Semantic wall reconstruction method based on topological graph expression

Abstract

The invention discloses a semantic wall surface reconstruction method based on topological graph expression, which focuses on semantic expression and reconstruction of a building wall surface, and specifically comprises the following steps: S100, constructing a wall surface layout graph model by using a composite topological graph with a two-layer structure, and expressing geometric, semantic and topological information of doors and windows on the wall surface by using nodes and edges respectively; s200, formulating a parameterized expression method and parameter constraints of the wall layout map model; s300, refining a building wall surface structure from the laser point cloud data, and extracting coarse contours of doors and windows on the wall surface; and S400, reconstructing a semantic wall surface model from the rough point cloud contours of the wall surface door and window by adopting a global optimization algorithm combining Gibbs sampling with simulated annealing. According to the method, the problems of different data quality, complex and non-uniform model description and difficulty in realizing global optimization in semantic reconstruction of the wall surface are effectively solved, the method can adapt to wall surface reconstruction tasks of different types, complexity and data quality, and geometric accurate and topological correct modeling is realized.

Description

Technical Field

[0001] The present invention relates to the field of semantic reconstruction of three-dimensional building models, and in particular to a semantic expression and reconstruction method of a building wall based on a topological graph. Background Art

[0002] In building structures, walls are not only their basic components, but also key elements for semantic reconstruction. Semantic modeling of walls can restore the geometric form and semantic attributes of each wall element, and can accurately express the spatial and topological associations between these elements. Although semantic reconstruction technology has made significant progress in recent years, there are still major challenges in automatically constructing semantic models of walls from images or point cloud data. First, due to the variability of shooting conditions, wall data collection is often affected by occlusion or incomplete information, which limits data-based semantic recognition methods in practical applications. Second, the structure of building walls is complex and changeable, and the walls of different buildings have significant differences in design and function, which makes it difficult to develop a universal model suitable for semantic reconstruction of all walls. Summary of the invention

[0003] In view of the difficulties faced in the automatic reconstruction of semantic wall models, the present invention proposes a semantic wall model expression method based on a composite topological graph, namely, a wall layout graph model. Different from traditional methods, the proposed graph structure is a layered graph model, which has the ability to illustrate the geometric and topological relationships between wall elements. The use of composite topological graphs effectively captures and displays the inherent hierarchical properties of wall structures, and has intuitive representation capabilities and robust explanatory structures, which improves machine readability. By adopting a parametric wall layout graph model, the present invention transforms the reconstruction process of the semantic wall model into an optimization problem of model parameters, and uses a Bayesian framework to achieve progressive automatic reconstruction.

[0004] To achieve the above object, the present invention adopts the following technical solution:

[0005] The semantic wall reconstruction method based on topological graph expression has the following specific steps:

[0006] S100. Construct a wall layout graph model using a two-layer composite topological graph, and use nodes and edges to express the geometric, semantic and topological information of doors and windows on the wall;

[0007] S200. Developing a parametric expression method and parameter constraints for a wall layout model;

[0008] S300. Refining the building wall structure from the laser point cloud data, extracting the rough outlines of the doors and windows on the wall;

[0009] S400. A global optimization algorithm based on Gibbs sampling and simulated annealing is used to reconstruct a semantic wall model from the rough point cloud contours of doors and windows on the wall.

[0010] On the basis of the above technical solution, the semantic wall reconstruction method based on topological graph expression, in step S100, comprises the following steps:

[0011] S101. In the two-layer composite topology diagram used, the first layer diagram represents the layout of the floors, and its nodes respectively record the number of floors, width, height, edge width, distance between the wall element and the bottom edge of the floor, and the second layer topology diagram;

[0012] S102. The second layer diagram represents the layout of the wall elements in the floor, wherein the nodes are composed of one or more wall elements, and the number, width and height of the elements are recorded respectively;

[0013] S103. The edge of the second-layer graph consists of two parts: one is the intra-cluster interval, which records the distance between elements inside the node; the other is the inter-cluster interval, which records the interval between nodes.

[0014] On the basis of the above technical solution, the semantic wall reconstruction method based on topological graph expression, in step S200, comprises the following steps:

[0015] S201. Using a typical array of characters to record the first and second floor maps in the wall layout map, wherein the parameter type is represented by a key ("Key"), and its content is described as a value ("Value"), in particular, the second floor map is recorded as a value in the "eachfloor" key of the first floor map node;

[0016] S202. Parameter constraints include 3 categories:

[0017] (1) Single parameter constraints, obtained from building design codes and standards;

[0018] (2) Structural constraints, obtained by the Gestalt principles of similarity and proximity;

[0019] (3) Global constraints, which are determined by the topological properties of the building wall itself.

[0020] On the basis of the above technical solution, the semantic wall reconstruction method based on topological graph expression, in step S300, comprises the following steps:

[0021] S301. For the unordered point cloud, a multi-scale threshold random sampling consensus algorithm is used to extract the wall plane, wherein the large-scale threshold is used to determine the main plane of the wall, and the small-scale threshold is used to extract the contour structure of the wall components such as doors and windows;

[0022] S302. For the ordered point cloud, a scan line-based structure contour extraction method is used, and the specific steps are as follows;

[0023] (1) Calculate the spatial coordinate difference and Euclidean distance of adjacent laser points according to the time series;

[0024] (2) The inverse of the calculated Euclidean distance is used as the modulus, and a threshold for changing the plane is set. When the modulus value is less than the threshold, it indicates that the two adjacent laser points are in the same plane.

[0025] (3) determine the plane and calculate the direction of change of its normal;

[0026] (4) Number the normal directions and assign them to the octree space to determine the outline of the wall structure.

[0027] On the basis of the above technical solution, the semantic wall reconstruction method based on topological graph expression, in step S400, comprises the following steps:

[0028] S401. The prior conditions in the optimization model include:

[0029] (1) The rationality control conditions of wall layout are composed of window width, number and wall width;

[0030] (2) The reasonableness judgment criteria of the window-to-wall area ratio determined by the building wall lighting energy-saving specifications;

[0031] (3) The rationality constraints of the random generation model for detecting the inherent topological relationship between wall elements;

[0032] S402. The likelihood function calculation process in the optimization model includes:

[0033] (1) The point counting function is determined by the proportion of laser points that fall into the random generation model to the total number of point clouds;

[0034] (2) The point density calculation function is defined by the average number of laser points contained in each wall element;

[0035] The S403 Gibbs sampling process is controlled by three kernel functions, whose selection is determined by the posterior probability and the number of iterations:

[0036] (1) From the beginning of the iteration to the third time or when the posterior probability is less than 0.5, the top kernel mainly controls the sampling process of the parameters in the first layer graph;

[0037] (2) When the iteration is from one third to two thirds or the posterior probability is less than 0.7, the bottom kernel mainly controls the sampling process of the node parameters in the second layer graph;

[0038] (3) When the number of iterations is greater than two-thirds or the posterior probability is greater than 0.7, the optimization kernel is used to control the sampling process of the parameters of the edges in the second-layer graph;

[0039] S404. Simulated annealing is used to control the direction of Gibbs sampling. When the temperature is cooled to a set threshold, an optimal semantic wall model can be obtained.

[0040] The beneficial effects of the present invention are:

[0041] The present invention uses a composite topological graph to describe the layout structure of the building wall, fully expressing the geometric, topological and semantic consistency of the wall elements, solving the problems faced by the semantic reconstruction of the building wall, such as uneven data quality, complex and ununiform model description, and difficulty in achieving global optimization.

[0042] Geometrically accurate and topologically correct modeling is achieved in wall reconstruction tasks of various types, different complexities and different data qualities. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 The figure is a schematic diagram of a method flow of an embodiment of the present invention.

[0044] Figure 2 Schematic diagram of a wall layout model according to an embodiment of the present invention DETAILED DESCRIPTION

[0045] The specific implementation of the present invention is further described below in conjunction with the accompanying drawings of the embodiments of the present invention.

[0046] like Figure 1 As shown, the semantic wall reconstruction method based on topological graph expression of this embodiment includes the following steps:

[0047] S100. Construct a wall layout graph model using a two-layer composite topological graph, and use nodes and edges to express the geometric, semantic and topological information of doors and windows on the wall;

[0048] S200. Developing a parametric expression method and parameter constraints for a wall layout model;

[0049] S300. Refining the building wall structure from the laser point cloud data, extracting the rough outlines of the doors and windows on the wall;

[0050] S400. A global optimization algorithm based on Gibbs sampling and simulated annealing is used to reconstruct a semantic wall model from the rough point cloud contours of doors and windows on the wall.

[0051] The above step S100 specifically includes the following process:

[0052] S101. Figure 2, the design of the wall layout model adopts a double-layer composite structure:

[0053] (1) First-level structure: Focus on the overall layout of the floor. According to the "proximity principle" in Gestalt theory, when adjacent floors have similar elements, these elements can be effectively combined to form clusters with common characteristics.

[0054] (2) Second layer structure: This layer goes deep into the interior of the floor and describes the distribution of wall elements in detail. Each node is the result of clustering adjacent wall elements through Gestalt theory, representing a collection of one or more wall elements.

[0055] S102. First layer graph node attribute design:

[0056] (1) Number of floors: Each node represents a set of elements with at least one floor.

[0057] (2) Floor width: Usually the same as the wall width, but may vary depending on the wall design.

[0058] (3) Floor height: affected by many factors such as floor location, function, building type and energy efficiency.

[0059] (4) Edge width: the distance between the wall element and the wall edge.

[0060] (5) Distance between the wall element and the bottom edge of the floor: The height of the wall element is determined by its distance from the bottom of the floor. For example, in the ground floor, this refers to the distance between the wall element and the ground.

[0061] (6) Bottom-level layout graph: Each node in the top-level layout graph is linked to a second-level layout graph.

[0062] S103. Second-layer node configuration: The similarity and proximity principles of Gestalt theory are used to cluster wall elements with similar attributes and close distances. The attributes include:

[0063] (1) Number of elements: Each node in the underlying layout graph contains at least one wall element, which are consistent in geometric properties and spacing.

[0064] (2) Element width: As a description of the shape of the wall element, the width affects the total number of wall elements in a floor and, as a scale factor, determines the change in the spacing between adjacent elements.

[0065] (3) Element height: Wall elements of the same type on the same floor usually have the same height, which is determined by the floor height and the distance between the wall element and the floor;

[0066] (4) Element interval: records the minimum interval between two elements.

[0067] S104. In the second-level graph, an edge consists of two parts:

[0068] (1) Intra-cluster distance: records the distance between each wall element within the node.

[0069] (2) Inter-cluster distance: records the distance between different nodes. The inter-cluster distance is composed of the element distance between two adjacent nodes. Specifically, it refers to the distance between the last wall element of the current node and the first wall element of the next node.

[0070] The above step S200 specifically includes the following process:

[0071] S201. Using a typical array of characters to record the first and second floor maps in the wall layout map, wherein the parameter type is represented by a key ("Key"), and its content is described as a value ("Value"), in particular, the second floor map is recorded as a value in the "eachfloor" key of the first floor map node;

[0072] S202. Parameter constraints include 3 categories:

[0073] (1) Single parameter constraints, obtained from building design codes and standards;

[0074] (2) Structural constraints, obtained by the similarity and proximity principles in the Gestalt law; the number of windows n on the same floor window Subject to the window width w window The window is separated by sp The influence changes as shown in formula (1):

[0075] (1)

[0076] in,

[0077]

[0078]

[0079] The range of window width is affected by the total number of windows.

[0080] (3) Global constraints, determined by the topological properties of the building wall, the height of the window h window The sum of the height intervals cannot exceed the floor height h floor Therefore, the window height range can be determined as formula (2):

[0081] (2)

[0082] The above step S300 specifically includes the following process:

[0083] S301. For the unordered point cloud, a multi-scale threshold random sampling consensus algorithm is used to extract the wall plane, wherein the large-scale threshold is used to determine the wall range, and the small-scale threshold is used to extract the contour structure of the wall components such as doors and windows;

[0084] S302. For the ordered point cloud, a scan line-based structure contour extraction method is used, and the specific steps are as follows;

[0085] (1) Calculate the spatial coordinate difference and Euclidean distance of adjacent laser points according to the time series;

[0086] (2) Taking the inverse of the Euclidean distance as the modulus and setting a threshold for plane change, when the modulus value is less than the threshold, it indicates that the two adjacent laser points are in the same plane;

[0087] (3) determine the plane and calculate the direction of change of its normal;

[0088] (4) Number the normal directions and assign them to the octree space to determine the outline of the wall structure.

[0089] The above step S400 includes the following steps:

[0090] S401. The prior conditions in the optimization model include:

[0091] (1) The rationality control conditions of the wall layout are composed of the window width ww, the number nw and the wall width W;

[0092] (3)

[0093] (2) The reasonableness judgment criteria of the window-to-wall area ratio determined by the building wall lighting energy-saving specifications;

[0094] (4)

[0095] (3) The rationality constraints of the random generation model for detecting the inherent topological relationship between wall elements;

[0096] (5)

[0097] in, is the y coordinate of the lower left corner of the window rectangle, and They are the lower left corner coordinates and rectangle height of the balcony rectangle respectively.

[0098] S402. The likelihood function calculation process in the optimization model includes:

[0099] (1) The point counting function is composed of laser points falling into the randomly generated model Total number of point clouds The proportion of

[0100] (6)

[0101] (2) The point density calculation function is defined by the average number of laser points contained in each wall element;

[0102] (7)

[0103] The S403 Gibbs sampling process is controlled by three kernel functions, whose selection is determined by the posterior probability and the number of iterations:

[0104] (1) From the beginning of the iteration to the third time or when the posterior probability is less than 0.5, the top kernel mainly controls the sampling process of the parameters in the first layer graph;

[0105] (2) When the iteration is from one third to two thirds or the posterior probability is less than 0.7, the bottom kernel mainly controls the sampling process of the node parameters in the second layer graph;

[0106] (3) When the number of iterations is greater than two-thirds or the posterior probability is greater than 0.7, the optimization kernel is used to control the sampling process of the parameters of the edges in the second-layer graph;

[0107] S404. Simulated annealing is used to control the direction of Gibbs sampling. When the temperature is cooled to a set threshold, an optimal semantic wall model can be obtained.

[0108]

[0109] The present invention uses a composite topological graph to describe the layout structure of building walls, fully expressing the geometric, topological and semantic consistency of wall elements, solving the problems faced by semantic reconstruction of building walls, such as uneven data quality, complex and ununiform model description, and difficulty in achieving global optimization. It achieves geometrically accurate and topologically correct modeling in wall reconstruction tasks of multiple types, different complexities and different data qualities.

[0110] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A semantic wall reconstruction method based on topological graph expression, characterized in that: The specific steps are as follows: S100. Construct a wall layout graph model using a two-layer composite topological graph, and use nodes and edges to express the geometric, semantic and topological information of doors and windows on the wall; S200. Developing a parametric expression method and parameter constraints for a wall layout model; S300. Refining the building wall structure from the laser point cloud data, extracting the rough outlines of the doors and windows on the wall; S400. A global optimization algorithm based on Gibbs sampling and simulated annealing is used to reconstruct a semantic wall model from the rough point cloud contours of doors and windows on the wall.

2. The method according to claim 1, characterized in that: The step S100 is specifically as follows: S101. In the two-layer composite topology diagram used, the first layer diagram represents the layout of the floors, and its nodes respectively record the number of floors, width, height, edge width, distance between the wall element and the bottom edge of the floor, and the second layer topology diagram; S102. The second layer diagram represents the layout of the wall elements in the floor, wherein the nodes are composed of one or more wall elements, and the number, width and height of the elements are recorded respectively; S103. The edge of the second-layer graph consists of two parts: one is the intra-cluster interval, which records the distance between elements inside the node; the other is the inter-cluster interval, which records the interval between nodes.

3. The method according to claim 1, characterized in that The step S200 is specifically as follows: S201. A typical array of characters is used to record the first and second floor maps in the wall layout diagram, wherein the parameter type is represented by a key ("Key"), and its content is described as a value ("Value"). In particular, the second floor map is recorded as a value in the "eachfloor" key of the first floor map node; S202. Parameter constraints include 3 categories: (1) Single parameter constraints, obtained from building design codes and standards; (2) Structural constraints, obtained by the Gestalt principles of similarity and proximity; (3) Global constraints, which are determined by the topological properties of the building wall itself.

4. The method according to claim 1, characterized in that The step S300 is specifically as follows: S301. For the unordered point cloud, a multi-scale threshold random sampling consensus algorithm is used to extract the wall plane, wherein the large-scale threshold is used to determine the main plane of the wall, and the small-scale threshold is used to extract the contour structure of the wall components such as doors and windows; S302. For the ordered point cloud, a scan line-based structure contour extraction method is used, and the specific steps are as follows; (1) Calculate the spatial coordinate difference and Euclidean distance of adjacent laser points according to the time series; (2) The inverse of the calculated Euclidean distance is used as the modulus, and a threshold for changing the plane is set. When the modulus value is less than the threshold, it indicates that the two adjacent laser points are in the same plane. (3) determine the plane and calculate the direction of change of its normal; (4) Number the normal directions and assign them to the octree space to determine the outline of the wall structure.

5. The method according to claim 1, characterized in that: The step S400 is specifically as follows: S401. The prior conditions in the optimization model include: (1) The rationality control conditions of wall layout are composed of window width, number and wall width; (2) The reasonableness judgment criteria of the window-to-wall area ratio determined by the building wall lighting energy-saving specifications; (3) The rationality constraints of the random generation model for detecting the inherent topological relationship between wall elements; S402. The likelihood function calculation process in the optimization model includes: (1) The point counting function is determined by the proportion of laser points that fall into the random generation model to the total number of point clouds; (2) The point density calculation function is defined by the average number of laser points contained in each wall element; The S403 Gibbs sampling process is controlled by three kernel functions, whose selection is determined by the posterior probability and the number of iterations: (1) From the beginning of the iteration to the third time or when the posterior probability is less than 0.5, the top kernel mainly controls the sampling process of the parameters in the first layer graph; (2) When the iteration is from one third to two thirds or the posterior probability is less than 0.7, the bottom kernel mainly controls the sampling process of the node parameters in the second layer graph; (3) When the number of iterations is greater than two-thirds or the posterior probability is greater than 0.7, the optimization kernel is used to control the sampling process of the parameters of the edges in the second-layer graph; S404. Simulated annealing is used to control the direction of Gibbs sampling. When the temperature is cooled to a set threshold, an optimal semantic wall model can be obtained.

Images