A surface simplification method for house models generated by oblique photography

Abstract

The invention discloses a surface simplification method for a house model generated by oblique photography. For a point cloud extracted according to an original house model, the following steps are taken: step 1: removing discrete points; step 2: removing large horizontal planes; step 3: extracting planes; and step 4: merging planes.

Description

Technical Field

[0001] The invention belongs to the field of geological surveying and mapping, and in particular, to a surface simplification method for a house model generated by oblique photography. Background Art

[0002] In the process of real-life 3D modeling using oblique photogrammetry technology, buildings need to be singulated. Singulated buildings mean that you want to obtain a certain building separately and separate it from other buildings, and then perform semantic analysis on the single building. With manual modeling, objects that need to be managed separately will naturally be made into separate models and separated from other objects. However, the data obtained by oblique photography are connected together, and it is necessary to separate a single building. Therefore, automatically singulating buildings is a problem that must be solved. In the singulation process, the point cloud extracted based on the original house model needs to be simplified into several patches that are approximately vertical to the ground.

[0003] The main benefit of face simplification is that it can achieve semi-automatic and automatic vector collection. The data obtained by oblique photography is connected, and the walls of the buildings in the model (the best effect) are bumpy. The previous collection of house outline information required the collector to collect one face at a time based on the 3D model. This method is a step in a work chain, which is as follows:

[0004] 1. Manually specify a point on the house model, use the one-click model monomerization method to obtain the triangulated network of the house model and its surroundings, and then generate a point cloud through the triangulated network;

[0005] 2. Use this method to generate several patches that are approximately perpendicular to the ground based on the point cloud;

[0006] 3. Generate a simple model based on the patch;

[0007] 4. Calculate the house outline based on the model.

[0008] The purpose of this simplification is to improve the speed and accuracy of model generation in step 3 and realize semi-automatic collection. The collector only needs to specify a point on the house model to quickly obtain the house outline. Furthermore, the computer can realize the manual designation steps in step 1 to realize automatic collection. Summary of the invention

[0009] In view of the above problems and shortcomings, the present invention proposes a rapid singularization technology based on model matching.

[0010] In order to achieve the above-mentioned invention object, this paper proposes the following technical solutions:

[0011] A surface simplification method for a house model generated by oblique photography, taking the following steps for a point cloud extracted from an original house model:

[0012] Step 1: Remove discrete points;

[0013] According to the prominent feature that there are fewer points near discrete points, the point cloud is clustered by the Euclidean distance between points. The points contained in the class with less points than the threshold are considered as discrete points, and the discrete points are removed from the point cloud;

[0014] Step 2: Remove large horizontal planes;

[0015] Remove points that form large horizontal planes from the point cloud;

[0016] Step 3: Extract the plane;

[0017] Use sampling consistency segmentation method to divide the point cloud into multiple groups, extract planes for each group, and remove inappropriate planes;

[0018] Step 4: Merge the planes;

[0019] When extracting a plane, if the same plane is divided into several parts, the same faces are merged.

[0020] In the above-mentioned method for simplifying the surface of a house model generated by oblique photography, the step 1 specifically comprises:

[0021] Given the point cloud C extracted from the original house model, the discrete point number threshold Nt, the search radius R, and the point set Pr to be removed, Pr is initially empty; traverse the point cloud C and perform the following processing on each point P:

[0022] Step 11: Skip if P has been processed;

[0023] Step 12: Construct the cluster point set Pc and the point set to be processed Pw, and put P into Pw;

[0024] Step 121: Pw is empty and the step 12 is ended;

[0025] Step 122: Take the first point p of Pw and remove p from Pw. If p has been processed, start from step 121;

[0026] Step 123: record p as processed, put p into Pc, search for all points Ps within the radius R of p, traverse Ps and put the unmarked processed points into Pw;

[0027] Step 124: Start from step 121;

[0028] Step 13: If the number of points in Pc is less than Nt, add the points in Pc to Pr; if Pr is not empty, remove the points in Pr from C.

[0029] In the above-mentioned method for simplifying the surface of a house model generated by oblique photography, the step 2 is specifically as follows:

[0030] The conditions for a large horizontal plane are that the angle between the plane normal and the Z axis is less than 5 degrees and the number of points is greater than N;

[0031] Step 21: Perform consistency segmentation on the point cloud C to satisfy the large horizontal plane, and obtain the point set Pp of the largest surface;

[0032] Step 22: If Pp is not empty, remove Pp from C and execute step 21 again.

[0033] In the above-mentioned method for simplifying the surface of a house model generated by oblique photography, the step 3 is specifically as follows:

[0034] Given the minimum number of points N on a valid plane, save the set F of extracted faces, save the linked list Rl of failed extractions, the number of points in the initial point cloud C is Cn, and the ending coefficient Ss;

[0035] Step 31: If the number of C is less than Ss*Cn, jump to step 8;

[0036] Step 32: If R1 is greater than 10 and the number of "failures" in the last ten is greater than 7, jump to step 8;

[0037] Step 33: Perform consistent segmentation on C, take the point set Pp and normal n of the largest face, if Pp is empty, jump to step 8;

[0038] Step 34: Remove Pp from C. If the number of Pp is less than N, add "failure" to the end of Rl and jump to step 1.

[0039] Step 35: Perform Euclidean distance classification on Pp and divide it into point set array Vpo, and add "success" at the end of Rl;

[0040] Step 36: Traverse Vpo, Poi is the i-th point set in Vpo, and perform the following processing on Poi:

[0041] Step 361: Fitting the plane f and its normal nf according to Poi;

[0042] Step 362: If the angle between nf and n is less than 10 degrees and the angle between nf and the Z axis is greater than 86 degrees, then put f into F, otherwise add Poi to C;

[0043] Step 37: Jump to step 31;

[0044] Step 38: End;

[0045] In the above-mentioned method for simplifying the surface of a house model generated by oblique photography, the step 4 is specifically as follows:

[0046] Step 41: Given the set F of extracted faces, record the set M of index pairs of mergeable faces; traverse the set F from the beginning, and Fi is the i-th element in F;

[0047] Step 411: Take the normal Ni and centroid Ci of Fi;

[0048] Step 412: traverse the set F starting from i+1, where Fj is the jth element in F;

[0049] Step 4121: Take the normal Nj and centroid Cj of Fj;

[0050] Step 4122: vector d=Cj-Ci;

[0051] Step 4123: If the angle between the normals of Ni and Nj is less than 5 degrees and the maximum length of the projection of d on Ni and Nj is less than 0.1, then add the index pair (i, j) to M;

[0052] Step 42: merge the index pair set M of mergeable faces into a set Mv of mergeable index groups; record the merged faces as Mf; traverse Mv, where m is an element in Mv;

[0053] Step 421: Add the point sets corresponding to the faces of all indices in m to the point set Ct;

[0054] Step 422: Fit a new surface according to Ct and put it into Mf;

[0055] Step 423: Output Mf.

[0056] Most of the existing surface simplification algorithms are based on point cloud clustering, and then extracting surface objects from the clustering results according to a set threshold. The biggest problem with the surfaces extracted by this method is that the accuracy cannot meet the needs of vector extraction. Using the method described in this patent, the point cloud can be divided and conquered. After separating the noise and ground points, most of the remaining points are wall points. Then, according to the wall normal consistency condition, the point cloud is further separated to separate different vertical surfaces. In this process, strict plane distance conditions can be set to ensure that the accuracy of the remaining points meets the requirements of vector mapping.

[0057] This method also has the advantage of short calculation time (processing time for 10,000 points is less than 1 second).

[0058] Compared with the conventional method of directly generating models from point clouds, the calculation time is longer (tens to dozens of minutes) and the accuracy is not good. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 What is shown are the discrete points in step 1. The points in the box in the figure are discrete points;

[0060] Figure 2 What is shown is the large horizontal plane mentioned in step 2, which is usually the bottom or the roof;

[0061] Figure 3 The diagram shows the planes before merging. The faces with different shades (displayed in different colors in the color map) are different faces. Some of the same faces are shown with different shades before merging.

[0062] Figure 4 The figure shows the schematic diagram of the plane after merging. The faces with different shades of shadows (displayed in different colors in the color map) are different faces. Figure 3 The faces shown as different shades of shadows in the figure become the same shade of shadows after being merged, that is, they become the same face after being merged;

[0063] Figure 5 The flowchart for step 1 is shown;

[0064] Figure 6 The flowchart for step 2 is shown;

[0065] Figure 7 The flowchart for step 3 is shown;

[0066] Figure 8 The flowchart for step 4 is shown; Specific implementation methods

[0067] The specific implementation of the present disclosure is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described herein is only used to illustrate and explain the present disclosure, and is not used to limit the present disclosure.

[0068] The present invention proposes a surface simplification method for a house model generated by oblique photography, and the following steps are taken for a point cloud extracted according to the original house model:

[0069] Step 1: Remove discrete points;

[0070] According to the prominent feature that there are fewer points near discrete points, the point cloud is clustered by the Euclidean distance between points. The points contained in the class with less points than the threshold are considered as discrete points, and the discrete points are removed from the point cloud;

[0071] Step 2: Remove large horizontal planes;

[0072] Remove points that form large horizontal planes from the point cloud;

[0073] Step 3: Extract the plane;

[0074] Use sampling consistency segmentation method to divide the point cloud into multiple groups, extract planes for each group, and remove inappropriate planes;

[0075] Step 4: Merge the planes;

[0076] When extracting a plane, if the same plane is divided into several parts, the same faces are merged.

[0077] In the above-mentioned method for simplifying the surface of a house model generated by oblique photography, the step 1 specifically comprises:

[0078] Given the point cloud C extracted from the original house model, the discrete point number threshold Nt, the search radius R, and the point set Pr to be removed, Pr is initially empty; traverse the point cloud C and perform the following processing on each point P:

[0079] Step 11: Skip if P has been processed;

[0080] Step 12: Construct the cluster point set Pc and the point set to be processed Pw, and put P into Pw;

[0081] Step 121: Pw is empty and the step 12 is ended;

[0082] Step 122: Take the first point p of Pw and remove p from Pw. If p has been processed, start from step 121;

[0083] Step 123: record p as processed, put p into Pc, search for all points Ps within the radius R of p, traverse Ps and put the unmarked processed points into Pw;

[0084] Step 124: Start from step 121;

[0085] Step 13: If the number of points in Pc is less than Nt, add the points in Pc to Pr; if Pr is not empty, remove the points in Pr from C.

[0086] In the above-mentioned method for simplifying the surface of a house model generated by oblique photography, the step 2 is specifically as follows:

[0087] The conditions for a large horizontal plane are that the angle between the plane normal and the Z axis is less than 5 degrees and the number of points is greater than N;

[0088] Step 21: Perform consistency segmentation on the point cloud C to satisfy the large horizontal plane, and obtain the point set Pp of the largest surface;

[0089] Step 22: If Pp is not empty, remove Pp from C and execute step 21 again.

[0090] In the above-mentioned method for simplifying the surface of a house model generated by oblique photography, the step 3 is specifically as follows:

[0091] Given the minimum number of points N on a valid plane, save the set F of extracted faces, save the linked list Rl of failed extractions, the number of points in the initial point cloud C is Cn, and the ending coefficient Ss;

[0092] Step 31: If the number of C is less than Ss*Cn, jump to step 8;

[0093] Step 32: If R1 is greater than 10 and the number of "failures" in the last ten is greater than 7, jump to step 8;

[0094] Step 33: Perform consistent segmentation on C, take the point set Pp and normal n of the largest face, if Pp is empty, jump to step 8;

[0095] Step 34: Remove Pp from C. If the number of Pp is less than N, add "failure" to the end of Rl and jump to step 1.

[0096] Step 35: Perform Euclidean distance classification on Pp and divide it into point set array Vpo, and add "success" at the end of Rl;

[0097] Step 36: Traverse Vpo, Poi is the i-th point set in Vpo, and perform the following processing on Poi:

[0098] Step 361: Fitting the plane f and its normal nf according to Poi;

[0099] Step 362: If the angle between nf and n is less than 10 degrees and the angle between nf and the Z axis is greater than 86 degrees, then put f into F, otherwise add Poi to C;

[0100] Step 37: Jump to step 31;

[0101] Step 38: End;

[0102] In the above-mentioned method for simplifying the surface of a house model generated by oblique photography, the step 4 is specifically as follows:

[0103] Step 41: Given the set F of extracted faces, record the set M of index pairs of mergeable faces; traverse the set F from the beginning, and Fi is the i-th element in F;

[0104] Step 411: Take the normal Ni and centroid Ci of Fi;

[0105] Step 412: traverse the set F starting from i+1, where Fj is the jth element in F;

[0106] Step 4121: Take the normal Nj and centroid Cj of Fj;

[0107] Step 4122: vector d=Cj-Ci;

[0108] Step 4123: If the angle between the normals of Ni and Nj is less than 5 degrees and the maximum length of the projection of d on Ni and Nj is less than 0.1, then add the index pair (i, j) to M;

[0109] Step 42: merge the index pair set M of mergeable faces into a set Mv of mergeable index groups; record the merged faces as Mf; traverse Mv, where m is an element in Mv;

[0110] Step 421: Add the point sets corresponding to the faces of all indices in m to the point set Ct;

[0111] Step 422: Fit a new surface according to Ct and put it into Mf;

[0112] Step 423: Output Mf. The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A surface simplification method for a house model generated by oblique photography, characterized in that: For the point cloud extracted from the original house model, take the following steps: Step 1: Remove discrete points; According to the prominent feature that there are fewer points near discrete points, the point cloud is clustered by the Euclidean distance between points. The points contained in the class with less points than the threshold are considered as discrete points, and the discrete points are removed from the point cloud; Step 2: Remove large horizontal planes; Remove points that form large horizontal planes from the point cloud; Step 3: Extract the plane; Use sampling consistency segmentation method to divide the point cloud into multiple groups, extract planes for each group, and remove inappropriate planes; The step 3 is specifically as follows: Given the minimum number of points N on a valid plane, save the set F of extracted faces, save the linked list Rl of failed extractions, the number of points in the initial point cloud C is Cn, and the ending coefficient Ss; Step 31: If the number of C is less than Ss*Cn, jump to step 38; Step 32: If R1 is greater than 10 and the number of "failures" in the last ten is greater than 7, jump to step 38; Step 33: Perform consistent segmentation on C, take the point set Pp and normal n of the largest face, if Pp is empty, jump to step 38; Step 34: Remove Pp from C. If the number of Pp is less than N, add "failure" to the end of R1 and jump to step 31. Step 35: Perform Euclidean distance classification on Pp and divide it into point set array Vpo, and add "success" at the end of Rl; Step 36: Traverse Vpo, Poi is the i-th point set in Vpo, and perform the following processing on Poi: Step 361: Fitting the plane f and its normal nf according to Poi; Step 362: If the angle between nf and n is less than 10 degrees and the angle between nf and the Z axis is greater than 86 degrees, then put f into F, otherwise add Poi to C; Step 37: Jump to step 31; Step 38: End; Step 4: Merge the planes; When extracting a plane, if the same plane is divided into several parts, the same faces are merged.

2. The surface simplification method of a house model generated by oblique photography according to claim 1, characterized in that: The step 1 is specifically as follows: Given the point cloud C extracted from the original house model, the discrete point number threshold Nt, the search radius R, and the point set Pr to be removed, Pr is initially empty; traverse the point cloud C and perform the following processing on each point P: Step 11: Skip if P has been processed; Step 12: Construct the cluster point set Pc and the point set to be processed Pw, and put P into Pw; Step 121: Pw is empty and the step 12 is ended; Step 122: Take the first point p of Pw and remove p from Pw. If p has been processed, start from step 121; Step 123: record p as processed, put p into Pc, search for all points Ps within the radius R of p, traverse Ps and put the unmarked processed points into Pw; Step 124: Start from step 121; Step 13: If the number of points in Pc is less than Nt, add the points in Pc to Pr; if Pr is not empty, remove the points in Pr from C.

3. The surface simplification method of a house model generated by oblique photography according to claim 1 or 2, characterized in that: The step 2 is specifically as follows: The conditions for a large horizontal plane are that the angle between the plane normal and the Z axis is less than 5 degrees and the number of points is greater than N; Step 21: Perform consistency segmentation on the point cloud C to satisfy the large horizontal plane, and obtain the point set Pp of the largest surface; Step 22: If Pp is not empty, remove Pp from C and execute step 21 again.

4. The surface simplification method of a house model generated by oblique photography according to claim 1, characterized in that: The step 4 is specifically as follows: Step 41: Given the set F of extracted faces, record the set M of index pairs of mergeable faces; traverse the set F from the beginning, and Fi is the i-th element in F; Step 411: Take the normal Ni and centroid Ci of Fi; Step 412: traverse the set F starting from i+1, where Fj is the jth element in F; Step 4121: Take the normal Nj and centroid Cj of Fj; Step 4122: vector d=Cj-Ci; Step 4123: If the angle between the normals of Ni and Nj is less than 5 degrees and the maximum length of the projection of d on Ni and Nj is less than 0.1, then add the index pair (i, j) to M; Step 42: merge the index pair set M of mergeable faces into a set Mv of mergeable index groups; record the merged faces as Mf; traverse Mv, where m is an element in Mv; Step 421: Add the point sets corresponding to the faces of all indices in m to the point set Ct; Step 422: Fit a new surface according to Ct and put it into Mf; Step 423: Output Mf.

Images