A method for indoor 3D reconstruction taking into account indoor curved surface structures

Abstract

The present invention discloses an indoor three-dimensional reconstruction method taking into account indoor curved surface structures, and relates to the technical field of map making. After inputting the indoor three-dimensional point cloud, the present invention first divides the indoor three-dimensional point cloud into surface points and plane supervoxel sets based on supervoxels; for the surface points and plane supervoxel sets, model fitting is performed respectively, and the indoor plane and surface models are extracted, and the indoor two-dimensional space is divided based on the 3D-2D projection model; the indoor two-dimensional plane diagram is constructed based on superposition analysis and Markov field; and finally the indoor three-dimensional model is established based on constrained Delaunay triangulation. The present invention performs indoor three-dimensional reconstruction based on three-dimensional point clouds, realizes the three-dimensional reconstruction of indoor curved surface structures, especially cylindrical curved surface structures, and can quickly, effectively and accurately realize indoor scene reconstruction, meeting the navigation and modeling requirements of complex indoor scenes.

Description

Technical Field

[0001] The present invention relates to the field of map making technology, and in particular to an indoor three-dimensional reconstruction method taking into account indoor curved surface structures. Background Art

[0002] With the rapid development of economic activities, the number of buildings has increased significantly, and human indoor activities have increased significantly. The internal structure of buildings is also showing a trend of large-scale and complex, gathering various production and living supplies, and indoor 3D modeling is imperative. With the increasing popularity of laser hardware and mobile scanning equipment, the 3D point cloud generated by 3D laser scanning has become an important way to obtain indoor scene information. How to realize indoor real scene modeling through 3D point cloud has become a hot topic of research today. However, the current modeling methods are generally aimed at outdoor plane structures. There is no good solution for indoor scenes, especially indoor scenes with curved structures. Summary of the invention

[0003] In view of this, the present invention proposes a method for indoor 3D reconstruction that takes into account indoor curved surface structures. The method performs indoor 3D reconstruction based on 3D point clouds, realizes the 3D reconstruction of indoor curved surface structures, especially cylindrical curved surface structures, and can quickly, effectively and accurately realize indoor scene reconstruction, meeting the needs of complex indoor scene navigation and modeling.

[0004] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0005] An indoor three-dimensional reconstruction method taking into account indoor curved surface structures comprises the following steps;

[0006] Step 1: input an indoor 3D point cloud, and segment the indoor 3D point cloud into an indoor plane supervoxel set S and an indoor surface point set C based on a supervoxel segmentation method;

[0007] Step 2: For the indoor plane supervoxel set S, fit the plane model based on the improved supervoxel random sampling consistency method to construct a plane model set ;

[0008] Step 3: Taking each surface point in the indoor surface point set C as a unit, and cylinder and sphere as geometric simplex, the surface model is fitted based on the random sampling consistency method to construct a surface model set ;

[0009] Step 4: Assemble the plane models With surface model collection Projecting onto a 2D plane, dividing the 2D plane into a set of two-dimensional units ;

[0010] Step 5: Construct a 2D unit set based on superposition analysis and Markov field Many-to-many mapping to indoor rooms to construct indoor two-dimensional floor plan;

[0011] Step 6: Based on the constrained Delaunay triangulation method, the indoor two-dimensional plan view is constructed into an indoor three-dimensional model to complete the indoor three-dimensional reconstruction taking into account the indoor curved surface structure.

[0012] Furthermore, the specific method of step 2 is:

[0013] Step 2.1, set the angle threshold , distance threshold With the number threshold, construct a plane model set ;

[0014] Step 2.2, for each indoor plane supervoxel in the indoor plane supervoxel set S, calculate its curvature value, and sort all indoor plane supervoxels in ascending order of curvature value;

[0015] Step 2.3, select the indoor plane supervoxel with the smallest curvature value, calculate its center point and normal vector, and establish the corresponding plane model based on the center point and normal vector , record the currently selected indoor plane supervoxel as the plane model A restrained supervoxel of;

[0016] Step 2.4, traverse the remaining indoor plane supervoxels, if the selected indoor plane supervoxel is consistent with the plane model The normal vector angle is less than the angle threshold , and in the plane model The distance on the normal vector is less than the distance threshold , then the currently selected indoor plane supervoxel is recorded as the plane model A restrained supervoxel of;

[0017] Step 2.5, if the plane model If the number of introverted supervoxels is greater than the threshold, the current plane model is retained. , otherwise delete the current plane model ;

[0018] Step 2.6, for the indoor plane supervoxels that do not belong to the inward supervoxels of any plane model, repeat steps 2.3 to 2.5 until all indoor plane supervoxels are recorded as inward supervoxels of a corresponding plane model;

[0019] Step 2.7, store the plane model with the largest number of introverted supervoxels into the plane model set PS, and delete the introverted supervoxels corresponding to the plane model from the indoor plane supervoxel set S; delete the remaining plane models;

[0020] Step 2.8, repeat steps 2.2 to 2.7 until the number of plane models in the plane model set PS stops increasing.

[0021] Furthermore, the specific method of step 4 is:

[0022] Step 4.1, forming a 2D boundary range according to the maximum and minimum coordinates of the indoor 3D point cloud;

[0023] Step 4.2, project the plane model set PS onto the 2D plane, extend each line segment into a straight line, and select similar straight lines therein for deletion; specifically, select similar straight lines therein for deletion and perform the following steps:

[0024] Step 4.2.1, for straight line With straight line , substitute into the following formula to calculate:

[0025] ;

[0026] ;

[0027] in, Indicates modulus, and Straight lines With straight line The vector direction of represents vector dot product, arccos represents the inverse cosine function;

[0028] Step 4.2.2, if both of the above formulas are true, then determine the straight line With straight line are similar lines, and one of the lines is selected to be deleted, otherwise the line is determined With straight line are non-similar straight lines;

[0029] Step 4.3: Assemble the surface models Project the cylinder set in to the 2D plane and delete the approximate circle;

[0030] Step 4.4: Divide the 2D boundary range into a set of two-dimensional units based on all extension lines and circle lines. .

[0031] Furthermore, the specific method of step 5 is:

[0032] Step 5.1, projecting the indoor three-dimensional point cloud onto a 2D plane, wherein the indoor three-dimensional point cloud is discretized into a two-dimensional projection grid map;

[0033] Step 5.2, segmenting the two-dimensional projection grid image using a morphological room segmentation method to form a two-dimensional room segmentation image;

[0034] Step 5.3: Generate a random point set within the 2D boundary formed in step 4.1. Each random point in the random point set corresponds to a room in the 2D room segmentation map according to its own position, and corresponds to a 2D unit set A two-dimensional unit in

[0035] Step 5.4, based on the random point set, calculate the two-dimensional unit Corresponding to room Probability :

[0036] ;

[0037] in, A two-dimensional unit set The i-th two-dimensional unit in, i=1,2,3,…,M, M is the set of two-dimensional units The number of 2D cells in ; represents the t-th room in the two-dimensional room partitioning graph, t=1,2,3,…,K, K is the number of rooms in the two-dimensional room partitioning graph; represents the number of random points in the random point set that correspond to the i-th two-dimensional unit and the t-th room; Represents the number of random points corresponding to the i-th two-dimensional unit in the random point set; Two-dimensional unit The preliminary corresponding room;

[0038] Step 5.5, use the α expansion optimization method to minimize the following formula and obtain the final M-row K-column matrix , and then get the two-dimensional unit set The final corresponding room for each 2D cell in :

[0039] ;

[0040] Where x is the optimization variable, and x is a matrix with M rows and K columns. Each row element in the matrix x contains one 1 and K-1 0s. represents the element in the i-th row and t-th column of the matrix x, Represents a two-dimensional cell The area of ​​Nei is a two-dimensional unit set The set of two-to-two adjacent two-dimensional unit pairs in , is the number of two-dimensional unit pairs in the set of two-to-two adjacent two-dimensional unit pairs, and is a pair of two-to-two adjacent two-dimensional units in the set Nei of two-to-two adjacent two-dimensional unit pairs, j,q=1,2,3,…,M, and Two-dimensional unit and The initial corresponding room, if ,but =1, =0, otherwise =0, =1; Represents a two-dimensional cell and The walls between Indicates the wall Area in 2D plane;

[0041] The final optimization variable x is a matrix with M rows and K columns. , if the matrix The value of the element in the i-th row and t-th column of is 1, which means that the two-dimensional unit Finally, it corresponds to the tth room in the two-dimensional room segmentation map;

[0042] Step 5.6, the two-dimensional units corresponding to the same room Mark the 2D units with the same color, corresponding to different rooms Mark different colors and output the indoor 2D floor plan.

[0043] Furthermore, the specific method of step 6 is:

[0044] Step 6.1, using the indoor two-dimensional plan generated in step 5.6, combined with the indoor three-dimensional point cloud in each room area, the height of each room is calculated through the point cloud elevation histogram in the vertical direction, and the location of the door is marked on the wall between adjacent rooms according to the indoor two-dimensional plan;

[0045] Step 6.2, use the constrained Delaunay triangulation method to triangulate the indoor ceiling, walls, doors and floors, construct the final indoor 3D model, and output it in the form of a vector mesh to complete the indoor 3D reconstruction that takes into account the indoor curved structure.

[0046] Due to the adoption of the above technical solution, the beneficial effects of the present invention compared with the prior art are:

[0047] 1. The present invention performs indoor 3D reconstruction based on 3D point cloud, realizes the 3D reconstruction of indoor curved surface structures, especially cylindrical curved surface structures, and improves the 3D modeling scheme in the field of indoor modeling.

[0048] 2. Based on the 3D-2D projection law, the present invention regards indoor scene reconstruction as a many-to-many mapping between indoor units and indoor room labels, and solves the problem through the α expansion optimization algorithm. An indoor scene reconstruction framework based on indoor three-dimensional point cloud is constructed, which can realize indoor scene reconstruction quickly, effectively and with high precision. The invention is also applicable to indoor positioning and indoor mapping. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 Schematic diagram of the principle of an indoor 3D reconstruction method taking into account indoor curved surface structures in an embodiment of the present invention.

[0050] Figure 2 Schematic diagram of an indoor three-dimensional point cloud input in an embodiment of the present invention.

[0051] Figure 3 This is a schematic diagram of an indoor 3D reconstruction result output by an embodiment of the present invention. DETAILED DESCRIPTION

[0052] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0053] A method for indoor 3D reconstruction taking into account indoor curved surface structures, such as Figure 1 As shown, the following steps are included;

[0054] Step 1: input an indoor 3D point cloud, and segment the indoor 3D point cloud into an indoor plane supervoxel set S and an indoor surface point set C based on a supervoxel segmentation method;

[0055] Specifically, the supervoxel segmentation method used in this embodiment refers to Su Fei, et al. Boundary-aware supervoxel segmentation for indoor 3D point clouds. IEEE Access. 2023,11, 73478-73489; and Figure 1 The input data in Figure 2 As shown, there are indoor three-dimensional point clouds of two embodiments;

[0056] Specifically, the supervoxel segmentation method includes the following steps: 1) judging the edge area of ​​the 3D point cloud and removing the edge point cloud by the change of the normal vector of the 3D point cloud; 2) , cluster the remaining point clouds to form a point cloud cluster set, each point cloud cluster in the set consists of multiple points; 3) classify the point cloud cluster set into an indoor plane supervoxel set S and an indoor surface point set C according to the curvature.

[0057] Step 2: For the indoor plane supervoxel set S, fit the plane model based on the improved supervoxel random sampling consistency method to construct a plane model set ;

[0058] Specifically, the improved supervoxel random sampling consistent plane fitting method includes the following steps: 1) sorting the plane supervoxel set S according to the supervoxel curvature, selecting the supervoxel with the smallest curvature, and defining a plane model according to the center point and normal vector direction of the supervoxel. 2) calculating the distance between all supervoxels and the plane, and the distance is less than the threshold is the introverted supervoxel of the model; 3) Among the non-introverted supervoxels, select the supervoxel with the second smallest curvature, repeat steps 1 and 2 until the model with the most introverted supervoxels is selected, record the plane, and delete the introverted supervoxel from S; 4) Repeat 1-3 until all supervoxels have a model, and the plane model set is completed Build.

[0059] Step 3: Taking each surface point in the indoor surface point set C as a unit, and cylinder and sphere as geometric simplex, the surface model is fitted based on the random sampling consistency method to construct a surface model set ;

[0060] Specifically, the random sampling consistency method used in this embodiment refers to Schnabel, R.; Wahl, R.; Klein, R. Efficient RANSAC for point-cloud shape detection. Computer GraphicsForum. 2007, 26(2), pp. 214-226. doi: 10.1111 / j.1467-8659.2007.01016.x.

[0061] Specifically, the surface fitting method based on random sampling consistency includes the following steps: 1) surface point set , select multiple points at random and fit a spherical surface (or cylindrical surface) according to the least squares method; 2) calculate the distance between all points and the surface, and the distance is less than the threshold is the inward point of the surface; 3) Repeat steps 1 and 2 among the non-inward points until the spherical surface (or cylindrical surface) model with the most inward points is selected, record the surface, and add the inward points from 4) Repeat 1-3 until each point has a corresponding model or the number of remaining points is less than 1000, and the model set is completed. Build.

[0062] Step 4: Assemble the plane models With surface model collection Projecting onto a 2D plane, dividing the 2D plane into a set of two-dimensional units ;

[0063] Step 5: Construct a 2D unit set based on superposition analysis and Markov field Many-to-many mapping to indoor rooms to construct indoor two-dimensional floor plan;

[0064] Step 6: Based on the constrained Delaunay triangulation method, the indoor two-dimensional plan view is constructed into an indoor three-dimensional model to complete the indoor three-dimensional reconstruction taking into account the indoor curved surface structure.

[0065] Furthermore, the specific method of step 2 is:

[0066] Step 2.1, set the angle threshold , distance threshold With the number threshold, construct a plane model set ;

[0067] Step 2.2, for each indoor plane supervoxel in the indoor plane supervoxel set S, calculate its curvature value, and sort all indoor plane supervoxels in ascending order of curvature value;

[0068] Step 2.3, select the indoor plane supervoxel with the smallest curvature value, calculate its center point and normal vector, and establish the corresponding plane model based on the center point and normal vector , record the currently selected indoor plane supervoxel as the plane model A restrained supervoxel of;

[0069] Step 2.4, traverse the remaining indoor plane supervoxels, if the selected indoor plane supervoxel is consistent with the plane model The normal vector angle is less than the angle threshold , and in the plane model The distance on the normal vector is less than the distance threshold , then the currently selected indoor plane supervoxel is recorded as the plane model A restrained supervoxel of;

[0070] Step 2.5, if the plane model If the number of introverted supervoxels is greater than the threshold, the current plane model is retained. , otherwise delete the current plane model ;

[0071] Step 2.6, for the indoor plane supervoxels that do not belong to the inward supervoxels of any plane model, repeat steps 2.3 to 2.5 until all indoor plane supervoxels are recorded as inward supervoxels of a corresponding plane model;

[0072] Step 2.7, store the plane model with the largest number of introverted supervoxels into the plane model set PS, and delete the introverted supervoxels corresponding to the plane model from the indoor plane supervoxel set S; delete the remaining plane models;

[0073] Step 2.8, repeat steps 2.2 to 2.7 until the number of plane models in the plane model set PS stops increasing.

[0074] Furthermore, the specific method of step 4 is:

[0075] Step 4.1, forming a 2D boundary range according to the maximum and minimum coordinates of the indoor 3D point cloud;

[0076] Step 4.2, project the plane model set PS onto the 2D plane, extend each line segment into a straight line, and select similar straight lines therein for deletion; specifically, select similar straight lines therein for deletion and perform the following steps:

[0077] Step 4.2.1, for straight line With straight line , substitute into the following formula to calculate:

[0078] ;

[0079] ;

[0080] in, Indicates modulus, and Straight lines With straight line The vector direction of represents vector dot product, arccos represents the inverse cosine function;

[0081] Step 4.2.2, if both of the above formulas are true, then determine the straight line With straight line are similar lines, and one of the lines is selected to be deleted, otherwise the line is determined With straight line are non-similar straight lines;

[0082] Step 4.3: Assemble the surface models Project the cylinder set in to the 2D plane and delete the approximate circle;

[0083] Step 4.4: Divide the 2D boundary range into a set of two-dimensional units based on all extension lines and circle lines. .

[0084] Furthermore, the specific method of step 5 is:

[0085] Step 5.1, projecting the indoor three-dimensional point cloud onto a 2D plane, wherein the indoor three-dimensional point cloud is discretized into a two-dimensional projection grid map;

[0086] Specifically, in this embodiment, if a certain grid image does not contain any point, the grid value is 0, otherwise the grid value is 1, forming a two-dimensional projection grid image;

[0087] Step 5.2, segmenting the two-dimensional projection grid image using a morphological room segmentation method to form a two-dimensional room segmentation image;

[0088] Step 5.3: Generate a random point set within the 2D boundary formed in step 4.1. Each random point in the random point set corresponds to a room in the 2D room segmentation map according to its own position, and corresponds to a 2D unit set A two-dimensional unit in

[0089] Step 5.4, based on the random point set, calculate the two-dimensional unit Corresponding to the room Probability :

[0090] ;

[0091] in, A two-dimensional unit set The i-th two-dimensional unit in, i=1,2,3,…,M, M is the set of two-dimensional units The number of 2D cells in ; represents the t-th room in the two-dimensional room partitioning graph, t=1,2,3,…,K, K is the number of rooms in the two-dimensional room partitioning graph; represents the number of random points in the random point set that correspond to the i-th two-dimensional unit and the t-th room; Represents the number of random points corresponding to the i-th two-dimensional unit in the random point set; Two-dimensional unit The preliminary corresponding room;

[0092] Step 5.5, use the α expansion optimization method to minimize the following formula and obtain the final M-row K-column matrix , and then get the two-dimensional unit set The final corresponding room for each 2D cell in :

[0093] ;

[0094] Where x is the optimization variable, and x is a matrix with M rows and K columns. Each row element in the matrix x contains one 1 and K-1 0s. represents the element in the i-th row and t-th column of the matrix x, Represents a two-dimensional cell The area of ​​Nei is a two-dimensional unit set The set of two-to-two adjacent two-dimensional unit pairs in , is the number of two-dimensional unit pairs in the set of two-to-two adjacent two-dimensional unit pairs Nei, and is a pair of two-to-two adjacent two-dimensional units in the set of two-to-two adjacent two-dimensional unit pairs Nei, j, q = 1, 2, 3, ..., M, and Two-dimensional unit and The initial corresponding room, if ,but =1, =0, otherwise =0, =1; Represents a two-dimensional cell and The walls between Indicates the wall Area in 2D plane;

[0095] The final optimization variable x is a matrix with M rows and K columns. , if the matrix The value of the element in the i-th row and t-th column of is 1, which means that the two-dimensional unit Finally, it corresponds to the tth room in the two-dimensional room segmentation map;

[0096] Step 5.6, the two-dimensional units corresponding to the same room Mark the 2D units with the same color, corresponding to different rooms Mark different colors and output the indoor 2D floor plan.

[0097] Furthermore, the specific method of step 6 is:

[0098] Step 6.1, using the indoor two-dimensional plan generated in step 5.6, combined with the indoor three-dimensional point cloud in each room area, the height of each room is calculated through the point cloud elevation histogram in the vertical direction, and the location of the door is marked on the wall between adjacent rooms according to the indoor two-dimensional plan;

[0099] Step 6.2, using the constrained Delaunay triangulation method, triangulate the indoor ceiling, walls, doors and floors to construct the final indoor 3D model and output it in the form of a vector mesh. Figure 3 As shown, the indoor three-dimensional reconstruction taking into account the indoor curved surface structure is completed.

[0100] In summary, the present invention performs indoor 3D reconstruction based on 3D point clouds, realizes the 3D reconstruction of indoor curved surface structures, especially cylindrical curved surface structures, and can quickly, effectively and accurately realize indoor scene reconstruction to meet the needs of complex indoor scene navigation and modeling.

[0101] Those skilled in the art will appreciate that the embodiments described are intended to help readers understand the principles of the present invention, and should be understood that the scope of protection of the present invention is not limited to the embodiments described. For those skilled in the art, the present invention may have various changes and variations. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A method for indoor 3D reconstruction taking into account indoor curved surface structures, characterized in that: The steps include: Step 1: input an indoor 3D point cloud, and segment the indoor 3D point cloud into an indoor plane supervoxel set S and an indoor surface point set C based on a supervoxel segmentation method; Step 2: For the indoor plane supervoxel set S, fit the plane model based on the improved supervoxel random sampling consistency method to construct a plane model set ; Step 3: Set indoor surface points Each surface point in the model is taken as a unit, and the cylinder and sphere are taken as geometric shapes. The surface model is fitted based on the random sampling consistency method to construct a surface model set. ; Step 4: Assemble the plane models With surface model collection Projecting onto a 2D plane, dividing the 2D plane into a set of two-dimensional units : Step 4.1, forming a 2D boundary range according to the maximum and minimum coordinates of the indoor 3D point cloud; Step 4.2, project the plane model set PS onto the 2D plane, extend each line segment into a straight line, and select similar straight lines therein for deletion; specifically, select similar straight lines therein for deletion and perform the following steps: Step 4.2.1, for straight line With straight line , substitute into the following formula to calculate: ； ； in, Indicates modulus, and Straight lines With straight line The vector direction of represents vector dot product, arccos represents the inverse cosine function; is the angle threshold, is the distance threshold; Step 4.2.2, if both of the above formulas are true, then determine the straight line With straight line are similar lines, and one of the lines is selected to be deleted, otherwise the line is determined With straight line are non-similar straight lines; Step 4.3: Assemble the surface models Project the cylinder set in to the 2D plane and delete the approximate circle; Step 4.4: Divide the 2D boundary range into a set of two-dimensional units based on all extension lines and circle lines. ; Step 5: Construct a 2D unit set based on superposition analysis and Markov field Many-to-many mapping to indoor rooms to construct an indoor two-dimensional floor plan: Step 5.1, projecting the indoor three-dimensional point cloud onto a 2D plane, wherein the indoor three-dimensional point cloud is discretized into a two-dimensional projection grid map; Step 5.2, segmenting the two-dimensional projection grid image using a morphological room segmentation method to form a two-dimensional room segmentation image; Step 5.3: Generate a random point set within the 2D boundary formed in step 4.

1. Each random point in the random point set corresponds to a room in the 2D room segmentation map according to its own position, and corresponds to a 2D unit set A two-dimensional unit in Step 5.4, based on the random point set, calculate the two-dimensional unit Corresponding to room Probability : ； in, A two-dimensional unit set The i-th two-dimensional unit in, i=1,2,3,......,M, M is the set of two-dimensional units The number of 2D cells in ; represents the t-th room in the two-dimensional room partitioning graph, t=1,2,3,......,K, K is the number of rooms in the two-dimensional room partitioning graph; represents the number of random points in the random point set that correspond to the i-th two-dimensional unit and the t-th room; Represents the number of random points corresponding to the i-th two-dimensional unit in the random point set; Two-dimensional unit The preliminary corresponding room; Step 5.5, use the α expansion optimization method to minimize the following formula and obtain the final M-row K-column matrix , and then get the two-dimensional unit set The final corresponding room for each 2D cell in : ； Where x is the optimization variable, and x is a matrix with M rows and K columns. Each row element in the matrix x contains one 1 and K-1 0s. represents the element in the i-th row and t-th column of the matrix x, Represents a two-dimensional cell The area of ​​Nei is a two-dimensional unit set The set of two-to-two adjacent two-dimensional unit pairs in , is the number of two-dimensional unit pairs in the set of two-to-two adjacent two-dimensional unit pairs, and is a pair of two-to-two adjacent two-dimensional cells in the set of two-to-two adjacent two-dimensional cell pairs Nei, j, q = 1, 2, 3, ..., M, and Two-dimensional unit and The initial corresponding room, if ,but =1, =0, otherwise =0, =1; Represents a two-dimensional cell and The walls between Indicates the wall Area in 2D plane; The final optimization variable x is a matrix with M rows and K columns. , if the matrix The value of the element in the i-th row and t-th column of is 1, which means that the two-dimensional unit Finally, it corresponds to the tth room in the two-dimensional room segmentation map; Step 5.6, the two-dimensional units corresponding to the same room Mark the 2D units with the same color, corresponding to different rooms Mark different colors and output the indoor two-dimensional plan; Step 6: Based on the constrained Delaunay triangulation method, the indoor two-dimensional plan view is constructed into an indoor three-dimensional model to complete the indoor three-dimensional reconstruction taking into account the indoor curved surface structure.

2. The indoor 3D reconstruction method taking into account indoor curved surface structures according to claim 1, characterized in that: The specific method of step 2 is: Step 2.1, set the angle threshold , distance threshold With the number threshold, construct a plane model set ; Step 2.2, for each indoor plane supervoxel in the indoor plane supervoxel set S, calculate its curvature value, and sort all indoor plane supervoxels in ascending order of curvature value; Step 2.3, select the indoor plane supervoxel with the smallest curvature value, calculate its center point and normal vector, and establish the corresponding plane model based on the center point and normal vector , record the currently selected indoor plane supervoxel as the plane model A restrained supervoxel of; Step 2.4, traverse the remaining indoor plane supervoxels, if the selected indoor plane supervoxel is consistent with the plane model The normal vector angle is less than the angle threshold , and in the plane model The distance on the normal vector is less than the distance threshold , then the currently selected indoor plane supervoxel is recorded as the plane model A restrained supervoxel of; Step 2.5, if the plane model If the number of introverted supervoxels is greater than the threshold, the current plane model is retained. , otherwise delete the current plane model ; Step 2.6, for the indoor plane supervoxels that do not belong to the inward supervoxels of any plane model, repeat steps 2.3 to 2.5 until all indoor plane supervoxels are recorded as inward supervoxels of a corresponding plane model; Step 2.7, storing the plane model with the largest number of introverted supervoxels into the plane model set PS, and deleting the introverted supervoxels corresponding to the plane model from the indoor plane supervoxel set S; Delete the remaining plane models; Step 2.8, repeat steps 2.2 to 2.7 until the number of plane models in the plane model set PS stops increasing.

3. The indoor 3D reconstruction method taking into account indoor curved surface structures according to claim 2, characterized in that: The specific method of step 6 is: Step 6.1, using the indoor two-dimensional plan generated in step 5.6, combined with the indoor three-dimensional point cloud in each room area, the height of each room is calculated through the point cloud elevation histogram in the vertical direction, and the location of the door is marked on the wall between adjacent rooms according to the indoor two-dimensional plan; Step 6.2, use the constrained Delaunay triangulation method to triangulate the indoor ceiling, walls, doors and floors, construct the final indoor 3D model, and output it in the form of a vector mesh to complete the indoor 3D reconstruction that takes into account the indoor curved structure.

Images