Three-dimensional model construction method and device, storage medium and computer equipment

Abstract

The invention discloses a three-dimensional model construction method and device, a storage medium and computer equipment, and relates to the technical field of financial science and technology and the technical field of digital medical treatment. Comprising the following steps: acquiring a two-dimensional image, and dividing the two-dimensional image into a plurality of panes; predicting a local approximate plane normal in a view angle range corresponding to each pane, determining a three-dimensional modeling starting point, determining a three-dimensional coordinate of the three-dimensional modeling starting point, taking the pane where the three-dimensional modeling starting point is located as a starting point pane, and taking the three-dimensional coordinate of the three-dimensional modeling starting point as a starting point center coordinate of a starting point local approximate plane corresponding to the starting point pane; determining a neighborhood pane adjacent to the starting point pane, and determining a neighborhood center coordinate of a neighborhood local approximate plane corresponding to the neighborhood pane based on the starting point center coordinate, a local approximate plane normal of the starting point pane and a local approximate plane normal of the neighborhood pane; and based on the starting point center coordinate and the neighborhood center coordinate, splicing the starting point local approximate plane and the neighborhood local approximate plane to obtain a three-dimensional model.

Description

Methods, apparatus, storage media and computer equipment for constructing 3D models Technical Field

[0001] This invention relates to the fields of financial technology and digital healthcare technology, and in particular to a method, apparatus, storage medium, and computer equipment for constructing a three-dimensional model. Background Technology

[0002] Two-dimensional images obtained through medical imaging equipment such as CT and MRI can be used to reconstruct three-dimensional models of organs, blood vessels, or tumors. These three-dimensional models help doctors understand the location, shape, and size of lesions more intuitively, thus enabling them to develop more precise surgical plans. In scenarios such as vehicle mortgages and leasing, converting two-dimensional images into three-dimensional images can enhance information transparency.

[0003] Currently, 3D point clouds are typically obtained through image depth estimation before 3D models are constructed. However, the accuracy of image depth estimation directly affects the precision of the 3D point cloud, resulting in lower accuracy of the constructed 3D model. Furthermore, point cloud data usually only contains spatial point location information and lacks detailed information such as texture, which also affects the quality of the constructed 3D model. Summary of the Invention

[0004] This invention provides a method, apparatus, storage medium, and computer equipment for constructing three-dimensional models, which mainly improves the accuracy and quality of three-dimensional model construction.

[0005] According to a first aspect of the present invention, a method for constructing a three-dimensional model is provided, comprising: in response to a construction signal for a three-dimensional model to be constructed, acquiring a two-dimensional image corresponding to the three-dimensional model to be constructed, and dividing the two-dimensional image into multiple panes; predicting the local approximate plane normal within the viewpoint range corresponding to each pane in the three-dimensional model to be constructed, determining a three-dimensional modeling starting point in the two-dimensional image, and determining the three-dimensional coordinates of the three-dimensional modeling starting point, taking the pane containing the three-dimensional modeling starting point as the starting pane, and taking the three-dimensional coordinates of the three-dimensional modeling starting point as the starting center coordinates of the starting local approximate plane within the viewpoint range of the starting pane; determining neighboring panes adjacent to the starting pane, and based on the starting center coordinates... The local approximate plane normals of the starting pane and the local approximate plane normals of the neighboring panes are used to determine the neighborhood center coordinates of the neighborhood local approximate planes within the viewing range of the neighboring panes. Based on the starting center coordinates and the neighborhood center coordinates, the starting local approximate planes and the neighborhood local approximate planes are stitched together. After stitching, the neighboring pane is used as the new starting pane, and new neighboring panes adjacent to the new starting pane are determined. The stitching of the starting local approximate planes within the viewing range of the new starting panes and the neighborhood local approximate planes within the viewing range of the new neighboring panes continues until the local approximate planes within the viewing range of each pane are stitched together, thus obtaining the three-dimensional model corresponding to the two-dimensional image.

[0006] Optionally, predicting the local approximate plane normal within the viewpoint range corresponding to each pane in the 3D model to be constructed includes: obtaining a preset normal prediction model, wherein the preset normal prediction model includes a CNN layer for extracting image features, a perspective transformation layer for image transformation, and a prediction layer for normal prediction; inputting the local approximate plane within the viewpoint range corresponding to each pane in the 3D model to be constructed into the preset normal prediction model; extracting image features of the local approximate plane through the CNN layer; performing perspective transformation on the image features through the perspective transformation matrix in the perspective transformation layer; and performing normal prediction on the perspective-transformed image features through the prediction layer to obtain the local approximate plane normal within the viewpoint range corresponding to each pane in the 3D model to be constructed.

[0007] Optionally, determining the neighborhood center coordinates of the neighborhood local approximation plane within the viewing range of the neighborhood pane, based on the starting point center coordinates, the local approximation plane normal of the starting point pane, and the local approximation plane normal of the neighborhood pane, includes: determining the sphere attribute information of the virtual sphere corresponding to the 3D modeling starting point based on the local approximation plane normal of the starting point pane and the local approximation plane normal of the neighborhood pane; and determining the neighborhood center coordinates of the neighborhood local approximation plane within the viewing range of the neighborhood pane, based on the starting point center coordinates of the starting point local approximation plane, the sphere attribute information, the local approximation plane normal of the starting point pane, and the local approximation plane normal of the neighborhood pane.

[0008] Optionally, based on the local approximate plane normal of the starting pane and the local approximate plane normal of the neighboring pane, the sphere attribute information of the virtual sphere corresponding to the starting point of the 3D modeling is determined, including: moving the local approximate plane normal of the starting pane and the local approximate plane normal of the neighboring pane in opposite directions by a preset distance respectively, determining the intersection point of the translated local approximate plane normal and the translated local approximate plane normal, and taking the intersection point as the center of the virtual sphere; determining the starting projection point of the intersection point between the local approximate plane normal of the starting pane and the starting local approximate plane in the 2D image, determining the neighboring projection point of the intersection point between the local approximate plane normal of the neighboring pane and the neighboring local approximate plane in the 2D image, and determining the projection distance between the starting projection point and the neighboring projection point in the 2D image, and taking the projection distance as the arc length of the virtual sphere; determining the angle between the local approximate plane normal of the starting pane and the local approximate plane normal of the neighboring pane, and taking the ratio between the arc length and the angle as the sphere radius of the virtual sphere.

[0009] Optionally, determining the neighborhood center coordinates of the neighborhood local approximation plane within the viewing range of the neighborhood pane, based on the starting center coordinates of the starting local approximation plane, the sphere attribute information, the local approximation plane normal of the starting pane, and the local approximation plane normal of the neighborhood pane, includes: determining the starting direction vector of the local approximation plane normal of the starting pane based on the sphere center and the starting center coordinates, and determining the neighborhood direction vector of the local approximation plane normal of the neighborhood pane based on the sphere center and the starting direction vector; moving the sphere radius along the neighborhood direction vector from the sphere center to the target point, and using the coordinates of the target point as the neighborhood center coordinates of the neighborhood local approximation plane within the viewing range of the neighborhood pane.

[0010] Optionally, based on the starting point center coordinates and the neighborhood center coordinates, the local approximation plane of the starting point and the local approximation plane of the neighborhood are spliced ​​together, including: determining a translation vector between the local approximation plane of the neighborhood and the local approximation plane of the starting point based on the starting point center coordinates and the neighborhood center coordinates; determining the normal angle and rotation axis between the normal of the local approximation plane of the starting point pane and the normal of the local approximation plane of the neighborhood pane, and determining a rotation matrix based on the normal angle and the rotation axis; and splicing the local approximation plane of the neighborhood onto the local approximation plane of the starting point based on the translation vector and the rotation matrix.

[0011] Optionally, stitching together the starting local approximation plane within the new starting pane's view range and the neighboring local approximation plane within the new neighboring pane's view range includes: using the three-dimensional coordinates of the center pixel of the new starting pane as the new starting center coordinates; determining the new neighboring center coordinates of the neighboring local approximation plane within the new neighboring pane's view range based on the new starting center coordinates, the local approximation plane normal of the new starting pane, and the local approximation plane normal of the new neighboring pane; and stitching together the starting local approximation plane within the new starting pane's view range and the neighboring local approximation plane within the new neighboring pane's view range based on the new starting center coordinates and the new neighboring center coordinates.

[0012] According to a second aspect of the present invention, a three-dimensional model construction apparatus is provided, comprising: an acquisition unit, configured to acquire a two-dimensional image corresponding to the three-dimensional model to be constructed in response to a construction signal of the three-dimensional model to be constructed, and divide the two-dimensional image into multiple panes; a prediction unit, configured to predict the local approximate plane normal within the viewpoint range corresponding to each pane in the three-dimensional model to be constructed, determine a three-dimensional modeling starting point in the two-dimensional image, and determine the three-dimensional coordinates of the three-dimensional modeling starting point, taking the pane containing the three-dimensional modeling starting point as the starting point pane, and taking the three-dimensional coordinates of the three-dimensional modeling starting point as the starting center coordinates of the starting local approximate plane within the viewpoint range of the starting point pane; and a determination unit, configured to determine neighboring panes adjacent to the starting point pane, based on the... The system uses the starting point center coordinates, the local approximate plane normal of the starting point pane, and the local approximate plane normal of the neighboring pane to determine the neighborhood center coordinates of the neighboring local approximate plane within the viewing range of the neighboring pane. A stitching unit is used to stitch the starting point local approximate plane and the neighboring local approximate plane based on the starting point center coordinates and the neighborhood center coordinates. After stitching, the neighboring pane is used as the new starting point pane, and a new neighboring pane adjacent to the new starting point pane is determined. The system continues to stitch the starting point local approximate plane within the viewing range of the new starting point pane and the neighboring local approximate plane within the viewing range of the new neighboring pane until all local approximate planes within the viewing range of each pane are stitched together, thus obtaining the three-dimensional model corresponding to the two-dimensional image.

[0013] According to a third aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method for constructing the above-described three-dimensional model.

[0014] According to a fourth aspect of the present invention, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described method for constructing a three-dimensional model.

[0015] According to the present invention, a method, apparatus, storage medium, and computer device for constructing a three-dimensional model are provided. Compared with the current method of obtaining a three-dimensional point cloud through image depth estimation and then constructing a three-dimensional model, the present invention divides a two-dimensional image into multiple panes and predicts the local approximate plane normal within the viewpoint range of each pane in the three-dimensional model to be constructed. Then, it determines the starting center coordinates of the starting local approximate plane within the viewpoint range of the starting pane, and the neighborhood center coordinates of the neighborhood local approximate plane within the viewpoint range of the neighboring panes adjacent to the starting pane. Finally, based on the starting center coordinates and the neighborhood center coordinates, the starting local approximate plane and the neighborhood local approximate plane are stitched together. The neighboring panes are used as new starting panes, and the panes adjacent to the neighboring panes are used as new neighboring panes. The stitching of local approximate planes is continued in the above manner until the local approximate planes corresponding to all panes participate in the stitching, and finally a three-dimensional model is obtained. Therefore, this invention can preserve the texture, color and other details of the image through planar stitching, thereby improving the construction quality of the 3D model. At the same time, the local approximate plane normal and center coordinates are easily and accurately determined. Thus, this invention can improve the construction accuracy of the 3D model by using the local approximate plane normal and center coordinates. Attached Figure Description

[0016] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with the description thereof, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings: Figure 1 shows a flowchart of a method for constructing a three-dimensional model according to an embodiment of the present invention; Figure 2 shows a flowchart of another method for constructing a three-dimensional model according to an embodiment of the present invention; Figure 3 shows a schematic structural diagram of a device for constructing a three-dimensional model according to an embodiment of the present invention; Figure 4 shows a schematic structural diagram of another device for constructing a three-dimensional model according to an embodiment of the present invention; Figure 5 shows a schematic structural diagram of a computer device according to an embodiment of the present invention. Detailed Implementation

[0017] The present invention will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the present application can be combined with each other.

[0018] Currently, the method of obtaining 3D point clouds through image depth estimation and then reconstructing 3D models suffers from the fact that the accuracy of image depth estimation directly affects the accuracy of 3D point clouds, resulting in low accuracy of 3D model construction. At the same time, point cloud data lacks detailed information such as texture, which in turn affects the quality of 3D model construction.

[0019] To address the aforementioned issues, this invention provides a method for constructing a three-dimensional model, as shown in Figure 1. The method includes: 101. In response to a construction signal for a three-dimensional model to be constructed, acquiring a two-dimensional image corresponding to the three-dimensional model to be constructed, and dividing the two-dimensional image into multiple panes.

[0020] Two-dimensional images include, but are not limited to, CT images of patients in the medical field. Specifically, two-dimensional images are obtained from databases or websites, and then divided into multiple panes based on the structural features of objects within the image. For example, if a patient's two-dimensional CT image is obtained from a hospital's medical image archive or communication system, and this image is a head CT image, then the two-dimensional image is divided into panes based on the patient's head structure. For instance, the skull is divided into one pane, the brain tissue into another, and important organs such as the eyes and ears into separate panes. In the vehicle mortgage scenario in the financial field, the two-dimensional image is a planar image of the vehicle. By converting this planar image into a three-dimensional model, the vehicle's spatial structure can be reconstructed, thereby enhancing vehicle information transparency, optimizing risk management processes, and improving customer experience. When dividing the planar image of the vehicle into panes, it is necessary to divide it into multiple panes based on the structural features of the vehicle within the image. For example, the windows are divided into one pane, the rearview mirrors into another, and so on.

[0021] 102. Predict the local approximate plane normal of each pane within the viewpoint range of the 3D model to be constructed. Determine the starting point of 3D modeling in the 2D image and determine the 3D coordinates of the starting point. Take the pane where the starting point of 3D modeling is located as the starting pane and take the 3D coordinates of the starting point as the starting center coordinates of the starting local approximate plane within the viewpoint range of the starting pane.

[0022] In the 3D model, the outer surface of the skull can be approximated as a curved surface, but within a local viewing angle, it can be considered a plane. For this embodiment of the invention, the device parameters of the 2D image capturing device, such as the camera's basic parameters, including pixel spacing and slice thickness, are first obtained. Taking the skull pane as an example, corresponding regions of the skull in several adjacent CT images are selected. In each image layer, feature points of the skull edge are determined. Then, using triangulation and plane fitting methods in multi-view geometry, these feature points are fitted in 3D space according to the pixel spacing and slice thickness parameters to obtain the plane equation of the skull surface in that region. The normal direction of the local approximate plane is calculated based on the plane equation, and then the normal is determined based on the normal direction. Thus, the local approximate plane normal within the viewing angle of each pane can be determined as described above.

[0023] Furthermore, a feature point with significant geometric structure is selected in the 2D image. The 2D image has a known layer thickness and pixel spacing; assuming a layer thickness of 1 mm and a pixel spacing of 0.5 mm, the 3D coordinates can be calculated based on the coordinates of the 3D modeling starting point in the 2D image and these parameters. For example, if the 2D coordinates of the 3D modeling starting point are (256, 200), the actual size in the X direction is 256 × 0.5 = 128 mm = 12.8 cm, the actual size in the Y direction is 200 × 0.5 = 100 mm = 10 cm, and the Z-coordinate is determined according to the number of scan layers in the image. Assuming the point is in layer 10, the z-coordinate is 10 × 1 = 10 mm = 1 cm. Therefore, the 3D coordinates of the 3D modeling starting point are (12.8, 10, 1). If the 3D modeling starting point is located in the skull pane, then the skull pane is the starting point pane. The three-dimensional coordinates (12.8, 10, 1) of the starting point of the three-dimensional modeling are used as the center coordinates of the starting point of the local approximate plane (skull surface) within the starting point pane view range.

[0024] 103. Determine the neighboring panes adjacent to the starting pane. Based on the starting center coordinates, the local approximate plane normal of the starting pane, and the local approximate plane normal of the neighboring panes, determine the neighborhood center coordinates of the local approximate plane within the viewing range of the neighboring panes.

[0025] In this embodiment of the invention, in order to convert a two-dimensional image into a three-dimensional model, it is first necessary to determine the coordinates of the neighborhood center of the local approximate plane within the viewing range of the neighborhood pane. Based on this, step 103 specifically includes: determining the sphere attribute information of the virtual sphere corresponding to the starting point of the three-dimensional modeling based on the local approximate plane normal of the starting point pane and the local approximate plane normal of the neighborhood pane; and determining the neighborhood center coordinates of the local approximate plane within the viewing range of the neighborhood pane based on the starting center coordinates of the starting point local approximate plane, the sphere attribute information, the local approximate plane normal of the starting point pane, and the local approximate plane normal of the neighborhood pane. The method for determining the sphere's attribute information includes: moving the local approximate plane normal of the starting pane and the local approximate plane normal of the neighboring pane in opposite directions by a preset distance, determining the intersection point of the translated local approximate plane normal and the translated local approximate plane normal, and taking the intersection point as the center of the virtual sphere; determining the starting projection point of the intersection point between the local approximate plane normal of the starting pane and the starting local approximate plane in the two-dimensional image, determining the neighboring projection point of the intersection point between the local approximate plane normal of the neighboring pane and the neighboring local approximate plane in the two-dimensional image, and determining the projection distance between the starting projection point and the neighboring projection point in the two-dimensional image, and taking the projection distance as the arc length of the virtual sphere; determining the angle between the local approximate plane normal of the starting pane and the local approximate plane normal of the neighboring pane, and taking the ratio between the arc length and the angle as the radius of the virtual sphere.

[0026] Specifically, for the local approximate planes within the viewing range corresponding to two adjacent panes (the starting pane and the neighboring pane), the normals located at the center of the image are each shifted by a pair of opposite distances and then intersect at point O in three-dimensional space. Here, opposite distances mean that the two normals are shifted by equal magnitudes but opposite directions. For example, let the normals at the center of the local approximate planes A and B corresponding to adjacent panes be n... A and n B , will n A Translate n a distance d in a certain direction. BThe two normals are translated a distance *d* in opposite directions until they intersect at a point *O* in space. This intersection point is considered the center of the sphere. The projection distance of the intersection points of the two normals with their respective planes in the two-dimensional image is approximated as the arc length on the sphere. The included angle between the two normals is determined, and the ratio of the arc length to the included angle is taken as the radius of the sphere. The arc length, sphere radius, and sphere center are taken as the sphere attribute information. Further, after determining the sphere attribute information, it is necessary to determine the neighborhood center coordinates of the neighborhood local approximate plane within the corresponding viewpoint range of the neighborhood pane based on the sphere attribute information. Based on this, the method includes: determining the starting direction vector of the local approximate plane normal of the starting pane based on the sphere center and the starting center coordinates, and determining the neighborhood direction vector of the local approximate plane normal of the neighborhood pane based on the sphere center and the starting direction vector; moving the sphere radius along the neighborhood direction vector from the sphere center to the target point, and taking the coordinates of the target point as the neighborhood center coordinates of the neighborhood local approximate plane within the viewpoint range of the neighborhood pane.

[0027] Specifically, since the center coordinates of the local approximation plane corresponding to the starting window are known, the center coordinates of the other plane (the local approximation plane of the neighborhood) can be determined based on the sphere's center position, radius, and the relative positional relationship (such as normal direction and angle between the planes) of the two planes (the local approximation plane of the starting point and the local approximation plane of the neighborhood). For example, if the center coordinates of the starting point are... The coordinates of the sphere's center are Given a sphere with radius r, and the known directions and positional relationships of the normals between two planes, the center coordinates of the other plane can be determined using vector operations. Specifically, if the coordinates of the starting point center are (12.8, 10, 1), and the coordinates of the sphere center O are (0, 0, 0), then the starting point direction vector... If the right pane adjacent to the starting pane is selected as the neighboring pane, the starting direction vector can be transformed to obtain the neighborhood direction vector of the local approximate plane normal of the neighboring pane. For example, it can be assumed that the neighborhood direction vector is offset from the starting direction vector in the X direction, while maintaining a relative relationship in the Y and Z directions. Assuming the offset is 2 cm (which can be adjusted according to the actual situation), the neighborhood direction vector... If the radius of a sphere is r = 5 cm, and a target point is reached by moving a distance equal to the radius of the sphere along the neighborhood direction vector from the center of the sphere, the coordinates of the target point are calculated as the coordinates of the sphere's center plus the normalized neighborhood direction vector multiplied by the sphere's radius. First, the magnitude of the neighborhood direction vector is calculated. Then, the neighborhood direction vectors are normalized, that is, the ratio of the magnitude of the neighborhood direction vectors to the magnitude of the neighborhood direction vectors is used as the normalized result. The final target point coordinates (neighborhood center coordinates) .

[0028] 104. Based on the starting point center coordinates and the neighborhood center coordinates, the starting point local approximation plane and the neighborhood local approximation plane are stitched together. After stitching, the neighborhood pane is used as the new starting point pane, and the new neighborhood pane adjacent to the new starting point pane is determined. The stitching of the starting point local approximation plane and the neighborhood local approximation plane within the view range of the new starting point pane and the view range of the new neighborhood pane is continued until the local approximation plane within the view range of each pane is stitched together, thus obtaining the three-dimensional model corresponding to the two-dimensional image.

[0029] Specifically, in determining the starting point center coordinates and the neighborhood center coordinates, it is necessary to stitch two local approximate planes together based on the aforementioned coordinates. Therefore, step 104 specifically includes: determining the translation vector between the neighborhood local approximate plane and the starting point local approximate plane based on the starting point center coordinates and the neighborhood center coordinates; determining the normal angle and rotation axis between the normal of the local approximate plane of the starting point pane and the normal of the local approximate plane of the neighborhood pane, and determining the rotation matrix based on the normal angle and the rotation axis; and stitching the neighborhood local approximate plane onto the starting point local approximate plane based on the translation vector and the rotation matrix.

[0030] Specifically, if the local approximate plane at the starting point is The local approximate plane in the neighborhood is ,flat The coordinates of the center point are ,flat The coordinates of the center point are ,flat The direction of the normal is ,flat The direction of the normal is The angle between the two normals is Then the translation vector ,Will and The perpendicular line formed by the plane is determined as the axis of rotation, such as if it can be obtained through... and The cross product is used to obtain the direction of the rotation axis, thus determining the rotation axis. Then, based on the rotation axis and the included angle, a rotation matrix is ​​constructed and applied to the plane. For all points on the plane, apply the translation vector to the rotated plane. (that is, according to the translation vector and rotation matrix, the plane) (Perform translation and rotation) to move it to its plane. The relative positions. Further adjustments to the plane can be made to improve splicing accuracy. The direction, such as making the plane One of the edges of the plane Align with one of the edges.

[0031] Furthermore, the neighboring pane is used as a new starting pane, and the panes adjacent to the neighboring pane are used as new neighboring panes. Following the above method, the local approximate planes within the viewpoint range corresponding to the new neighboring panes are stitched onto the local metallic planes within the viewpoint range corresponding to the new starting pane. This process is repeated to gradually stitch the local approximate planes within the viewpoint range of each pane until all the planes within the viewpoint range corresponding to all panes are in their final positions, ultimately obtaining the three-dimensional model corresponding to the two-dimensional image. Therefore, this embodiment of the invention, through planar stitching, can preserve the texture, color, and other detailed information in the image, thereby improving the construction quality of the three-dimensional model. Simultaneously, the local approximate plane normals and center coordinates are easily and accurately determined. Therefore, this invention uses local approximate plane normals and center coordinates to construct the three-dimensional model, improving the construction accuracy of the three-dimensional model. In addition, this embodiment of the invention starts from a small pane in the image, obtains the normal direction based on planar normal prediction technology, and stitches the planar boundaries to construct the three-dimensional structure using specific extension methods (translating normal intersections, calculating the sphere radius, etc.). This method is more sensitive to local undulations, thus improving the construction accuracy of the three-dimensional model.

[0032] According to the present invention, a method for constructing a 3D model, compared with the current method of obtaining 3D point clouds through image depth estimation and then constructing a 3D model, the present invention divides a 2D image into multiple panes and predicts the local approximate plane normal within the viewpoint range corresponding to each pane in the 3D model to be constructed. Then, it determines the starting center coordinates of the starting local approximate plane within the viewpoint range of the starting pane, and the neighborhood center coordinates of the neighborhood local approximate planes within the viewpoint range of the neighboring panes adjacent to the starting pane. Finally, based on the starting center coordinates and the neighborhood center coordinates, the starting local approximate plane and the neighborhood local approximate planes are stitched together, with the neighboring panes used as new starting panes and the panes adjacent to the neighboring panes used as new neighboring panes. This process of stitching local approximate planes continues until all local approximate planes corresponding to all panes participate in the stitching, ultimately obtaining a 3D model. Therefore, the present invention, through planar stitching, can preserve the texture, color, and other detailed information in the image, thereby improving the construction quality of the 3D model. Simultaneously, the local approximate plane normals and center coordinates are easily and accurately determined. Therefore, the present invention, by using local approximate plane normals and center coordinates to construct the 3D model, can improve the construction accuracy of the 3D model.

[0033] Furthermore, in order to better illustrate the above process of constructing a three-dimensional model, as a refinement and extension of the above embodiment, this embodiment of the invention provides another method for constructing a three-dimensional model, as shown in Figure 2. The method includes: 201. In response to a construction signal of a three-dimensional model to be constructed, acquiring a two-dimensional image corresponding to the three-dimensional model to be constructed, and dividing the two-dimensional image into multiple panes.

[0034] Specifically, when a 3D model construction signal is received, the 2D image is divided into multiple panes according to the structural information of the objects in the 2D image.

[0035] 202. Obtain a preset normal prediction model, wherein the preset normal prediction model includes a CNN layer for extracting image features, a perspective transformation layer for image transformation, and a prediction layer for normal prediction.

[0036] 203. Input the local approximate plane within the viewpoint range corresponding to each pane in the 3D model to be constructed into the preset normal prediction model. Extract the image features of the local approximate plane through the CNN layer, perform perspective transformation on the image features through the perspective transformation matrix in the perspective transformation layer, and perform normal prediction on the perspective-transformed image features through the prediction layer to obtain the local approximate plane normal of each pane within the viewpoint range corresponding to the 3D model to be constructed.

[0037] Specifically, to improve the normal prediction accuracy of the preset normal prediction model, it is first necessary to train and construct the preset normal prediction model. Based on this, the method includes: constructing a preset initial normal prediction model; obtaining a sample dataset, wherein the sample dataset includes local approximate planes of sample 2D images with normal labels within the viewpoint range of the corresponding sample 3D model; dividing the sample dataset into a training set and a test set, using the training set to train the preset initial normal prediction model, and using the test set to test the trained preset initial normal prediction model, finally selecting the trained preset initial normal prediction model that meets the test conditions as the preset normal prediction model. Specifically, during model training, firstly, a preset initial normal prediction model is constructed, the structure of which is the same as the preset normal prediction model. Secondly, a sample dataset is obtained. It is ensured that the dataset contains all necessary files, including multiple local approximate planes (image patches) of 2D images within the viewpoint range of the corresponding sample 3D model. The data is converted to a format that the preset initial normal prediction model can understand, and finally, the model is trained and tested. Specifically, the dataset can be divided first: using random or specific strategies (such as stratified sampling), the sample dataset can be divided into a training set and a test set. The training set is then used to train the model, and the test set is used to test the trained model and evaluate its performance on unseen data. Metrics such as mCP, precision, and recall on the test set are calculated and recorded. If the model performance does not meet the requirements, it can return to the training phase for further iterations or adjustments. This yields a preset normal prediction model that meets the requirements. Then, local approximate planes (image patches) are input into the preset normal prediction model. The preset normal prediction model can directly output the normal direction of the local approximate plane within the viewpoint range of each pane, thus determining the normal based on the normal direction. This embodiment of the invention uses a model for normal prediction, which can improve the prediction accuracy and efficiency of normals, thereby improving the efficiency and accuracy of 3D modeling. Furthermore, the model in this embodiment contains a perspective transformation matrix, which can perform perspective transformation on image features, enhancing image quality and further improving the prediction accuracy of normals.

[0038] 204. Determine the starting point of the 3D modeling in the 2D image, and determine the 3D coordinates of the starting point. Use the pane where the starting point of the 3D modeling is located as the starting point pane, and use the 3D coordinates of the starting point as the center coordinates of the starting point of the local approximate plane within the view range of the starting point pane.

[0039] Specifically, the point with significant geometric features at the center of the two-dimensional image can be used as the starting point for three-dimensional modeling, and the three-dimensional structure can be expanded starting from the pane corresponding to the starting point of the three-dimensional modeling.

[0040] 205. Determine the neighboring panes adjacent to the starting pane. Based on the center coordinates of the starting pane, the local approximate plane normal of the starting pane, and the local approximate plane normal of the neighboring panes, determine the center coordinates of the neighboring local approximate plane within the viewing range of the neighboring panes.

[0041] Specifically, by utilizing the fact that normals can be translated, the normals of the local approximate planes within the viewpoint of adjacent panes are each shifted by a pair of opposite distances, and then intersect in three-dimensional space. The intersection point is considered the center of a sphere. The projection distance of the intersection points of the two current normals with their respective planes onto the screen is approximated as the arc length on the sphere. Combined with the included angle of the normals, the radius of the sphere is obtained. This yields the relative three-dimensional information of the two local approximate planes. Since the center coordinates of one local approximate plane are known, the center coordinates of the other local approximate plane can be determined from this information.

[0042] 206. Based on the center coordinates of the starting point and the center coordinates of the neighborhood, the local approximate plane of the starting point and the local approximate plane of the neighborhood are stitched together. After stitching, the neighborhood pane is used as the new starting pane, and the new neighborhood pane adjacent to the new starting pane is determined. The local approximate plane of the starting point within the view range of the new starting pane and the local approximate plane of the neighborhood within the view range of the new neighborhood pane are stitched together until the local approximate planes within the view range of each pane are stitched together, thus obtaining the three-dimensional model corresponding to the two-dimensional image.

[0043] In this embodiment of the invention, the translation vector and rotation matrix between two planes can be determined based on information such as the starting center coordinates, the neighborhood center coordinates, the sphere radius, and the included angle of the normals. Based on the translation vector and rotation matrix, the local approximate plane within the viewing range of the neighborhood pane is moved towards the local approximate plane within the viewing range of the starting window, thereby achieving the stitching of the two planes. After stitching, it is necessary to continue stitching the local approximate planes within the remaining viewing range of the panes. Therefore, the method includes: using the three-dimensional coordinates of the center pixel of the new starting window as the new starting center coordinates; determining the new neighborhood center coordinates of the neighborhood local approximate planes within the viewing range of the new neighborhood pane based on the new starting center coordinates, the normal of the local approximate plane of the new starting window, and the normal of the local approximate plane of the new neighborhood pane; and stitching the starting local approximate planes within the viewing range of the new starting window and the neighborhood local approximate planes within the viewing range of the new neighborhood pane based on the new starting center coordinates and the new neighborhood center coordinates.

[0044] Specifically, the neighboring pane is used as a new starting pane, and the panes adjacent to the neighboring pane are used as new neighboring panes. A 3D modeling starting point is determined in the new starting pane, and its 3D coordinates are determined. These 3D coordinates are used as the starting center coordinates of the local approximate plane within the view range of the new starting pane. Based on the starting center coordinates, the local approximate plane normal of the new starting pane, and the local approximate plane normal of the new neighboring pane, the neighborhood center coordinates of the local approximate plane within the view range of the new neighboring pane are determined. Based on the starting center coordinates and the neighborhood center coordinates, the starting local approximate plane and the neighborhood local approximate plane are stitched together. After stitching, the new neighboring pane is used as the next new starting pane, and the next new neighboring pane adjacent to this next new starting pane is determined. The stitching of the starting local approximate plane within the view range of the next new starting pane and the neighborhood local approximate plane within the view range of the next new neighboring pane continues. This process is repeated to stitch together the local approximate planes within the view range of each pane until all the local approximate planes within the view range of each pane are stitched together, resulting in a 3D model corresponding to the 2D image. For example, if there are six consecutive adjacent panes, first, take the first pane as the starting pane and the second pane as the neighboring pane. Then, stitch the local approximate plane within the view range of the second pane onto the local approximate plane within the view range of the first pane. Next, take the second pane as the starting pane, take the third pane as the neighboring pane, and stitch the local approximate plane within the view range of the third pane onto the local approximate plane within the view range of the second pane. Then, take the third pane as the starting pane, take the fourth pane as the neighboring pane, and stitch the local approximate plane within the view range of the fourth pane onto the local approximate plane within the view range of the third pane. Next, take the fourth pane as the starting pane, take the fifth pane as the neighboring pane, and stitch the local approximate plane within the view range of the fifth pane onto the local approximate plane within the view range of the fourth pane. Finally, take the fifth pane as the starting pane, take the sixth pane as the neighboring pane, and stitch the local approximate plane within the view range of the sixth pane onto the local approximate plane within the view range of the fifth pane. This will result in a 3D model. Therefore, by using planar stitching, details such as texture and color in the image can be preserved, thereby improving the construction quality of the 3D model. At the same time, the local approximate plane normal and center coordinates can be easily and accurately determined. Thus, this invention uses the local approximate plane normal and center coordinates to construct the 3D model, thereby improving the construction accuracy of the 3D model.

[0045] According to another method for constructing a 3D model provided by the present invention, compared with the current method of obtaining 3D point clouds through image depth estimation and then constructing a 3D model, the present invention divides a 2D image into multiple panes and predicts the local approximate plane normal within the viewpoint range corresponding to each pane in the 3D model to be constructed. Then, it determines the starting center coordinates of the starting local approximate plane within the viewpoint range of the starting pane, and the neighborhood center coordinates of the neighborhood local approximate planes within the viewpoint range of the neighboring panes adjacent to the starting pane. Finally, based on the starting center coordinates and the neighborhood center coordinates, the starting local approximate plane and the neighborhood local approximate planes are stitched together, with the neighboring panes used as new starting panes and the panes adjacent to the neighboring panes used as new neighboring panes. This process of stitching local approximate planes continues until all local approximate planes corresponding to all panes participate in the stitching, ultimately obtaining a 3D model. Therefore, the present invention, through planar stitching, can preserve the texture, color, and other detailed information in the image, thereby improving the construction quality of the 3D model. Simultaneously, the local approximate plane normals and center coordinates are easily and accurately determined. Therefore, the present invention, by using local approximate plane normals and center coordinates to construct the 3D model, can improve the construction accuracy of the 3D model.

[0046] Furthermore, as a specific implementation of Figure 1, this embodiment of the invention provides a three-dimensional model construction device, as shown in Figure 3. The device includes: an acquisition unit 31, a prediction unit 32, a determination unit 33, and a splicing unit 34.

[0047] The acquisition unit 31 can be used to acquire a two-dimensional image corresponding to the three-dimensional model to be constructed in response to the construction signal of the three-dimensional model to be constructed, and divide the two-dimensional image into multiple panes.

[0048] The prediction unit 32 can be used to predict the local approximate plane normal of each pane in the view range corresponding to the three-dimensional model to be constructed, determine the starting point of three-dimensional modeling in the two-dimensional image, determine the three-dimensional coordinates of the starting point of three-dimensional modeling, take the pane where the starting point of three-dimensional modeling is located as the starting pane, and take the three-dimensional coordinates of the starting point of three-dimensional modeling as the starting center coordinates of the starting local approximate plane in the view range of the starting pane.

[0049] The determining unit 33 can be used to determine the neighboring panes adjacent to the starting pane, and to determine the neighborhood center coordinates of the neighborhood local approximate plane within the viewing range of the neighboring pane based on the starting center coordinates, the local approximate plane normal of the starting pane, and the local approximate plane normal of the neighboring pane.

[0050] The stitching unit 34 can be used to stitch the local approximate plane of the starting point and the local approximate plane of the neighborhood based on the center coordinates of the starting point and the center coordinates of the neighborhood. After stitching, the neighborhood pane is used as the new starting pane, and a new neighborhood pane adjacent to the new starting pane is determined. The stitching continues on the local approximate plane of the starting point within the view range of the new starting pane and the local approximate plane of the neighborhood within the view range of the new neighborhood pane, until the local approximate planes within the view range of each pane are stitched together, thereby obtaining the three-dimensional model corresponding to the two-dimensional image.

[0051] In specific application scenarios, in order to determine the local approximate plane normal, as shown in Figure 4, the prediction unit 32 includes an acquisition module 321 and a prediction module 322.

[0052] In specific application scenarios, in order to determine the coordinates of the neighborhood center, the determining unit 33 can be used to determine the sphere attribute information of the virtual sphere corresponding to the starting point of the 3D modeling based on the local approximate plane normal of the starting point pane and the local approximate plane normal of the neighborhood pane; and to determine the neighborhood center coordinates of the neighborhood local approximate plane within the viewing range of the neighborhood pane based on the starting point center coordinates of the starting point local approximate plane, the sphere attribute information, the local approximate plane normal of the starting point pane, and the local approximate plane normal of the neighborhood pane.

[0053] In specific application scenarios, to determine the sphere's attribute information, the determining unit 33 can be specifically used to move the local approximate plane normal of the starting pane and the local approximate plane normal of the neighboring pane in opposite directions by a preset distance, determine the intersection point of the translated local approximate plane normal and the translated local approximate plane normal, and take the intersection point as the center of the virtual sphere; determine the starting projection point of the intersection point between the local approximate plane normal of the starting pane and the starting local approximate plane in the two-dimensional image, determine the neighboring projection point of the intersection point between the local approximate plane normal of the neighboring pane and the neighboring local approximate plane in the two-dimensional image, and determine the projection distance between the starting projection point and the neighboring projection point in the two-dimensional image, and take the projection distance as the arc length of the virtual sphere; determine the angle between the local approximate plane normal of the starting pane and the local approximate plane normal of the neighboring pane, and take the ratio between the arc length and the angle as the radius of the virtual sphere.

[0054] In a specific application scenario, in order to determine the coordinates of the neighborhood center, the determining unit 33 can be used to determine the starting direction vector of the local approximate plane normal of the starting window based on the coordinates of the sphere center and the starting point center, and to determine the neighborhood direction vector of the local approximate plane normal of the neighborhood window based on the sphere center and the starting direction vector; after moving the sphere radius from the sphere center along the neighborhood direction vector to the target point, the coordinates of the target point are used as the neighborhood center coordinates of the neighborhood local approximate plane within the viewing range of the neighborhood window.

[0055] In specific application scenarios, in order to stitch together local approximate planes, the stitching unit 34 includes a determining module 341 and a stitching module 342.

[0056] The determining module 341 can be used to determine the translation vector between the local approximate plane of the neighborhood and the local approximate plane of the starting point based on the center coordinates of the starting point and the center coordinates of the neighborhood.

[0057] The determining module 341 can also be used to determine the normal angle and rotation axis between the local approximate plane normal of the starting window and the local approximate plane normal of the neighboring window, and to determine the rotation matrix based on the normal angle and the rotation axis.

[0058] The splicing module 342 can be used to splice the neighborhood local approximation plane onto the starting local approximation plane based on the translation vector and the rotation matrix.

[0059] In specific application scenarios, in order to stitch together the local approximate planes in the neighborhood, the determining module 341 can also be used to take the three-dimensional coordinates of the center pixel of the new starting point pane as the new starting point center coordinates, and determine the new neighborhood center coordinates of the local approximate planes within the viewing range of the new neighborhood panes based on the new starting point center coordinates, the local approximate plane normal of the new starting point pane, and the local approximate plane normal of the new neighborhood panes.

[0060] The splicing module 342 can also be used to splice the local approximate plane of the starting point within the view range of the new starting point pane and the local approximate plane of the neighborhood within the view range of the new neighborhood pane, based on the new starting point center coordinates and the new neighborhood center coordinates.

[0061] It should be noted that other corresponding descriptions of the functional modules involved in the three-dimensional model construction device provided in this embodiment of the invention can be found in the corresponding description of the method shown in Figure 1, and will not be repeated here.

[0062] Based on the method shown in Figure 1, this embodiment of the invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, performs the following steps: In response to a construction signal for a three-dimensional model to be constructed, acquire a two-dimensional image corresponding to the three-dimensional model to be constructed, and divide the two-dimensional image into multiple panes; predict the local approximate plane normal within the viewpoint range corresponding to each pane in the three-dimensional model to be constructed; determine the three-dimensional modeling starting point in the two-dimensional image, and determine the three-dimensional coordinates of the three-dimensional modeling starting point; take the pane containing the three-dimensional modeling starting point as the starting pane; take the three-dimensional coordinates of the three-dimensional modeling starting point as the starting center coordinates of the starting local approximate plane within the viewpoint range of the starting pane; determine the coordinates of the starting pane... For adjacent neighboring panes, based on the starting center coordinates, the local approximate plane normal of the starting pane, and the local approximate plane normal of the neighboring panes, the neighborhood center coordinates of the local approximate planes within the viewing range of the neighboring panes are determined. Based on the starting center coordinates and the neighborhood center coordinates, the starting local approximate plane and the neighboring local approximate planes are stitched together. After stitching, the neighboring pane is used as the new starting pane, and new neighboring panes adjacent to the new starting pane are determined. The stitching of the starting local approximate planes within the viewing range of the new starting panes and the neighboring local approximate planes within the viewing range of the new neighboring panes continues until the local approximate planes within the viewing range of each pane are stitched together, thus obtaining the three-dimensional model corresponding to the two-dimensional image.

[0063] Based on the embodiments of the method shown in Figure 1 and the device shown in Figure 3, this embodiment of the invention also provides a physical structure diagram of a computer device, as shown in Figure 5. The computer device includes: a processor 41, a memory 42, and a computer program stored in the memory 42 and executable on the processor. Both the memory 42 and the processor 41 are mounted on a bus 43. When the processor 41 executes the program, it performs the following steps: in response to a construction signal for a three-dimensional model to be constructed, it acquires a two-dimensional image corresponding to the three-dimensional model to be constructed and divides the two-dimensional image into multiple panes; it predicts the local approximate plane normal within the viewpoint range corresponding to each pane in the three-dimensional model to be constructed; it determines the three-dimensional modeling starting point in the two-dimensional image and determines the three-dimensional coordinates of the three-dimensional modeling starting point; it uses the pane where the three-dimensional modeling starting point is located as the starting pane; and it sets the three-dimensional coordinates of the three-dimensional modeling starting point... The starting center coordinates of the starting local approximation plane within the view range of the starting pane are determined; neighboring panes adjacent to the starting pane are determined, and based on the starting center coordinates, the local approximation plane normal of the starting pane, and the local approximation plane normal of the neighboring panes, the neighborhood center coordinates of the neighboring local approximation plane within the view range of the neighboring panes are determined; based on the starting center coordinates and the neighborhood center coordinates, the starting local approximation plane and the neighboring local approximation plane are stitched together, and the neighboring pane is used as the new starting pane, and new neighboring panes adjacent to the new starting pane are determined, and the stitching of the starting local approximation plane within the view range of the new starting pane and the neighboring local approximation plane within the view range of the new neighboring pane continues until the local approximation planes within the view range of each pane are stitched together, thus obtaining the three-dimensional model corresponding to the two-dimensional image.

[0064] The present invention divides a two-dimensional image into multiple panes and predicts the local approximate plane normal within the viewpoint range of each pane in the three-dimensional model to be constructed. Then, it determines the starting center coordinates of the starting local approximate plane within the viewpoint range of the starting pane, and the neighborhood center coordinates of the neighborhood local approximate planes within the viewpoint range of the neighboring panes adjacent to the starting pane. Finally, based on the starting center coordinates and the neighborhood center coordinates, the starting local approximate plane and the neighborhood local approximate planes are stitched together, with the neighboring panes becoming the new starting panes and the panes adjacent to the neighboring panes becoming the new neighboring panes. This process of stitching local approximate planes continues until all local approximate planes corresponding to all panes participate in the stitching, ultimately obtaining the three-dimensional model. Therefore, the present invention, through planar stitching, can preserve the texture, color, and other detailed information in the image, thereby improving the construction quality of the three-dimensional model. Simultaneously, the local approximate plane normals and center coordinates are easily and accurately determined. Therefore, the present invention, by using local approximate plane normals and center coordinates to construct the three-dimensional model, can improve the construction accuracy of the three-dimensional model.

[0065] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0066] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for constructing a three-dimensional model, characterized in that, include: In response to a construction signal for a 3D model to be constructed, a 2D image corresponding to the 3D model to be constructed is acquired, and the 2D image is divided into multiple panes; Predict the local approximate plane normal within the viewpoint range corresponding to each pane in the 3D model to be constructed. Determine the starting point of 3D modeling in the 2D image and its 3D coordinates. Use the pane containing the starting point as the starting pane and its 3D coordinates as the starting center coordinates of the starting local approximate plane within the viewpoint range of the starting pane. Determine the neighboring panes adjacent to the starting pane. Based on the starting center coordinates, the local approximate plane normal of the starting pane, and the local approximate plane normal of the neighboring panes, determine the neighborhood center coordinates of the neighborhood local approximate plane within the viewpoint range of the neighboring panes. Based on the starting center coordinates and the neighborhood center coordinates, stitch the starting local approximate plane and the neighborhood local approximate plane. After stitching, use the neighboring pane as the new starting pane and determine the new neighboring panes adjacent to it. Continue stitching the starting local approximate plane within the viewpoint range of the new starting pane and the neighborhood local approximate plane within the viewpoint range of the new neighboring pane until the local approximate planes within the viewpoint range of each pane are stitched together to obtain the 3D model corresponding to the 2D image.

2. The method according to claim 1, characterized in that, The method for predicting the local approximate plane normals within the viewpoint range corresponding to each pane in the 3D model to be constructed includes: obtaining a preset normal prediction model, wherein the preset normal prediction model includes a CNN layer for extracting image features, a perspective transformation layer for image transformation, and a prediction layer for normal prediction; inputting the local approximate planes within the viewpoint range corresponding to each pane in the 3D model to be constructed into the preset normal prediction model; extracting image features of the local approximate planes through the CNN layer; performing perspective transformation on the image features through the perspective transformation matrix in the perspective transformation layer; and performing normal prediction on the perspective-transformed image features through the prediction layer to obtain the local approximate plane normals within the viewpoint range corresponding to each pane in the 3D model to be constructed.

3. The method according to claim 1, characterized in that, Based on the starting point center coordinates, the local approximate plane normal of the starting point pane, and the local approximate plane normal of the neighboring pane, the neighborhood center coordinates of the neighborhood local approximate plane within the viewing range of the neighboring pane are determined, including: determining the sphere attribute information of the virtual sphere corresponding to the 3D modeling starting point based on the local approximate plane normal of the starting point pane and the local approximate plane normal of the neighboring pane; and determining the neighborhood center coordinates of the neighborhood local approximate plane within the viewing range of the neighboring pane based on the starting point center coordinates, the sphere attribute information, the local approximate plane normal of the starting point pane, and the local approximate plane normal of the neighboring pane.

4. The method according to claim 3, characterized in that, Based on the local approximate plane normal of the starting pane and the local approximate plane normal of the neighboring pane, the sphere attribute information of the virtual sphere corresponding to the starting point of the 3D modeling is determined, including: moving the local approximate plane normal of the starting pane and the local approximate plane normal of the neighboring pane in opposite directions by a preset distance respectively, determining the intersection point of the translated local approximate plane normal and the translated local approximate plane normal, and taking the intersection point as the center of the virtual sphere; determining the starting projection point of the intersection point between the local approximate plane normal of the starting pane and the starting local approximate plane in the 2D image, determining the neighboring projection point of the intersection point between the local approximate plane normal of the neighboring pane and the neighboring local approximate plane in the 2D image, and determining the projection distance between the starting projection point and the neighboring projection point in the 2D image, and taking the projection distance as the arc length of the virtual sphere; determining the angle between the local approximate plane normal of the starting pane and the local approximate plane normal of the neighboring pane, and taking the ratio between the arc length and the angle as the radius of the virtual sphere.

5. The method according to claim 4, characterized in that, Based on the starting center coordinates of the starting local approximation plane, the sphere attribute information, the local approximation plane normal of the starting pane, and the local approximation plane normal of the neighboring pane, the neighborhood center coordinates of the neighborhood local approximation plane within the viewing range of the neighboring pane are determined, including: based on the sphere center and the starting center coordinates, determining the starting direction vector of the local approximation plane normal of the starting pane, and based on the sphere center and the starting direction vector, determining the neighborhood direction vector of the local approximation plane normal of the neighboring pane; moving the sphere radius along the neighborhood direction vector from the sphere center to the target point, and using the coordinates of the target point as the neighborhood center coordinates of the neighborhood local approximation plane within the viewing range of the neighboring pane.

6. The method according to claim 1, characterized in that, Based on the starting point center coordinates and the neighborhood center coordinates, the local approximation plane of the starting point and the local approximation plane of the neighborhood are spliced ​​together, including: determining the translation vector between the local approximation plane of the neighborhood and the local approximation plane of the starting point based on the starting point center coordinates and the neighborhood center coordinates; determining the normal angle and rotation axis between the local approximation plane normal of the starting point pane and the local approximation plane normal of the neighborhood pane, and determining the rotation matrix based on the normal angle and the rotation axis; and splicing the local approximation plane of the neighborhood onto the local approximation plane of the starting point based on the translation vector and the rotation matrix.

7. The method according to claim 1, characterized in that, The process of stitching together the local approximate plane of the starting point within the new starting point pane's viewpoint and the local approximate plane of the neighboring pane's viewpoint includes: using the three-dimensional coordinates of the center pixel of the new starting point pane as the new starting point center coordinates; determining the new neighboring center coordinates of the local approximate plane of the neighboring pane's viewpoint based on the new starting point center coordinates, the local approximate plane normal of the new starting point pane, and the local approximate plane normal of the new neighboring pane's viewpoint; and stitching together the local approximate plane of the starting point within the new starting point pane's viewpoint and the local approximate plane of the neighboring pane's viewpoint based on the new starting point center coordinates and the new neighboring center coordinates.

8. A device for constructing a three-dimensional model, characterized in that, include: The acquisition unit is used to acquire a two-dimensional image corresponding to the three-dimensional model to be constructed in response to the construction signal of the three-dimensional model to be constructed, and divide the two-dimensional image into multiple panes; the prediction unit is used to predict the local approximate plane normal of each pane within the view range corresponding to the three-dimensional model to be constructed, determine the three-dimensional modeling starting point in the two-dimensional image, determine the three-dimensional coordinates of the three-dimensional modeling starting point, take the pane where the three-dimensional modeling starting point is located as the starting point pane, and take the three-dimensional coordinates of the three-dimensional modeling starting point as the starting center coordinates of the starting local approximate plane within the view range of the starting point pane; A determining unit is used to determine neighboring panes adjacent to the starting pane. Based on the center coordinates of the starting pane, the local approximate plane normal of the starting pane, and the local approximate plane normal of the neighboring panes, it determines the center coordinates of the neighboring local approximate planes within the viewing range of the neighboring panes. A stitching unit is used to stitch the starting local approximate plane and the neighboring local approximate planes based on the center coordinates of the starting pane and the center coordinates of the neighboring panes. After stitching, the neighboring pane is used as a new starting pane. A new neighboring pane adjacent to the new starting pane is determined. The stitching of the starting local approximate plane and the neighboring local approximate plane within the viewing range of the new starting pane and the new neighboring panes continues until the local approximate planes within the viewing range of each pane are stitched together, thus obtaining the three-dimensional model corresponding to the two-dimensional image.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images