0105] Among them, Ti = [Ri|ti] represents image I i The camera pose.

[0106] Combining the error equations of relative orientation and absolute orientation, the comprehensive objective function is obtained:

[0107]

[0108] in, is the weight of GCPs and can be set according to the measurement accuracy of GCPs.

[0109] Furthermore, for the iterative closest point algorithm, in order to minimize the average distance between corresponding points, the following objective function is optimized:

[0110]

[0111] Among them, p k is the corresponding point in the original point cloud data, q k is the corresponding point in the target point cloud data.

[0112] The minimization problem can be solved by using the singular value decomposition (SVD) method. First, calculate the centroid of the original point cloud data and the target point cloud data:

[0113]

[0114] Where K is the number of corresponding points. centroid Represents the centroid of the original point cloud data. centroid Represents the centroid of the target point cloud data. Then, construct the covariance matrix:

[0115]

[0116] Perform singular value decomposition on the covariance matrix H:

[0117] H=USVT

[0118] Where U is an orthogonal matrix, and its column vectors are called left singular vectors of H. If the dimension of H is m×n, then the dimension of U is m×m, and the column vector of U is HH T The eigenvectors of. S is a diagonal matrix. The elements on the diagonal are the singular values ​​of H. The singular values ​​are non-negative and are arranged on the diagonal of S in descending order. If the dimension of H is m×n, then the dimension of S is m×n. V is an orthogonal matrix. Its column vectors are called the right singular vectors of H. The dimension of V is n×n. The column vectors of V are H T The eigenvector of H.

[0119] The rotation matrix R is calculated by the following formula:

[0120] R=VU T

[0121] Finally, the translation vector t is calculated by the following formula:

[0122] t=q centroid -Rp centroid

[0123] In each iteration, the obtained R and t are applied to the original point cloud data and the corresponding points are recalculated until the change in the average distance is less than the preset threshold.

[0124] Furthermore, by combining the above relative orientation, absolute orientation and ICP algorithm, a comprehensive optimization framework can be constructed to accurately solve the posture parameters of image data and the coordinates of three-dimensional points. The comprehensive objective function is as follows:

[0125]

[0126] Here, λ is a regularization parameter that weighs the relative orientation, absolute orientation, and ICP errors.

[0127] By minimizing the above comprehensive objective function, the optimal image posture parameters and 3D point coordinates can be obtained. By combining the advantages of POS data, feature point matching, GCPs and ICP algorithms, the orientation accuracy of image data and the accuracy of 3D reconstruction can be effectively improved.

[0128] Specifically, the controller further performs orientation processing and image processing on the image data, extracts and matches feature points, calculates initial parameters using the least squares method or iterative closest point algorithm, and iteratively processes after bundle adjustment optimization until preset conditions are met to obtain aerial triangulation results.

[0129] In an optional manner, the step of obtaining and constructing an irregular triangulated mesh TIN model using a three-dimensional coordinate point set according to the aerial triangulation result includes:

[0130] Based on the aerial triangulation results, a three-dimensional coordinate point set containing the terrain location information and the location information of the objects in the target area is extracted, and the three-dimensional coordinate point set is screened to remove noise points and redundant points to obtain a screened three-dimensional coordinate point set;

[0131] The Delaunay triangulation algorithm is used to connect the screened three-dimensional coordinate point set into a target triangular mesh to form an irregular triangular mesh TIN model; wherein the target triangular mesh is a continuous and non-overlapping triangular mesh.

[0132] It should be noted that aerial triangulation is based on the relative position and posture relationship between image data. In aerial triangulation, the bundle method regional network is used for adjustment optimization. By continuously adjusting the relative position and posture parameters between image data, the feature points on all image data can be accurately projected into three-dimensional space, thereby obtaining an accurate three-dimensional coordinate point set.

[0133] The Delaunay triangulation algorithm is a triangulation method based on the maximum and minimum angle criteria. The Delaunay triangulation algorithm can ensure that each triangle in the generated triangular mesh is as close to an equilateral triangle as possible, thereby avoiding the appearance of long and thin triangles. The advantage of the above algorithm is that it can generate high-quality triangular meshes and improve the accuracy and stability of the model.

[0134] After the screening, the three-dimensional coordinate point set is connected into the target triangle mesh using the Delaunay triangulation algorithm to form an irregular triangle mesh TIN model. In the process of constructing the irregular triangle mesh TIN model, it is necessary not only to ensure that the target triangle mesh is continuous and non-overlapping, but also to combine the aerial triangulation results and image data for further optimization.

[0135] In this embodiment, first, the high-precision information in the aerial triangulation results is used to accurately orient and align the screened three-dimensional coordinate point set, thereby ensuring the accurate position of each point in the actual space. Then, the appropriate image matching unit is analyzed and selected, feature matching and dense matching are performed, and dense point cloud data is generated. The point cloud data not only contains the information of the original three-dimensional coordinate point set, but also obtains more details and texture information through image matching. Then, three-dimensional modeling is performed based on the generated dense point cloud data. The Delaunay triangulation algorithm automatically selects the best image pair model at different viewing angles, generates a dense point cloud of three-dimensional scale, and automatically converts it into an irregular triangular mesh TIN model. The Delaunay triangulation algorithm ensures that the generated triangular mesh is both continuous and non-overlapping, and the shape of each triangle is relatively regular.

[0136] Furthermore, in order to achieve accurate 3D scene reconstruction, process large-scale image data, enhance robustness, and support multi-sensor fusion, camera pose estimation can also be performed based on SfM. First, the SIFT (Scale-Invariant Feature Transform) algorithm is used to extract features from each image. The SIFT algorithm can maintain good invariance under scale and rotation changes, and the expression is as follows:

[0137] For an image I, in the scale space L(x, y, σ), the key point detection is achieved by finding the extreme point in the scale space:

[0138]

[0139] By generating feature descriptors, a 128-dimensional vector of each feature point can be obtained for subsequent feature matching.

[0140] Then, the RANSAC (Random Sample Consensus) algorithm is used to optimize the initial matching and obtain the inlier matching set. Based on the above matching points, the camera pose is estimated by minimizing the reprojection error:

[0141]

[0142] Among them, x i is the image coordinate, X i is the world coordinate, π is the camera projection function, R is the rotation matrix, and t is the translation vector.

[0143] Furthermore, dense point clouds can be generated based on MVS. First, the cost volume C(x, y, d) is constructed using the principle of multi-view geometry, where d represents the disparity:

[0144]

[0145] Where V is the number of views, W(x, y) is the window function, and SAD is the Sum of Absolute Differences.

[0146] Next, the Winner-Takes-All (WTA) strategy is used to perform cost aggregation and disparity estimation to estimate the optimal disparity:

[0147]

[0148] Then, a dense point cloud can be calculated by triangulation. Therefore, dense point cloud generation based on MVS (Multi-View Stereo) can achieve high-precision and high-density 3D reconstruction by extracting high-detail 3D structures from multiple image data.

[0149] After the irregular triangular mesh TIN model is constructed, it is smoothed and optimized. Based on the intrinsic geometric relationship (the side length, angle, area of ​​the triangle and the adjacent relationship between triangles), the irregular triangular mesh TIN model is adjusted (re-dividing the triangular mesh, adjusting the side length and angle of the triangle, merging or splitting the triangle) to make it more consistent with the ups and downs and changes of the actual terrain. At the same time, according to the spatial position information of the three-dimensional irregular triangular mesh TIN model, the image texture of the best viewing angle is obtained, and the model texture is given, so as to obtain an irregular triangular mesh TIN model with high precision, high stability and rich texture information, so as to achieve a true reflection of the actual scene.

[0150] Specifically, the controller further extracts a set of three-dimensional coordinate points of the target area based on the aerial triangulation results, and after screening and removing noise and redundant points, uses the Delaunay triangulation algorithm to construct a continuous and non-overlapping triangular mesh to form an irregular triangular mesh TIN model.

[0151] In an optional manner, the step of performing texture mapping based on the irregular triangular mesh TIN model to generate a three-dimensional scene model of the target area includes:

[0152] According to the spatial position information and geometric characteristics of the irregular triangular mesh TIN model, the texture information corresponding to the surface of the irregular triangular mesh TIN model is extracted from the image data, and the texture information is mapped to the surface of the irregular triangular mesh TIN model using the texture mapping algorithm to generate a three-dimensional scene model of the target area.

[0153] It should be noted that texture mapping is to enrich the surface details of the irregular triangular mesh TIN model by using the texture information in the image data. In addition, by analyzing the geometric features of the irregular triangular mesh TIN model, the part of the image data corresponding to the surface of the irregular triangular mesh TIN model can be determined, thereby extracting the corresponding texture information.

[0154] Texture mapping algorithm maps the texture information in two-dimensional image data onto the surface of three-dimensional model. When applying texture mapping algorithm, it is necessary to consider the geometric features of the model, the characteristics of texture information and the applicability of mapping algorithm to ensure the accuracy and authenticity of the mapping results.

[0155] In this embodiment, professional modeling software is used to load the constructed irregular triangular mesh TIN model and the corresponding image data. The image data includes multiple data sources such as drone oblique photography, ground close-range photogrammetry, and three-dimensional laser scanning. The geometric analysis function of the modeling software is used to analyze the surface morphology, slope, slope aspect and other geometric features of the irregular triangular mesh TIN model. The above-mentioned geometric features are used to determine the optimal viewing angle and projection method of the image data during texture mapping. According to the spatial position information and geometric features of the irregular triangular mesh TIN model, the corresponding texture information is extracted from the image data. The extraction process involves operations such as cropping, rotation, and scaling of the image data to ensure that the texture information matches the geometric features of the surface of the irregular triangular mesh TIN model.

[0156] According to the complexity of the irregular triangular mesh TIN model and the characteristics of the texture information, select the appropriate texture mapping algorithm (projection mapping or parametric mapping). Establish a texture coordinate system on the irregular triangular mesh TIN model to determine the position and direction of the texture information on the model surface. The texture coordinate system is associated with the geometric coordinate system of the model to ensure the accuracy of texture mapping. Using the selected texture mapping algorithm, the extracted texture information is mapped onto the surface of the irregular triangular mesh TIN model. The mapping process involves operations such as texture splicing, adjustment, and optimization to ensure the continuity and authenticity of the texture on the model surface. Check and adjust the texture mapping results to ensure that the texture matches the geometric features of the model surface and there is no obvious distortion, deformation, or mapping error.

[0157] Check the irregular triangular mesh TIN model after texture mapping to ensure that the texture information is correctly and realistically mapped onto the model surface. In the modeling software, integrate and optimize the irregular triangular mesh TIN model after texture mapping and other data sources (laser point cloud data, ground measurement data, etc.) to generate the final 3D scene model, and perform visual display and quality inspection of the results.

[0158] Specifically, the controller further extracts texture information from the image data, maps the texture information to the surface of the irregular triangular mesh TIN model using a texture mapping algorithm, and generates a three-dimensional scene model of the target area.

[0159] It should be noted that by loading the dense matching point cloud and the laser point cloud in the modeling software one after another, seamless fusion can be achieved. The modeling software will automatically build the TIN triangulation network, automatically perform texture slicing and mapping, and finally generate a real-life 3D model. At the same time, after the modeling software completes the aerial triangulation of the oblique image, it imports the laser point cloud data and trajectory files obtained by the mobile measurement data processing, which enables the two point cloud data to be spliced ​​and fused under the same spatial coordinates. After fusion, they are submitted to the modeling software for automatic modeling. Figure 2 As shown in the figure, the overall effect of the irregular triangular mesh TIN model can be significantly improved by fusion modeling of the above two types of point cloud data, such as under the eaves, the bottom of trees, etc. In addition, by encrypting the laser point cloud data, the TIN triangulation network can be improved and the real structure of the main body of the object can be restored.

[0160] In addition, the modeling software can also produce individual naked-eye 3D models. It is a software that can perform fast and accurate 3D modeling based on multiple image data. Figure 3 The modeling software shown can directly outline interactive modeling against the original image, breaking through the traditional stereo image pair operation method. It does not require wearing stereo glasses and automatically retrieves multi-angle images. It can quickly push and pull structures on any surface.

[0161] In visual display and quality inspection of results, such as Figure 4 As shown, the indoor and outdoor integrated fine real-life 3D visualization results integrate the constructed fine monomer model, the refined real-life 3D scene and other spatial data to form an indoor and outdoor integrated 3D scene for display. In addition, the real-life 3D mesh model and other data can be checked by human-computer interaction using the real-life 3D quality inspection algorithm, which can realize the automatic collection of detection point data, automatic mathematical accuracy detection, and automatic accuracy report export, etc., to improve the efficiency and automation of inspection, improve the efficiency and quality of the quality inspection work of the results, and better ensure the implementation of quality inspection tasks. The human-computer interaction inspection includes the following contents:

[0162] 1. Overall inspection of model data:

[0163] (1) The integrity of the model data, including whether there are missing blocks, missing structures, missing textures, etc.;

[0164] (2) Whether the model blocks are misaligned or delaminated;

[0165] (3) Check whether the block size and block center point of the model meet the requirements;

[0166] (4) Whether there are large abnormal insertions or gaps at the junction of the fine individual reconstruction model and the scene model;

[0167] (5) Check whether the center point of the model is consistent with the regulations.

[0168] 2. Model texture check:

[0169] (1) Check whether the model texture is obviously distorted or deformed;

[0170] (2) Whether there are large areas of texture missing or mapping errors.

[0171] 3. Model structure check:

[0172] (1) Check whether the model structure proportions are consistent with the real scene;

[0173] (2) Whether the spatial position of the model is consistent with the real-life 3D model;

[0174] (3) Model accuracy should meet relevant specifications, standards and contract requirements.

[0175] 4. Model color inspection:

[0176] (1) Whether the model elements such as terrain, buildings (structures), transportation ancillary facilities, and water systems are realistic, and whether the model accurately reflects the shape, texture, color, brightness, and light-dark relationship of the modeled objects;

[0177] (2) The overall color and lighting effects of the scene should be coordinated and consistent.

[0178] In this embodiment, the following evaluation indicators are used to quantitatively evaluate the reconstruction effect:

[0179] Reprojection error: measures the accuracy of the camera pose estimate.

[0180] Completeness: The degree of match between the reconstructed point cloud and the real point cloud.

[0181] Accuracy: The deviation between the reconstructed point cloud and the true point cloud.

[0182] The calculation formula of the reprojection error is as follows:

[0183]

[0184] Where n is the number of matching points, x i is the image coordinate, X i is the world coordinate, π is the camera projection function, R is the rotation matrix, and t is the translation vector.

[0185] The calculation formula for completeness is as follows:

[0186]

[0187] The accuracy is calculated as follows:

[0188]

[0189] Among them, P reconstructed is the reconstructed point cloud, P groundtruth It is a real point cloud.

[0190] Through evaluation, it can be known that the traditional SfM+MVS method has a large reconstruction error in detail parts (such as object edges); the deep learning enhancement method using only SIFT features has improved detail reconstruction, but there is still some blur and missing. The present invention combines the deep learning enhancement of SfM, MVS, SIFT features and the multi-scale image pyramid fusion method to achieve better performance in detail reconstruction and overall integrity, and can clearly reconstruct the edge details and texture details of the object.

[0191] Figure 5 FIG. 2 is a schematic diagram showing a structure of an embodiment of a scene 3D reconstruction system 200 provided by the present invention. Figure 5 As shown, the system 200 includes: an acquisition module 210 and an execution module 220 .

[0192] The acquisition module 210 is used to collect image data of various image types of the target area using various image sensors, and perform calculations on all the image data to obtain aerial triangulation results; wherein each image sensor collects image data of one image type respectively.

[0193] The execution module 220 is used to obtain and construct an irregular triangular mesh TIN model using a three-dimensional coordinate point set according to the aerial triangulation result, and perform texture mapping based on the irregular triangular mesh TIN model to generate a three-dimensional scene model of the target area.

[0194] In an optional manner, the acquisition module 210 is specifically used to:

[0195] Preprocess each image data separately to obtain multiple preprocessed images;

[0196] Extracting multiple feature points from each preprocessed image respectively, matching multiple feature points of any preprocessed image with multiple feature points of a preprocessed image adjacent to or overlapping with the preprocessed image, and obtaining a set of similar feature point pairs corresponding to the preprocessed image;

[0197] Based on the similar feature points, the least squares operation or iterative closest point operation is performed on the set to obtain the relative posture relationship between the multiple pre-processed images, and the initial parameters of the aerial triangulation of the relative posture relationship are determined;

[0198] Based on the initial parameters, the bundle adjustment optimization algorithm is used to optimize the relative posture relationship between multiple preprocessed images, and the optimization parameters corresponding to each preprocessed image are obtained. The corresponding preprocessed images are optimized to obtain the optimized images corresponding to each preprocessed image. The optimized images corresponding to each preprocessed image are used as the corresponding preprocessed images, and the steps of extracting multiple feature points of each preprocessed image are returned until the preset iteration conditions are met, and the optimization parameters corresponding to each preprocessed image are determined as the aerial triangulation results.

[0199] In an optional manner, preprocessing includes: orientation processing and image processing; orientation processing includes: relative orientation and absolute orientation; image processing includes: at least one of denoising, enhancement, correction, filtering, alignment and redundancy removal; operation processing includes: at least one of least squares method and iterative closest point algorithm.

[0200] The technical solution of this embodiment improves the accuracy of aerial triangulation, enhances the precision of three-dimensional reconstruction, and at the same time improves the realism and detail expression of the model.

[0201] The above parameters and steps for each module to implement corresponding functions in the scene 3D reconstruction system 200 of this embodiment can refer to the parameters and steps in the above embodiment of the scene 3D reconstruction method, and will not be repeated here.

[0202] like Figure 6 As shown, an electronic device 300 according to an embodiment of the present invention includes a processor 320, the processor 320 is coupled to a memory 310, and the memory 310 stores at least one computer program 330, and the at least one computer program 330 is loaded and executed by the processor 320, so that the electronic device 300 implements any of the above-mentioned scene 3D reconstruction methods, specifically:

[0203] The electronic device 300 may have relatively large differences due to different configurations or performances, and may include one or more processors 320 (Central Processing Units, CPU) and one or more memories 310, wherein the one or more memories 310 store at least one computer program 330, and the at least one computer program 330 is loaded and executed by the one or more processors 320, so that the electronic device 300 implements any of the scene 3D reconstruction methods provided in the above embodiments. Of course, the electronic device 300 may also have components such as a wired or wireless network interface, a keyboard, and an input and output interface for input and output, and the electronic device 300 may also include other components for implementing device functions, which will not be described in detail here.

[0204] A computer-readable storage medium according to an embodiment of the present invention stores at least one computer program, and the at least one computer program is loaded and executed by a processor so that a computer implements any of the above-mentioned scene three-dimensional reconstruction methods.

[0205] Optionally, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc (CD-ROM), a magnetic tape, a floppy disk, an optical data storage device, and the like.

[0206] In an exemplary embodiment, a computer program product or a computer program is also provided, the computer program product or the computer program comprising computer instructions, the computer instructions being stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the electronic device performs any of the above-mentioned scene 3D reconstruction methods.

[0207] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects and represent the definition of a specific order or sequence. The order of use of similar objects can be interchanged where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than the order shown or described.

[0208] Those skilled in the art know that the present invention can be implemented as a system, method or computer program product. Therefore, the present disclosure can be specifically implemented in the following forms, namely: it can be complete hardware, it can be complete software (including firmware, resident software, microcode, etc.), or it can be a combination of hardware and software, which is generally referred to as "circuit", "module" or "system" herein. In addition, in some embodiments, the present invention can also be implemented in the form of a computer program product in one or more computer-readable media, and the computer-readable medium contains computer-readable program code.

[0209] Any combination of one or more computer-readable media may be used. A computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, device, or device.

[0210] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A method for three-dimensional reconstruction of a scene, characterized in that: include: Acquire image data of multiple image types of the target area collected by multiple image sensors, and perform calculations on all the image data to obtain aerial triangulation results; wherein each image sensor collects image data of one image type respectively; According to the aerial triangulation result, an irregular triangular mesh TIN model is obtained and constructed using a three-dimensional coordinate point set, and texture mapping is performed based on the irregular triangular mesh TIN model to generate a three-dimensional scene model of the target area.

2. The scene 3D reconstruction method according to claim 1, characterized in that: The steps of computing and processing all image data to obtain aerial triangulation results include: Preprocess each image data separately to obtain multiple preprocessed images; Extracting multiple feature points from each preprocessed image respectively, matching multiple feature points of any preprocessed image with multiple feature points of a preprocessed image adjacent to or overlapping with the preprocessed image, and obtaining a set of similar feature point pairs corresponding to the preprocessed image; Based on the set of similar feature points, a least squares operation or an iterative closest point operation is performed to obtain a relative position and posture relationship between a plurality of pre-processed images, and initial parameters of aerial triangulation of the relative position and posture relationship are determined; Based on the initial parameters, the relative position and posture relationship between the multiple preprocessed images is optimized using the bundle adjustment optimization algorithm to obtain the optimization parameters corresponding to each preprocessed image, and the corresponding preprocessed images are optimized to obtain the optimized images corresponding to each preprocessed image. The optimized images corresponding to each preprocessed image are respectively used as the corresponding preprocessed images, and the step of respectively extracting multiple feature points of each preprocessed image is returned to execute until the preset iteration conditions are met, and the optimization parameters corresponding to each preprocessed image are determined as the aerial triangulation results.

3. The scene 3D reconstruction method according to claim 2, characterized in that: The preprocessing includes: orientation processing and image processing; the orientation processing includes: relative orientation and absolute orientation; the image processing includes: at least one of denoising, enhancement, correction, filtering, registration and redundancy removal; the operation processing includes: at least one of the least squares method and iterative closest point algorithm.

4. The scene 3D reconstruction method according to claim 1, characterized in that: According to the aerial triangulation results, an irregular triangulated mesh TIN model is obtained and constructed using a three-dimensional coordinate point set, including: Based on the aerial triangulation result, extract the three-dimensional coordinate point set including the terrain position information and the object position information of the target area, and filter the three-dimensional coordinate point set to remove noise points and redundant points to obtain a filtered three-dimensional coordinate point set; The filtered three-dimensional coordinate point set is connected into a target triangular mesh by using a Delaunay triangulation algorithm to form the irregular triangular mesh TIN model; wherein the target triangular mesh is a continuous and non-overlapping triangular mesh.

5. The scene 3D reconstruction method according to claim 1, characterized in that: The step of performing texture mapping based on the irregular triangular mesh TIN model to generate a three-dimensional scene model of the target area includes: According to the spatial position information and geometric features of the irregular triangular mesh TIN model, texture information corresponding to the surface of the irregular triangular mesh TIN model in the image data is extracted, and the texture information is mapped onto the surface of the irregular triangular mesh TIN model using a texture mapping algorithm to generate the three-dimensional scene model of the target area.

6. A scene three-dimensional reconstruction system, characterized in that: include: Get module and execute module; The acquisition module is used to: acquire image data of multiple image types of the target area collected by multiple image sensors, and perform calculation processing on all the image data to obtain aerial triangulation results; wherein each image sensor collects image data of one image type respectively; The execution module is used to obtain and construct an irregular triangular mesh TIN model using a three-dimensional coordinate point set according to the aerial triangulation result, and perform texture mapping based on the irregular triangular mesh TIN model to generate a three-dimensional scene model of the target area.

7. The scene 3D reconstruction system according to claim 6, characterized in that: The acquisition module is specifically used for: Preprocess each image data separately to obtain multiple preprocessed images; Extracting multiple feature points from each preprocessed image respectively, matching multiple feature points of any preprocessed image with multiple feature points of a preprocessed image adjacent to or overlapping with the preprocessed image, and obtaining a set of similar feature point pairs corresponding to the preprocessed image; Based on the set of similar feature points, a least squares operation or an iterative closest point operation is performed to obtain a relative position and posture relationship between a plurality of pre-processed images, and initial parameters of aerial triangulation of the relative position and posture relationship are determined; Based on the initial parameters, the relative position and posture relationship between the multiple preprocessed images is optimized using the bundle adjustment optimization algorithm to obtain the optimization parameters corresponding to each preprocessed image and optimize the corresponding preprocessed images to obtain the optimized images corresponding to each preprocessed image. The optimized images corresponding to each preprocessed image are respectively used as the corresponding preprocessed images, and the step of respectively extracting multiple feature points of each preprocessed image is returned to execute until the preset iteration conditions are met, and the optimization parameters corresponding to each preprocessed image are determined as the aerial triangulation results.

8. The scene 3D reconstruction system according to claim 7, characterized in that: The preprocessing includes: orientation processing and image processing; the orientation processing includes: relative orientation and absolute orientation; the image processing includes: at least one of denoising, enhancement, correction, filtering, registration and redundancy removal; the operation processing includes: at least one of the least squares method and iterative closest point algorithm.

9. An electronic device, characterized in that: The electronic device includes a processor, the processor is coupled to a memory, the memory stores at least one computer program, and the at least one computer program is loaded and executed by the processor so that the electronic device implements the scene three-dimensional reconstruction method as described in any one of claims 1 to 5.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores at least one computer program, and the at least one computer program is loaded and executed by the processor so that the computer-readable storage medium implements the scene three-dimensional reconstruction method according to any one of claims 1 to 5.

Images