Indoor scene layout estimation method based on structural prior

Abstract

The present invention relates to the field of computer vision technology, and more specifically, to a method for estimating the layout of indoor scenes based on structural priors. The method comprises: step 1: resampling a point cloud model; step 2: extracting planes and segmenting structures; step 3: generating candidate facade segments; step 4: clustering rooms based on an extensible graph; and step 5: optimizing the contour walls based on the results of steps 1 to 4, selecting a subset of the candidate segments, and generating a complete floor plan. The present invention significantly reduces the topological complexity of reconstruction, eliminates the mutual influence of wall candidate segments between different rooms, and thereby improves the efficiency and accuracy of reconstruction.

Description

Technical Field

[0001] The present invention relates to the technical field of computer vision, and more particularly to a method for estimating indoor scene layout based on structure prior. Background Art

[0002] With the deepening of social informatization, the digital transformation of information has become a trend. 3D reconstruction technology, as a digital vision technology that transforms real-world scenes into digital virtual environments, has attracted much attention. Vectorized reconstruction, as a key branch of 3D reconstruction, is crucial for converting complex 3D data into concise and clear vector models. However, traditional 3D surface reconstruction methods focus on generating accurate and dense 3D models, often ignoring structural and semantic information. Furthermore, the incompleteness and noise of point clouds acquired by 3D sensors pose challenges to layout estimation.

[0003] Currently, mainstream scene vectorization modeling methods can be broadly divided into two directions. The first is geometry-based methods, which are typically based on geometric principles and computer vision techniques. These methods use sensors (such as depth cameras and lidar) to acquire geometric information about indoor scenes and then estimate the room layout through geometric reasoning. This may include the detection, segmentation, and reconstruction of walls, furniture, and other objects. The second is end-to-end methods based on deep learning. End-to-end indoor layout estimation methods use architectures such as convolutional neural networks to learn features and layout information from raw input data to reconstruct a vectorized model of the scene. However, these two methods struggle to balance robustness and reconstruction accuracy. Geometric reasoning provides intuitive spatial structure information. For simple scenes, traditional methods are easier to implement and have lower computational overhead. However, in complex indoor scenes, traditional methods struggle to cope with occlusions and missing information. End-to-end methods based on deep learning can learn richer feature representations from large amounts of data, but they cannot guarantee the accuracy of the reconstruction results. They also lack interpretability, making it difficult to understand the network's reasoning process for layout estimation. Summary of the Invention

[0004] In order to solve the above technical difficulties, an embodiment of the present invention provides an indoor scene layout estimation method based on structural prior, the method comprising:

[0005] Step 1: Resample the point cloud model;

[0006] Step 2: Extract plane and segment structure;

[0007] Step 3: Generate candidate face segments;

[0008] Step 4: Cluster the rooms based on the scalable graph;

[0009] Step 5: Based on the results from steps 1 to 4, optimize the contour wall, select a subset from the candidate segments, and generate a complete floor plan.

[0010] Furthermore, the resampling point cloud model method includes:

[0011] The original point cloud model is reconstructed into a mesh model, and the proposed random sampling method is used to sample the surface units of the mesh model;

[0012] In the mesh model patch unit M, a three-dimensional point p is randomly sampled by a resampling method; a vertex is randomly selected as an endpoint O, and two adjacent edges OA and OB are used as basis vectors to resample the patch unit.

[0013] Furthermore, the method for extracting planes and segmenting structures includes:

[0014] In real-world architectural scenes, most can be described by simplified planar structures, such as piecewise planes or curved surfaces. To handle the discreteness of point cloud data and abstract the structural features of the model, this patent proposes a plane fitting method that uses a random sample consistency method to extract planar shape units from the point cloud model and obtain all planes that meet the fitting constraints. The fitting constraints are determined by RANSAC parameters, including the maximum distance from the point to the plane, the normal angle threshold, the minimum number of supporting points per plane, and the maximum probability of missing a plane candidate. The RANSAC parameters can be adjusted according to the quality of the point cloud data and are applicable to various datasets with different densities and noise levels.

[0015] Based on the results of plane fitting, the point cloud model is structured and segmented using plane normals, further achieving structural segmentation of the ceiling, floor, and facade. During the model preprocessing phase, the point cloud model's posture is corrected, allowing the segmentation plane set to be calculated by fitting the plane normals of the ceiling and floor planar shape point clusters and combining them with the coordinate axis direction vectors. For each plane in the plane set, structural segmentation is achieved by performing a dot product of the plane normal vector with the Z-axis direction vector, applying constraints.

[0016] Furthermore, the method for generating candidate face segments includes:

[0017] The Z coordinate includes height information, projects the vertical facade onto the horizontal plane, and focuses on generating a floor plan in 2D space. The plane and facade are extracted from the point cloud model using RANSAC and regularized angle threshold segmentation methods. Based on the floor-wall prior assumption, the plane and facade are ensured to be approximately perpendicular. Each fitted facade is considered part of the wall plane and contains a set of shape point clusters, namely support points. The fitted facade is projected onto the horizontal plane, and the corresponding 2D support points are projected onto the projection line to generate candidate wall segments using the projected facade endpoints. The wall candidate segments are further refined by merging line segments to obtain facade candidate segments.

[0018] The methods for merging line segments include:

[0019] If the minimum distance between two wall segments wi and wj is less than the threshold θ2, and the following conditions are met: the angle between the normal vectors of the two wall segments is less than θ1, and the overlapping length of the projections of the two wall segments is less than θ3 of the shorter line segment, then wi and wj are considered to belong to the same wall segment, wi and wj are merged, and unnecessary segmentation introduced by RANSAC is removed. θ2, θ1 and θ3 are preset thresholds; the method for merging line segments is expressed as follows:

[0020]

[0021] Where, represents the angle between the normal vectors of two wall segments wi and wj, represents the minimum distance between two wall segments wi and wj, Represents the overlapping length of the projections of two wall segments wi and wj.

[0022] Furthermore, the room clustering method based on the scalable graph includes:

[0023] The prior estimation is formalized as a projected line segment labeling problem, where a labeling problem represents a room or outdoor area; a line segment prior labeling method based on scalable graph clustering is used for the labeling problem;

[0024] Line segment prior labeling methods include:

[0025] Regions of mutual visibility among line segments representing planes are clustered, corresponding to rooms of the building. A visibility graph is generated by performing ray casting between the detected facade line segments. Subsequently, the nodes of the visibility graph are clustered using a scalable graph clustering algorithm that uses flow simulation methods to determine natural clusters in the graph.

[0026] A rough occupancy map is generated by passing a unit line segment through each column of the plane, and each occupied pixel is regarded as a patch with the same normal as the corresponding plane; a line segment of length 0.5m is set, pi is the i-th line segment, its center position is ci, and its normal is ni; for each pair pi, pj, i, j, a ray is cast between the points ci+A×ni and cj+A×nj, where A=10cm; if the line segment connecting the two is in the same direction as the normal vector of the two and there is no obstacle between the connecting line segments, the visibility between pi and pj is set to 1, otherwise it is set to 0, that is, a visibility graph is generated, and the nodes of the visibility graph are clustered using a clustering algorithm; the calculated visibility is interpreted as the flow between the node pairs corresponding to the corresponding point patches;

[0027] After obtaining n disjoint patch clusters, define a set of room labels R = {r1, ..., rn}, which will be used in the rest of the reconstruction process; each point of the point cloud is assigned to the room label of the corresponding line segment candidate;

[0028] Furthermore, the method for generating a complete floor plan includes:

[0029] Ray casting is used to test the visibility between points on the detected surface, thereby constructing a visibility graph. The nodes of the visibility graph are clustered using a clustering algorithm, which simulates the process to determine natural clusters in the graph. The mutual visibility of regions in the point cloud is utilized, and a scalable graph clustering algorithm is used to identify natural room clusters. Candidate wall segments si are obtained in units of rooms, and the corresponding 3D support point set P(si) is obtained. A contour wall optimization method is used to select a subset of the candidate segments to generate a complete floor plan.

[0030] The contour wall optimization method based on point support degree is any one of greedy algorithm, integer programming, particle swarm optimization and genetic algorithm;

[0031] The technical effects and advantages of the indoor scene layout estimation method based on structural prior provided by the present invention are as follows:

[0032] Depend on Figure 8 It can be observed that the method proposed in this patent excels in detail processing, generating accurate and regular models. When dealing with concave room structures, the method successfully restores the actual wall structure. This result is due to the patent's strategy for resampling corner points on the reconstructed model, which evenly distributes the point cloud density of small wall surfaces, making it easier to identify and recover candidate wall planes during the plane fitting process. Furthermore, this patent sets strict regularization parameters and performs line segment denoising and merging steps to ensure that the final generated model meets the expected accuracy and regularity.

[0033] Furthermore, this patent successfully eliminates the "paper wall" phenomenon commonly seen in reconstruction results, effectively restoring wall width information. Wall width information is calculated based on the model point cloud, making the generated wall closer to reality. This result is attributed to the scalable graph clustering method used in this patent, which can effectively cluster and process candidate wall segments in a single room. Compared to traditional methods, this local processing approach significantly reduces the topological complexity of the reconstruction, eliminates the mutual influence of wall candidate segments between different rooms, and thus improves the efficiency and accuracy of reconstruction. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 This is a flow chart of the indoor scene layout estimation method based on structure prior of the present invention;

[0035] Figure 2 Schematic diagram of grid model patch unit sampling of the indoor scene layout estimation method based on structure prior of the present invention;

[0036] Figure 3 Schematic diagram of the structure segmentation result of the indoor scene layout estimation method based on structure prior of the present invention;

[0037] Figure 4 A schematic diagram of generating candidate segments of facade projections for the indoor scene layout estimation method based on structure priors of the present invention;

[0038] Figure 5 The room clustering result of the scalable graph of the indoor scene layout estimation method based on structure prior of the present invention

[0039] Figure 6 is a schematic diagram of an electronic device of the present invention;

[0040] Figure 7 Schematic diagram of the storage medium of the present invention;

[0041] Figure 8 Schematic diagram of the reconstruction effect of the indoor scene layout estimation method based on structure prior of the present invention DETAILED DESCRIPTION

[0042] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0043] Example 1:

[0044] See also Figure 1As shown, the indoor scene layout estimation method based on structure prior in this embodiment includes:

[0045] Step 1: Resample the point cloud model;

[0046] Step 2: Extract plane and segment structure;

[0047] Step 3: Generate candidate face segments;

[0048] Step 4: Cluster the rooms based on the scalable graph;

[0049] Step 5: Based on the results from steps 1 to 4, optimize the contour wall, select a subset from the candidate segments, and generate a complete floor plan.

[0050] During the 3D model scanning process, the presence of non-uniform factors such as noise, density, and posture leads to uneven distribution of 3D model data. To enhance the robustness of the planar image reconstruction method, this resampling method is based on the Poisson surface reconstruction algorithm and aims to achieve uniform processing of point cloud data.

[0051] The resampling point cloud model method includes:

[0052] The original point cloud model is reconstructed into a mesh model, and the proposed random sampling method is used to sample the surface units of the mesh model;

[0053] like Figure 2 As shown, in the mesh model patch unit M, a three-dimensional point p is randomly sampled by the resampling method; a vertex is randomly selected as the endpoint O, and two adjacent edges OA and OB are used as basis vectors to resample the patch unit;

[0054] like Figure 4 As shown, the method for extracting planes and segmenting structures includes:

[0055] In real-world architectural scenes, most can be described by simplified planar structures, such as piecewise planes or curved surfaces. To handle the discreteness of point cloud data and abstract the structural features of the model, this patent proposes a plane fitting method that uses a random sample consistency method to extract planar shape units from the point cloud model and obtain all planes that meet the fitting constraints. The fitting constraints are determined by RANSAC parameters, including the maximum distance from the point to the plane, the normal angle threshold, the minimum number of supporting points per plane, and the maximum probability of missing a plane candidate. The RANSAC parameters can be adjusted according to the quality of the point cloud data and are applicable to various datasets with different densities and noise levels.

[0056] Based on the results of plane fitting, the point cloud model is segmented using the plane normal vector, and the structural segmentation of the ceiling, floor and facade is further realized. Figure 3 The figure shows the shape point clusters belonging to the facade and plane, respectively. During the model preprocessing stage, the point cloud model's pose is corrected, allowing the segmentation plane set to be calculated by fitting the plane normal vectors of the ceiling and floor plane shape point clusters and combining them with the coordinate axis direction vectors. For each plane in the plane set, structural segmentation is achieved by calculating the dot product of the plane normal vector with the Z-axis direction vector and applying constraints.

[0057] The method for generating candidate face segments includes:

[0058] Compared to ceilings and floors, wall structures in indoor scenes are more difficult to recover due to their complexity. Based on the observation that walls are perpendicular to the ground in most scenes and that the Z coordinate includes height information, we project the vertical facade onto the horizontal plane and focus on generating floor plans in 2D space. Compared with direct extraction in 3D space, extracting floor plans in 2D space can significantly reduce computational complexity and improve robustness. The generated floor plan can greatly assist in the reconstruction of ceilings, floors, and walls. The two core steps in generating floor plans are wall candidate segment generation and wall segment selection.

[0059] Using RANSAC and regularized angle threshold segmentation, planes and facades are extracted from the point cloud model. Based on the Floor-Wall prior assumption, the planes and facades are ensured to be approximately perpendicular. Each fitted facade is considered part of the wall plane and contains a set of shape point clusters, i.e., support points. The fitted facade is projected onto a horizontal plane, and the corresponding two-dimensional support points are projected onto the projection line to generate candidate wall segments using the projected facade endpoints.

[0060] It is particularly important to note that the method of merging line segments based on angle and distance improves the recognition and reconstruction of wall structures. Considering that noise and missing areas in point cloud data may interfere with the planes detected by the RANSAC algorithm, resulting in problems such as over-segmentation, it is necessary to further refine the wall candidate segments, that is, merge line segments, to obtain facade candidate segments in order to obtain clearer and more regular results.

[0061] The methods for merging line segments include:

[0062] If the minimum distance between two wall segments wi and wj is less than the threshold θ2, the wall segment is a type of line segment, and the following conditions are met: the angle between the normal vectors of the two wall segments is less than θ1, and the overlapping length of the projections of the two wall segments is less than θ3 of the shorter line segment, then wi and wj are considered to belong to the same wall segment and should be merged to remove unnecessary segmentation introduced by RANSAC. θ2, θ1 and θ3 are preset thresholds. The method of merging line segments is expressed as follows:

[0063]

[0064] Where, represents the angle between the normal vectors of two wall segments wi and wj, represents the minimum distance between two wall segments wi and wj, represents the overlapping length of the projections of two wall segments wi and wj;

[0065] The room clustering method based on the scalable graph includes:

[0066] The prior estimation is formalized as a projected line segment labeling problem, where a labeling problem represents a room or outdoor area; a line segment prior labeling method based on scalable graph clustering is used for the labeling problem;

[0067] Line segment prior labeling methods include:

[0068] Regions of mutual visibility among line segments representing planes are clustered, corresponding to rooms of the building. A visibility graph is generated by performing ray casting between the detected facade line segments. Subsequently, the nodes of the visibility graph are clustered using a scalable graph clustering algorithm that uses flow simulation methods to determine natural clusters in the graph.

[0069] A unit line segment generates a coarse occupancy bitmap through each column of planes, and each occupied pixel is considered to have the same normal as the corresponding plane; the line segment length is set to 0.5m;

[0070] It is particularly noteworthy that this patent uses line segments instead of all points or planes, which greatly reduces the number of nodes in the visibility graph, making the calculation feasible;

[0071] A rough occupancy map is generated by passing a unit line segment through each column of the plane, and each occupied pixel is regarded as a patch with the same normal as the corresponding plane; a line segment of length 0.5m is set, pi is the i-th line segment, its center position is ci, and its normal is ni; for each pair pi, pj, i, j, a ray is cast between the points ci+A×ni and cj+A×nj, where A=10cm; if the line segment connecting the two is in the same direction as the normal vector of the two and there is no obstacle between the connecting line segments, the visibility between pi and pj is set to 1, otherwise it is set to 0, that is, a visibility graph is generated, and the nodes of the visibility graph are clustered using a clustering algorithm; the calculated visibility is interpreted as the flow between the node pairs corresponding to the corresponding point patches;

[0072] It is important to note that the main advantage of this method is that it is unsupervised, so there is no need to manually specify the number of generated labels.

[0073] After obtaining n disjoint patch clusters, define a set of room labels R = {r1, ..., rn}, which will be used in the rest of the reconstruction process; each point of the point cloud is assigned to the room label of the corresponding line segment candidate;

[0074] Note in particular that the number of room labels n may be larger than the number of rooms actually included in the final reconstruction;

[0075] like Figure 5 As shown in Figure 2, the effect comparison of the projected candidate line segments before and after graph clustering is performed;

[0076] It is particularly noted that: the key to this method is to identify natural clusters of rooms inside a building through the formation of a visibility map;

[0077] Ray casting is used to test the visibility between points on the detected surface, thereby constructing a visibility graph. The nodes of the visibility graph are clustered using a clustering algorithm, which simulates the process to determine natural clusters in the graph. The mutual visibility of regions in the point cloud is utilized, and a scalable graph clustering algorithm is used to identify natural room clusters. Candidate wall segments si are obtained in units of rooms, and the corresponding three-dimensional support point set P(si) is obtained. A contour wall optimization method is used to select a subset of the set R from the candidate segments to generate a complete floor plan.

[0078] It is important to note that the candidate segment set for a single room may contain a large number of internal candidate segments, while only the outer contour edges need to be retained. To address this issue, this patent proposes a contour wall optimization method based on point support. The point support optimization method can be any of a greedy algorithm, integer programming, particle swarm optimization, and genetic algorithm. Each candidate segment is assigned a value based on its position and adjacency, converting the candidate segment selection problem into an optimization problem.

[0079] It is particularly noted that the various preset thresholds in the above analysis process are set by those skilled in the art based on actual conditions or obtained through simulation of a large amount of data;

[0080] Example 2

[0081] See also Figure 6 According to another aspect of the present application, an electronic device 500 is provided. The electronic device 500 may include one or more processors and one or more memories. The memories may store computer-readable code, which, when executed by the one or more processors, may execute the above-described method for estimating indoor scene layout based on structure priors.

[0082] The system according to the embodiment of the present application can also be used by Figure 6 The electronic device architecture shown in FIG. Figure 6 As shown, the electronic device 500 may include a bus 501, one or more CPUs 502, ROM 503, RAM 504, a communication port 505 connected to a network, an input / output 506, a hard disk 507, etc. The storage device in the electronic device 500, such as ROM 503 or hard disk 507, may store the indoor scene layout estimation method based on structural prior provided by the present application. Furthermore, the electronic device 500 may also include a user interface 508. Of course, Figure 6 The architecture shown is only exemplary and can be omitted according to actual needs when implementing different devices. Figure 6 One or more components of an electronic device are shown.

[0083] Example 3

[0084] See also Figure 7 , is a computer-readable storage medium according to one embodiment of the present application. Computer-readable instructions are stored on the computer-readable storage medium. When the computer-readable instructions are executed by a processor, the method for estimating indoor scene layout based on structural prior according to the embodiment of the present application described with reference to the above figures can be executed. The storage medium includes, but is not limited to, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0085] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

[0086] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for estimating indoor scene layout based on structural priors, characterized by: Methods include: Step 1: Resample the point cloud model; Step 2: Extract plane and segment structure; Step 3: Generate candidate face segments; Step 4: Clustering rooms based on the scalable graph. The clustering method of rooms based on the scalable graph includes: The prior estimation is formalized as a projected line segment labeling problem, where a labeling problem represents a room or outdoor area. The line segment prior labeling method based on scalable graph clustering is used for the labeling problem. The line segment prior labeling method includes: Regions of mutual visibility among line segments representing planes are clustered, corresponding to rooms of the building. A visibility graph is generated by performing ray casting between the detected facade line segments. Subsequently, the nodes of the visibility graph are clustered using a scalable graph clustering algorithm that uses flow simulation methods to determine natural clusters in the graph. A rough occupancy map is generated by passing a unit line segment through each column of the plane, and each occupied pixel is considered to have the same normal as the corresponding plane. The line segment is set to be 0.5m long, pi is the i-th line segment, its center position is ci, and its normal is ni. For each pair pi,pj,i,j, a ray is cast between the points ci+A×ni and cj+A×nj, where A=10cm. If the line segment connecting the two has the same normal vector direction and there is no obstacle between the connecting line segment, the visibility between pi and pj is set to 1, otherwise it is set to 0, that is, a visibility graph is generated, and the nodes of the visibility graph are clustered using a clustering algorithm. The calculated visibility is interpreted as the flow between the node pairs corresponding to the corresponding point patches. In step 5, based on the results of steps 1 to 4, the contour wall is optimized, a subset is selected from the candidate segments, and a complete floor plan is generated.

2. The method for estimating indoor scene layout based on structure prior according to claim 1, characterized in that: The resampling point cloud model method includes: The original point cloud model is reconstructed into a mesh model, and the proposed random sampling method is used to sample the surface units of the mesh model; In the mesh model patch unit M, a three-dimensional point p is randomly sampled by a resampling method; a vertex is randomly selected as an endpoint O, and two adjacent edges OA and OB are used as basis vectors to resample the patch unit.

3. The method for estimating indoor scene layout based on structure prior according to claim 1, characterized in that: The method for extracting a plane and segmenting a structure comprises: The random sample consistency method is used to extract planar shape units from the point cloud model to obtain all planes that meet the fitting constraints. The fitting constraints are determined by RANSAC parameters, including the maximum distance from the point to the plane, the normal angle threshold, the minimum number of support points for each plane, and the maximum probability of missing a plane candidate. Based on the results of plane fitting, the point cloud model is structurally segmented using the plane normal vector. For each plane in the plane set, the plane normal vector is multiplied by the Z-axis direction vector, and the constraint conditions are applied to achieve structural segmentation.

4. The method for estimating indoor scene layout based on structure prior according to claim 1, characterized in that: The method for generating candidate face segments includes: The Z coordinate includes height information, and the vertical facade is projected onto the horizontal plane, generating a plan view in 2D space. The plane and facade are extracted from the point cloud model using RANSAC and regularized angle threshold segmentation. Based on the Floor-Wall prior assumption, the plane and facade are ensured to be approximately perpendicular. Each fitted facade is considered part of the wall plane and contains a set of shape point clusters, namely support points. The fitted facade is projected onto the horizontal plane, and the corresponding 2D support points are projected onto the projection line, generating candidate wall segments using the projected facade endpoints. The wall candidate segments are further refined by merging line segments to obtain facade candidate segments. The methods for merging line segments include: Assume that the minimum distance between two wall segments wi and wj is less than the threshold θ2 and the following conditions are met: If the angle between the normal vectors of two wall segments is less than the threshold θ1, and the overlapping length of the projections of the two wall segments is less than the shorter line segment θ3, then wi and wj are considered to belong to the same wall segment. wi and wj are merged to remove unnecessary segmentation introduced by RANSAC. θ2, θ1 and θ3 are preset thresholds. Where, represents the angle between the normal vectors of two wall segments wi and wj, represents the minimum distance between two wall segments wi and wj, Represents the overlapping length of the projections of two wall segments wi and wj.

5. The method for estimating indoor scene layout based on structure prior according to claim 1, characterized in that: Methods for generating a complete floor plan include: Ray casting tests are used to detect the visibility between surface points and construct a visibility graph. Nodes in the visibility graph are clustered using a clustering algorithm, which simulates processes to identify natural clusters within the graph. The mutual visibility of regions within the point cloud is exploited, and a scalable graph-based clustering algorithm is used to identify natural room clusters. This yields candidate wall segments si, defined as rooms, and the corresponding 3D support point set P(si). A contour wall optimization method is used to select subsets from the candidate segments to generate a complete floor plan. The contour wall optimization method based on point support degree is any one of greedy algorithm, integer programming, particle swarm optimization algorithm and genetic algorithm.

Images