A three-dimensional data acquisition and fusion method and system for road surface detection

Abstract

The present invention discloses a 3D data acquisition and fusion method and system for road surface detection. In the process of detecting the road surface, a laser radar is used to collect the 3D point cloud spatial coordinate data of the road surface in real time, and a distance encoder is used to measure the precise distance information; based on the 3D point cloud spatial coordinate data of the road surface, adaptive grid division is performed to obtain the 3D grid primitive of the road surface; for the 3D grid primitive contained in the road surface, the precise distance information measured by the distance encoder for the 3D grid primitive is referred to and characterized as a layer entity set; the BiLSTM network is used to extract the features of the layer entity set of the road surface to obtain the layer feature vector of the 3D grid primitive of the road surface; and then the prototype network is used to identify the category of the 3D grid primitive of the road surface based on the layer feature vector; according to the category of the 3D grid primitive of the road surface, the 3D grid primitive of the road surface corresponding to the road surface disease category is determined and marked. The present invention realizes more efficient and more accurate 3D data acquisition of road surface conditions.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent detection technology, and in particular relates to a three-dimensional data acquisition and fusion method and system for road surface detection. Background Art

[0002] Pavement inspection is one of the important links in highway maintenance. By conducting regular or irregular inspections of road conditions, road defects can be discovered and dealt with in a timely manner to ensure the safety and smooth flow of roads. Pavement inspection data is an important basis for making maintenance decisions. Through scientific road inspection means and methods, accurate and rich road condition information can be obtained to provide strong support for scientific maintenance decisions. This helps to achieve the optimal allocation of maintenance resources and improve maintenance efficiency and quality. With the advancement of science and technology and the increase in highway maintenance needs, pavement inspection technology is also constantly innovating and developing. From traditional manual inspections to current automated and intelligent inspection methods, the continuous improvement of pavement inspection technology has provided more convenient, efficient and accurate services for highway maintenance.

[0003] At present, the application of advanced technologies such as the 3D laser point cloud road intelligent detection system has made road detection more comprehensive, detailed and accurate. The 3D laser point cloud technology uses laser radar to scan the ground or objects to generate a data set containing a large number of 3D coordinate points. This technology uses a laser to emit extremely short laser pulses, which are focused by the lens and radiated to the surface of the observed object. The receiver captures the reflected laser beam and calculates the spatial position information of the target object based on the speed of light and flight time. This technology can efficiently obtain the accurate spatial position information of the target object at multiple angles and viewing angles, generate high-precision, dense 3D model data, and is not affected by ambient light.

[0004] However, when LiDAR collects 3D point clouds of road surfaces, the main problems with data collection accuracy include: insufficient resolution and density of point clouds, which reduces measurement accuracy; dust, rain, fog and other particles in the environment absorb or scatter lasers, affecting measurement accuracy; and uneven road reflectivity can also cause deviations in collected data. LiDAR measurement errors have a cumulative effect, especially when measuring over long distances, where the adverse effects of error accumulation will be further magnified. Summary of the invention

[0005] In view of the deficiencies of the above-mentioned prior art, the technical problem to be solved by the present invention is: how to provide a three-dimensional data acquisition and fusion method and system for road surface detection, realize a road surface detection mode combining a three-dimensional laser point cloud and a distance encoder, and utilize the distance encoder to provide accurate distance information that complements the three-dimensional point cloud spatial coordinate data results of the laser radar, thereby improving the overall measurement accuracy through data fusion; thereby achieving more efficient and accurate three-dimensional data acquisition of road surface conditions, which is conducive to analyzing and extracting the precise three-dimensional coordinates of road surface disease areas.

[0006] The present invention adopts the following technical solution:

[0007] The present invention discloses a three-dimensional data acquisition and fusion method for road surface detection, which is characterized by comprising the following steps:

[0008] During the inspection of the road surface, the laser radar is used to collect the three-dimensional point cloud spatial coordinate data of the road surface in real time, and the distance encoder is used to measure the precise distance information;

[0009] Based on the three-dimensional point cloud spatial coordinate data of the road surface, adaptive meshing is performed to obtain the three-dimensional mesh primitives of the road surface;

[0010] For the three-dimensional grid primitives included in the road surface, referring to the precise distance information measured by the distance encoder for the three-dimensional grid primitives, the three-dimensional grid primitives are characterized as a layer entity set;

[0011] The BiLSTM network is used to extract features from the layer entity set of the road surface to obtain the layer feature vector of the three-dimensional grid primitive of the road surface; and then the prototype network is used to identify the category of the three-dimensional grid primitive of the road surface based on the layer feature vector;

[0012] According to the category of the pavement three-dimensional grid primitive, the pavement three-dimensional grid primitive corresponding to the pavement disease category is determined and marked.

[0013] Preferably, the specific steps of performing adaptive meshing on the road surface three-dimensional point cloud spatial coordinate data include:

[0014] The spatial coordinate information of the road surface three-dimensional point cloud is divided into mesh vertices based on the Delaunay triangulation algorithm to form a triangular initial mesh. The initial mesh is a polygon formed by connecting the feature points of the road surface three-dimensional point cloud as mesh vertices;

[0015] Regularize the initial grid;

[0016] For the initial grid after regularization, adaptive grid subdivision is performed according to the grid size and grid quality index;

[0017] The regularization of the initial mesh includes: supplementing the missing edges of the missing polygons, inserting new mesh vertices at the intersections of the polygon edges, and merging the edges of the overlapping polygons.

[0018] Preferably, for the three-dimensional grid primitive, the three-dimensional grid primitive is divided into layer entities according to the precise distance information measured by the distance encoder within the range of the three-dimensional grid primitive; the three-dimensional point cloud spatial coordinates within the range of the three-dimensional grid primitive and their corresponding precise distance information are traversed, and when the distance difference between the precise distance information of the three-dimensional point cloud spatial coordinates is greater than a threshold, they are attributed to different layer entities, thereby forming a layer entity set.

[0019] Preferably, the BiLSTM network is used to extract features from the layer entity set of the road surface to obtain the layer feature vector of the three-dimensional grid primitive of the road surface, which specifically includes: by inputting the layer entity set of the road surface as an input into the BiLSTM network, the layer feature vector of the three-dimensional grid primitive of the road surface can be obtained, which is expressed as ,Right now

[0020]

[0021] in It means that the layer entity set is input into BiLSTM for feature extraction, so that the layer entity combination is mapped into a layer feature vector of a road surface three-dimensional grid primitive;

[0022] in, Represents a set of layer entities of the road surface 3D grid primitive reference plane, and maps the road surface 3D grid primitive reference plane to the layer entities in the set according to its spatial position . and The layer entity sets respectively represent the lower adjacent layer and the upper adjacent layer of the reference plane of the road surface three-dimensional grid primitive.

[0023] Preferably, the prototype network calculates the layer feature vector of the input road surface three-dimensional grid primitive Relative to Each of the categories Distribution:

[0024]

[0025] in express categories do not belong to the category Other categories; the prototype network is based on the layer feature vector of the road 3D grid primitive Relative to Each of the categories The distribution of pavement 3D grid primitives is determined to determine the characteristic categories.

[0026] On the other hand, the present invention discloses a three-dimensional data acquisition and fusion system for road surface detection, characterized in that it includes:

[0027] The embedded integrated sensing unit integrates a laser radar and a distance encoder, which is used to collect the three-dimensional point cloud spatial coordinate data of the road surface in real time using the laser radar during the inspection process, and to measure the precise distance information using the distance encoder;

[0028] The point cloud decomposition module performs adaptive meshing based on the three-dimensional point cloud spatial coordinate data of the road surface to obtain the three-dimensional mesh primitives of the road surface;

[0029] A pavement primitive layering module is used to characterize the three-dimensional grid primitives contained in the pavement as a layer entity set with reference to the precise distance information measured by the distance encoder for the three-dimensional grid primitives;

[0030] The pavement primitive feature recognition module uses the BiLSTM network to extract features from the pavement layer entity set to obtain the layer feature vector of the pavement three-dimensional grid primitive; and then uses the prototype network to identify the category of the pavement three-dimensional grid primitive based on the layer feature vector;

[0031] The pavement primitive disease category marking module determines and marks the pavement three-dimensional grid primitive corresponding to the pavement disease category according to the category of the pavement three-dimensional grid primitive.

[0032] Preferably, the point cloud decomposition module performs adaptive meshing on the road surface three-dimensional point cloud spatial coordinate data, specifically including:

[0033] The spatial coordinate information of the road surface three-dimensional point cloud is divided into mesh vertices based on the Delaunay triangulation algorithm to form a triangular initial mesh. The initial mesh is a polygon formed by connecting the feature points of the road surface three-dimensional point cloud as mesh vertices;

[0034] Regularize the initial grid;

[0035] For the initial grid after regularization, adaptive grid subdivision is performed according to the grid size and grid quality index.

[0036] Preferably, the regularization of the initial mesh includes: supplementing the missing edges of the missing polygons, inserting new mesh vertices at the intersections of the edges of the polygons, and merging the edges of the overlapping polygons.

[0037] Preferably, the pavement primitive layering module divides the three-dimensional grid primitive into layer entities according to the precise distance information measured by the distance encoder within the range of the three-dimensional grid primitive; traverses the three-dimensional point cloud spatial coordinates and their corresponding precise distance information within the range of the three-dimensional grid primitive, and when the distance difference between the precise distance information of the three-dimensional point cloud spatial coordinates is greater than a threshold, attributes them to different layer entities, thereby forming a layer entity set.

[0038] Preferably, the pavement primitive feature recognition module uses a BiLSTM network to extract features from a pavement layer entity set to obtain a layer feature vector of a pavement three-dimensional grid primitive, specifically comprising: by inputting the pavement layer entity set as an input into the BiLSTM network, a layer feature vector of a pavement three-dimensional grid primitive can be obtained, which is expressed as ,Right now

[0039]

[0040] in It means that the layer entity set is input into BiLSTM for feature extraction, so that the layer entity combination is mapped into a layer feature vector of a road surface three-dimensional grid primitive;

[0041] in, Represents a set of layer entities of the road surface 3D grid primitive reference plane, and maps the road surface 3D grid primitive reference plane to the layer entities in the set according to its spatial position . and The layer entity sets respectively represent the lower adjacent layer and the upper adjacent layer of the reference plane of the road surface three-dimensional grid primitive.

[0042] Preferably, the prototype network calculates the layer feature vector of the input road surface three-dimensional grid primitive Relative to Each of the categories Distribution:

[0043]

[0044] in express categories do not belong to the category Other categories; the prototype network is based on the layer feature vector of the road 3D grid primitive Relative to Each of the categories The distribution of pavement 3D grid primitives is determined to determine the characteristic categories.

[0045] Beneficial effects of the present invention:

[0046] The present invention integrates the three-dimensional point cloud spatial coordinate data of the road surface and the precise distance information formed by the distance encoder, and utilizes the advantage of the distance encoder's low sensitivity to ambient light and particulate matter, so as to maintain a high measurement accuracy in harsh environments; the distance information obtained by the distance encoder can help correct the measurement error of the laser radar, especially in long-distance measurement; the addition of the distance encoder helps to improve the problems caused by the uneven reflectivity of the road surface, and reduces the impact of this unevenness on the acquisition results by optimizing the distance information; further, the present invention grids the three-dimensional point cloud spatial coordinate data of the road surface, and extracts layer entities with reference to the precise distance information, and then classifies and identifies the layer entities, so as to achieve more efficient and more accurate three-dimensional data acquisition of road surface conditions, which is conducive to analyzing and extracting the precise three-dimensional coordinates of road surface disease areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is a general flow chart of the three-dimensional data acquisition and fusion method for road surface detection of the present invention.

[0048] Figure 2 This is a framework diagram of the three-dimensional data acquisition and fusion system for road surface detection of the present invention. DETAILED DESCRIPTION

[0049] The present invention will be further described in detail below in conjunction with the accompanying drawings.

[0050] The present invention provides a three-dimensional data acquisition and fusion method based on road surface detection, and its specific process is as follows: Figure 1 As shown, including:

[0051] S100: During the inspection of the road surface, the laser radar is used to collect the three-dimensional point cloud spatial coordinate data of the road surface in real time; and the distance encoder is used to measure the precise distance information.

[0052] In order to implement step S100, an embedded integrated sensing unit is used to integrate the high-precision laser radar and the distance encoder, which can realize the raw data collection function of multiple sensors and perform necessary preprocessing and fusion processing. The integration realizes the processing function of a large amount of point cloud data or precise distance data, and then by fusing data from different sensors, more comprehensive and accurate data collection can be achieved.

[0053] S200: Based on the three-dimensional point cloud spatial coordinate data of the road surface, perform adaptive meshing to obtain the three-dimensional mesh primitives of the road surface.

[0054] Specifically, for the road surface three-dimensional point cloud spatial coordinate data, the specific process of performing adaptive meshing includes:

[0055] S200A, dividing the spatial coordinate information of the road surface three-dimensional point cloud to form an initial grid; the initial grid is a polygon formed by connecting the feature points of the road surface three-dimensional point cloud as grid vertices, and the grid vertices can be determined based on the Delaunay triangulation algorithm to form a triangular initial grid;

[0056] S200B, performing regularization processing on the initial mesh, including supplementing the missing edges of the missing edge polygons, inserting new mesh vertices at the intersection points of the edges of the polygons, and merging the edges of the overlapping polygons;

[0057] S200C, for the initial mesh after the regularization processing, adaptive mesh subdivision is performed according to the mesh size and mesh quality index; specifically, firstly, the meshes to be subdivided in the initial mesh after the regularization processing that need to be further adaptively subdivided are locked, and the meshes to be subdivided are meshes in the initial mesh whose any side length is greater than the set side length threshold, and any internal angle of the polygon is less than the angle threshold; then the center of the circumscribed circle of the mesh to be subdivided is inserted into the mesh to be subdivided as an inserted mesh vertex, and then the inserted mesh vertex and other mesh vertices of the mesh to be subdivided are connected to realize adaptive mesh subdivision; repeat the above steps until all the meshes to be subdivided are adaptively subdivided.

[0058] Back to Figure 1 , S300: For the three-dimensional grid primitives included in the road surface, refer to the precise distance information measured by the distance encoder for the three-dimensional grid primitives and characterize them as a layer entity set.

[0059] Specifically, for the road surface, the three-dimensional grid primitive finally divided in step S200 is used as the basic unit and represented as a layer entity ,in, Represents a set of layer entities of the road surface 3D grid primitive reference plane, and maps the road surface 3D grid primitive reference plane to the layer entities in the set according to its spatial position . and The layer entity sets respectively represent the lower adjacent layer and the upper adjacent layer of the reference plane of the road surface three-dimensional grid primitive.

[0060] The specific process of characterizing the above layer entity set includes: for the three-dimensional grid primitive, according to the precise distance information measured by the distance encoder within the range of the three-dimensional grid primitive, the three-dimensional grid primitive is divided into layer entities; the three-dimensional point cloud spatial coordinates and their corresponding precise distance information within the range of the three-dimensional grid primitive are traversed, and when the distance difference between the precise distance information of the three-dimensional point cloud spatial coordinates is greater than a threshold, they are attributed to different layer entities, thereby forming a layer entity set.

[0061] S400: Using the BiLSTM network to extract features from the layer entity set of the road surface, obtain the layer feature vector of the three-dimensional grid primitive of the road surface; then, based on the prototype network model trained with the support set, obtain the category of the three-dimensional grid primitive of the road surface according to the layer feature vector of the three-dimensional grid primitive of the road surface.

[0062] Specifically, by inputting the layer entity set of the road surface into the BiLSTM network as the input, the layer feature vector of the road surface three-dimensional grid primitive can be obtained, which is expressed as ,Right now

[0063]

[0064] in It means that the layer entity set is input into BiLSTM for feature extraction, so as to map the layer entity combination into a layer feature vector of a road surface three-dimensional grid primitive.

[0065] The prototype network is a model that can achieve feature classification after training with a small number of support set samples. After training, the prototype network model can make objects of the same category more similar in features, while objects of different categories have greater differences in feature representation. The support set of the prototype network can be expressed as ,in represents the sample features in the support set, Represents the category attributes corresponding to each sample in the support set. Assume that there are categories, each of which is represented by ,but , then for each category Support sets can be set , support set The sample category attributes of the samples in are , then calculate each category through the support set Prototype:

[0066]

[0067] in represents the feature vector obtained by extracting the sample features, which is the feature extraction performed by the BiLSTM network in the present invention. As a prototype, The average representation of the feature vectors of samples in a class.

[0068] Furthermore, the prototype network can calculate the layer feature vector of the input road 3D grid primitive Relative to Each of the categories Distribution:

[0069]

[0070] in express categories do not belong to the category Thus, the prototype network can be based on the layer feature vector of the road surface three-dimensional grid primitive. Relative to Each of the categories The distribution of pavement 3D grid primitives is determined to determine the characteristic categories.

[0071] S500: According to the category of the pavement three-dimensional grid primitive, determine and mark the pavement three-dimensional grid primitive corresponding to the pavement disease category.

[0072] Specifically, the feature type of the road surface 3D mesh primitive reference plane includes the feature category corresponding to the road surface disease. Therefore, according to the feature category of the road surface 3D mesh primitive determined by the prototype network, when the feature category of a road surface 3D mesh primitive is identified as belonging to the road surface disease category, the road surface 3D mesh primitive is marked.

[0073] To implement the above method, see Figure 2 The present invention provides a three-dimensional data acquisition and fusion system for road surface detection, which is characterized by comprising:

[0074] The embedded integrated sensing unit integrates a laser radar and a distance encoder, which is used to collect the three-dimensional point cloud spatial coordinate data of the road surface in real time using the laser radar during the inspection process, and to measure the precise distance information using the distance encoder;

[0075] The point cloud decomposition module performs adaptive meshing based on the three-dimensional point cloud spatial coordinate data of the road surface to obtain the three-dimensional mesh primitives of the road surface;

[0076] A pavement primitive layering module is used to characterize the three-dimensional grid primitives contained in the pavement as a layer entity set with reference to the precise distance information measured by the distance encoder for the three-dimensional grid primitives;

[0077] The pavement primitive feature recognition module uses the BiLSTM network to extract features from the pavement layer entity set to obtain the layer feature vector of the pavement three-dimensional grid primitive; and then uses the prototype network to identify the category of the pavement three-dimensional grid primitive based on the layer feature vector;

[0078] The pavement primitive disease category marking module determines and marks the pavement three-dimensional grid primitive corresponding to the pavement disease category according to the category of the pavement three-dimensional grid primitive.

[0079] Preferably, the point cloud decomposition module performs adaptive meshing on the road surface three-dimensional point cloud spatial coordinate data, specifically including:

[0080] The spatial coordinate information of the road surface three-dimensional point cloud is divided into mesh vertices based on the Delaunay triangulation algorithm to form a triangular initial mesh. The initial mesh is a polygon formed by connecting the feature points of the road surface three-dimensional point cloud as mesh vertices;

[0081] Regularize the initial grid;

[0082] For the initial grid after regularization, adaptive grid subdivision is performed according to the grid size and grid quality index.

[0083] Preferably, the regularization of the initial mesh includes: supplementing the missing edges of the missing polygons, inserting new mesh vertices at the intersections of the edges of the polygons, and merging the edges of the overlapping polygons.

[0084] Preferably, the pavement primitive layering module divides the three-dimensional grid primitive into layer entities according to the precise distance information measured by the distance encoder within the range of the three-dimensional grid primitive; traverses the three-dimensional point cloud spatial coordinates and their corresponding precise distance information within the range of the three-dimensional grid primitive, and when the distance difference between the precise distance information of the three-dimensional point cloud spatial coordinates is greater than a threshold, attributes them to different layer entities, thereby forming a layer entity set.

[0085] Preferably, the pavement primitive feature recognition module uses a BiLSTM network to extract features from a pavement layer entity set to obtain a layer feature vector of a pavement three-dimensional grid primitive, specifically comprising: by inputting the pavement layer entity set as an input into the BiLSTM network, a layer feature vector of a pavement three-dimensional grid primitive can be obtained, which is expressed as ,Right now

[0086]

[0087] in It means that the layer entity set is input into BiLSTM for feature extraction, so that the layer entity combination is mapped into a layer feature vector of a road surface three-dimensional grid primitive;

[0088] in, Represents a set of layer entities of the road surface 3D grid primitive reference plane, and maps the road surface 3D grid primitive reference plane to the layer entities in the set according to its spatial position . and The layer entity sets respectively represent the lower adjacent layer and the upper adjacent layer of the reference plane of the road surface three-dimensional grid primitive.

[0089] Preferably, the prototype network calculates the layer feature vector of the input road surface three-dimensional grid primitive Relative to Each of the categories Distribution:

[0090]

[0091] in express categories do not belong to the category Other categories; the prototype network is based on the layer feature vector of the road 3D grid primitive Relative to Each of the categories The distribution of pavement 3D grid primitives is determined to determine the characteristic categories.

[0092] Beneficial effects of the present invention:

[0093] The present invention integrates the three-dimensional point cloud spatial coordinate data of the road surface and the precise distance information formed by the distance encoder, and utilizes the advantage of the distance encoder's low sensitivity to ambient light and particulate matter, so as to maintain a high measurement accuracy in harsh environments; the distance information obtained by the distance encoder can help correct the measurement error of the laser radar, especially in long-distance measurement; the addition of the distance encoder helps to improve the problems caused by the uneven reflectivity of the road surface, and reduces the impact of this unevenness on the acquisition results by optimizing the distance information; further, the present invention grids the three-dimensional point cloud spatial coordinate data of the road surface, and extracts layer entities with reference to the precise distance information, and then classifies and identifies the layer entities, so as to achieve more efficient and more accurate three-dimensional data acquisition of road surface conditions, which is conducive to analyzing and extracting the precise three-dimensional coordinates of road surface disease areas.

[0094] The above are only preferred implementations of the present invention. It should be pointed out that a number of modifications and improved technical solutions made by those skilled in the art without departing from the technical solution should also be deemed to fall within the scope of protection required by the claims.

Claims

1. A three-dimensional data acquisition and fusion method for road surface detection, characterized in that: The following steps are involved: During the inspection of the road surface, the laser radar is used to collect the three-dimensional point cloud spatial coordinate data of the road surface in real time, and the distance encoder is used to measure the precise distance information; Based on the spatial coordinate data of the three-dimensional point cloud of the road surface, adaptive mesh division is performed to obtain the three-dimensional mesh primitive of the road surface, including: dividing the spatial coordinate information of the three-dimensional point cloud of the road surface based on the Delaunay triangulation algorithm to determine the mesh vertices to form an initial triangular mesh, wherein the initial mesh is a polygon formed by connecting the feature points of the three-dimensional point cloud of the road surface as mesh vertices; Regularize the initial grid; For the initial grid after regularization, adaptive grid subdivision is performed according to the grid size and grid quality index; The regularization of the initial mesh includes: supplementing the missing edges of the missing polygons, inserting new mesh vertices at the intersections of the edges of the polygons, and merging the edges of the overlapping polygons; For the three-dimensional grid primitives contained in the road surface, with reference to the precise distance information measured by the distance encoder for the three-dimensional grid primitives, the three-dimensional grid primitives are characterized as a layer entity set, wherein, for the three-dimensional grid primitives, according to the precise distance information measured by the distance encoder within the range of the three-dimensional grid primitives, the three-dimensional grid primitives are divided into layer entities; the three-dimensional point cloud spatial coordinates within the range of the three-dimensional grid primitives and their corresponding precise distance information are traversed, and when the distance difference between the precise distance information of the three-dimensional point cloud spatial coordinates is greater than a threshold, they are attributed to different layer entities, thereby forming a layer entity set; The BiLSTM network is used to extract features from the layer entity set of the road surface to obtain the layer feature vector of the three-dimensional grid primitive of the road surface; and then the prototype network is used to identify the category of the three-dimensional grid primitive of the road surface based on the layer feature vector; According to the category of the pavement three-dimensional grid primitive, the pavement three-dimensional grid primitive corresponding to the pavement disease category is determined and marked.

2. The three-dimensional data acquisition and fusion method for road surface detection according to claim 1 is characterized in that: The BiLSTM network is used to extract features from the layer entity set of the road surface to obtain the layer feature vector of the three-dimensional grid primitive of the road surface. Specifically, the layer entity set of the road surface is input into the BiLSTM network as an input to obtain the layer feature vector of the three-dimensional grid primitive of the road surface, which is expressed as ,Right now in It means that the layer entity set is input into BiLSTM for feature extraction, so that the layer entity combination is mapped into a layer feature vector of a road surface three-dimensional grid primitive; in, Represents a set of layer entities of the road surface 3D grid primitive reference plane, and maps the road surface 3D grid primitive reference plane to the layer entities in the set according to its spatial position and The layer entity sets respectively represent the lower adjacent layer and the upper adjacent layer of the reference plane of the road surface three-dimensional grid primitive.

3. A three-dimensional data acquisition and fusion system for road surface detection, characterized in that: include: The embedded integrated sensing unit integrates a laser radar and a distance encoder, which is used to collect the three-dimensional point cloud spatial coordinate data of the road surface in real time using the laser radar during the inspection process, and to measure the precise distance information using the distance encoder; The point cloud decomposition module performs adaptive meshing based on the three-dimensional point cloud spatial coordinate data of the road surface to obtain the three-dimensional mesh primitives of the road surface. The point cloud decomposition module performs adaptive meshing for the three-dimensional point cloud spatial coordinate data of the road surface, specifically including: The spatial coordinate information of the road surface three-dimensional point cloud is divided into mesh vertices based on the Delaunay triangulation algorithm to form a triangular initial mesh. The initial mesh is a polygon formed by connecting the feature points of the road surface three-dimensional point cloud as mesh vertices; Regularize the initial grid; For the initial grid after regularization, adaptive grid subdivision is performed according to the grid size and grid quality index; The regularization of the initial mesh includes: supplementing the missing edges of the missing polygons, inserting new mesh vertices at the intersections of the edges of the polygons, and merging the edges of the overlapping polygons; A pavement primitive layering module is used to characterize the three-dimensional grid primitives contained in the pavement as a layer entity set with reference to the precise distance information measured by the distance encoder for the three-dimensional grid primitives; The pavement primitive feature recognition module uses the BiLSTM network to extract features from the pavement layer entity set to obtain the layer feature vector of the pavement three-dimensional grid primitive; then uses the prototype network to identify the category of the pavement three-dimensional grid primitive based on the layer feature vector; the pavement primitive stratification module divides the three-dimensional grid primitive into layer entities according to the precise distance information measured by the distance encoder within the range of the three-dimensional grid primitive; traverses the three-dimensional point cloud space coordinates within the range of the three-dimensional grid primitive and the corresponding precise distance information, and when the distance difference between the precise distance information of the three-dimensional point cloud space coordinates is greater than a threshold, it is attributed to different layer entities, thereby forming a layer entity set; The pavement primitive disease category marking module determines and marks the pavement three-dimensional grid primitive corresponding to the pavement disease category according to the category of the pavement three-dimensional grid primitive.

4. The three-dimensional data acquisition and fusion system for road surface detection according to claim 3 is characterized in that: The pavement primitive feature recognition module uses the BiLSTM network to extract features from the pavement layer entity set to obtain the layer feature vector of the pavement three-dimensional grid primitive, which specifically includes: inputting the pavement layer entity set as input into the BiLSTM network to obtain the layer feature vector of the pavement three-dimensional grid primitive, which is expressed as ,Right now in It means that the layer entity set is input into BiLSTM for feature extraction, so that the layer entity combination is mapped into a layer feature vector of a road surface three-dimensional grid primitive; in, Represents a set of layer entities of the road surface 3D grid primitive reference plane, and maps the road surface 3D grid primitive reference plane to the layer entities in the set according to its spatial position and The layer entity sets respectively represent the lower adjacent layer and the upper adjacent layer of the reference plane of the road surface three-dimensional grid primitive.

Images