Road infrastructure BIM model reverse construction method based on data fusion

Abstract

The invention discloses a road infrastructure BIM model reverse construction method based on data fusion, and relates to the technical field of building design, and the method comprises the steps: obtaining first point cloud data of a plurality of road regions and infrastructures thereof; performing segmentation processing on the first point cloud data to obtain a plurality of first road sections; based on a random sampling consensus algorithm, performing local optimal matching on the point cloud data in the first road section and standard point cloud data of the infrastructure to obtain target point cloud data; screening the target point cloud data based on the reproduction degrees of the target point cloud data at different road positions so as to carry out denoising processing on the target point cloud data; and semantic segmentation and instance extraction processing are performed on the denoised target point cloud data through a deep learning algorithm, and a target road model containing different infrastructures is constructed. According to the method and the device, the road facilities in the constructed target road model better conform to the actual road condition.

Description

Technical Field

[0001] This application relates to the field of architectural design technology, and in particular to a method for reverse engineering a BIM model of road infrastructure based on data fusion. Background Technology

[0002] Road infrastructure refers to a series of fixed facilities that provide basic conditions for the passage of vehicles and pedestrians. Reverse modeling (Scan-to-BIM) technology digitally restores existing urban road facilities (such as roads, bridges, pipelines, etc.) to form a building information model (BIM) containing accurate geometric information and rich attribute information. It is of great significance for the operation and maintenance management of existing facilities, the planning and design of renovation and expansion, and the construction of smart cities.

[0003] Scan-to-BIM technology involves scanning existing urban road infrastructure into point clouds, performing initial denoising and registration on the point cloud data, and then using deep learning algorithms to segment the point cloud data into different road facilities, such as bridges, fences, and pipelines. However, the actual point cloud scanning process is affected by the non-smooth surfaces and structural errors of real-world infrastructure, such as infill particles on the road surface. This results in noise in the scanned 3D point cloud data, causing the road facilities segmented by the deep learning algorithm to differ from the actual infrastructure. Summary of the Invention

[0004] The main purpose of this application is to provide a method for reverse construction of BIM models of road infrastructure based on data fusion. This method aims to solve the technical problem in related technologies where, after scanning the point cloud of existing urban road infrastructure, the noise generated by the filling particles on the road surface cannot be removed by the preliminary denoising method, resulting in the road facilities divided by the deep learning algorithm not matching the actual situation.

[0005] To achieve the above objectives, embodiments of this application provide a method for reverse engineering a BIM model of road infrastructure based on data fusion, including: Acquire first-point cloud data of multiple road areas and their infrastructure; The first point cloud data is segmented to obtain multiple first road segments; Based on the random sampling consensus algorithm, the point cloud data in the first road segment is locally optimally matched with the standard point cloud data of the infrastructure to obtain the target point cloud data. Based on the reproducibility of each target point cloud data at different road locations, the target point cloud data is filtered to perform noise reduction processing. By using deep learning algorithms to perform semantic segmentation and instance extraction on the denoised target point cloud data, a target road model containing different infrastructure is constructed.

[0006] In one possible implementation of this application, the first point cloud data is segmented to obtain multiple first road segments, including: Along the road direction corresponding to the road area, the vertical plane perpendicular to the road direction is taken as the road section of each road area; Determine the number of point clouds corresponding to the first point cloud data at the data collection location within each road area, and use the number of point clouds as the cross-sectional density of each road section; Clustering of cross-sectional densities yields multiple first road segments.

[0007] In one possible embodiment of this application, the cross-sectional density is clustered to obtain multiple first road segments, including: Calculate the mean density of the cross-sectional density for all road areas, and use the mean density as the clustering threshold; If the difference in the mean cross-sectional density between any two clusters is greater than the clustering threshold, then these two clusters are determined to be divided into two first road segments, until all road areas are clustered, resulting in multiple first road segments.

[0008] In one possible implementation of this application, based on a random sampling consensus algorithm, local optimal matching is performed between the point cloud data in the first road segment and the standard point cloud data of the infrastructure to obtain target point cloud data, including: A predetermined number of second point cloud data are extracted from each first point cloud data using a random sampling consistency algorithm; Set the distance between the second point cloud data and the nearest point cloud data in the standard point cloud data corresponding to all infrastructures as the shortest matching distance of the second point cloud data. Sort the shortest matching distances corresponding to all second point cloud data in descending order to obtain the point cloud distance sorting sequence, and select the optimal matching sequence segment in the point cloud distance sorting sequence; Based on the point cloud data corresponding to the optimal matching sequence segment and the shortest matching distance, the reliability of the point cloud data at the current iteration number is calculated. If the reliability level is greater than the preset threshold, the point cloud data of the current iteration number will be used as the target point cloud data.

[0009] In one possible implementation of this application, selecting the optimal matching sequence segment from the point cloud distance sorting sequence includes: Calculate the distance difference between the two shortest matching distances adjacent to each other in the point cloud sorting sequence; Based on the location corresponding to the maximum value of the distance difference, the point cloud distance sorting sequence is divided to obtain the first sequence segment and the second sequence segment; The segment corresponding to the minimum average distance between the first and second sequence segments is taken as the optimal matching sequence segment.

[0010] In one possible implementation of this application, the reliability of the point cloud data at the current iteration number is calculated based on the point cloud data corresponding to the optimal matching sequence segment and the shortest matching distance, including: Determine the first proportion of point cloud data corresponding to the optimal matching sequence segment, and the first average value of each shortest matching distance; The reliability of the point cloud data at the current iteration number is calculated based on the ratio between the first proportion and the first average value.

[0011] In one possible implementation of this application, the target point cloud data is filtered based on the degree of reproducibility of each target point cloud data at different road locations, including: Along the road direction corresponding to the road area, construct slice planes at multiple road locations in each road area; Based on the distance between target point cloud data in different slice planes, the suspected reproducing point cloud data of each target point cloud data is determined; Based on the number of suspected reproducible point cloud data and the number of slice planes, the degree of reproducibility of each target point cloud data at different road locations is calculated. Determine the spatial distance between each target point cloud data and the fitted model constructed by the random sampling consistency algorithm in the current iteration number; Based on the degree of reproducibility, the spatial distance is weakened to obtain the weakened distance; Based on the weakened distance, the target point cloud data is divided into interior points and exterior points, and the interior points are retained.

[0012] In one possible implementation of this application, based on the distance between target point cloud data in different slice planes, the suspected reproducing point cloud data of each target point cloud data is determined, including: For any slice plane, project the target point cloud data between the current slice plane and the next adjacent slice plane onto the current slice plane, until point cloud data exists on all slice planes; Project the point cloud data of each slice plane in each road region onto the first slice plane with the road center as the origin; For any target point cloud data in the first slice plane, the point cloud data in the preset neighboring region of the current target point cloud data is taken as the suspected reproduction point cloud data of the current target point cloud data.

[0013] In one possible implementation of this application, the degree of reproduction of each target point cloud data at different road locations is calculated based on the number of suspected reproducible point cloud data and the number of slice planes, including: Arrange all suspected reproducible point cloud data of different target point cloud data in the order of road direction, and determine the number of planar intervals between any two adjacent suspected reproducible point cloud data of each target point cloud data. Calculate the average distance between each target point cloud data and its corresponding suspected reproducing point cloud data; Based on the ratio between the mean distance and the number of planar intervals, and the ratio between the number of suspected reproducible point cloud data and the number of slice planes, the degree of reproducibility of each target point cloud data at different road locations is calculated.

[0014] In one possible implementation of this application, the target point cloud data is divided into interior points and exterior points based on the weakened distance, and the interior points are retained, including: The weakened distance is compared with a preset distance threshold. If the reduced distance is less than the preset distance threshold, the target point cloud data corresponding to the reduced distance is classified as an interior point; otherwise, it is classified as an exterior point. Determine the iteration number with the largest number of interior points among multiple iterations of the random sampling consensus algorithm, retain the interior points at that iteration number, and filter out all exterior points at that iteration number.

[0015] This application provides a method for reverse engineering a BIM model of road infrastructure based on data fusion. Compared to related technologies where preliminary denoising methods fail to remove noise caused by surface particles after scanning point clouds of existing urban road infrastructure, leading to discrepancies between the road facilities segmented by deep learning algorithms and reality, this application acquires first point cloud data of multiple road areas and their infrastructure. By segmenting each first point cloud data point, multiple first road segments are obtained. A random sampling consensus algorithm is used to perform local optimal matching between the point cloud data in the first road segments and the standard point cloud data of the infrastructure to obtain target point cloud data. Based on the reproducibility of the target point cloud data at different road locations, the target point cloud data is filtered for denoising. Since the infrastructure and roads are embedded, this method allows for filtering of point cloud data within embedded structures, thus denoising the target point cloud data. Finally, a deep learning algorithm is used to perform semantic segmentation and instance extraction on the denoised target point cloud data to reconstruct a target road model containing different infrastructures. This results in road facilities in the segmented target road model that better reflect actual road conditions. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the first embodiment of the reverse engineering method for road infrastructure BIM models based on data fusion in this application. Figure 2 This is a flowchart illustrating the second embodiment of the data fusion-based road infrastructure BIM model reverse construction method of this application; Figure 3 This is a schematic diagram of the device structure of the hardware operating environment involved in the embodiments of this application. Detailed Implementation

[0017] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0018] This application provides a method for reverse engineering a road infrastructure BIM model based on data fusion. In the first embodiment of this method, referring to... Figure 1 The methods include: Step S10: Obtain first point cloud data of multiple road areas and their infrastructure; Step S20: The first point cloud data is segmented to obtain multiple first road segments; Step S30: Based on the random sampling consensus algorithm, perform local optimal matching between the point cloud data in the first road segment and the standard point cloud data of the infrastructure to obtain the target point cloud data; Step S40: Based on the reproducibility of each target point cloud data at different road locations, the target point cloud data is filtered to perform noise reduction processing on the target point cloud data; Step S50: Semantic segmentation and instance extraction are performed on the denoised target point cloud data using deep learning algorithms to construct a target road model containing different infrastructures.

[0019] This embodiment aims to filter target point cloud data based on the degree of reproducibility of the target point cloud data at different road locations, and to perform noise reduction processing on the target point cloud data. The infrastructure and roads are installed in an embedded manner. In this way, the point cloud data of mutually embedded structures can be filtered, thereby performing noise reduction processing on the target point cloud data, so that the road facilities in the divided target road model are more consistent with the actual road conditions.

[0020] The specific steps are as follows: Step S10: Obtain first point cloud data for multiple road areas and their infrastructure.

[0021] As an example, the method for reverse engineering a road infrastructure BIM model based on data fusion can be applied to a reverse engineering device for a road infrastructure BIM model based on data fusion. This reverse engineering device for a road infrastructure BIM model based on data fusion belongs to a reverse engineering system for a road infrastructure BIM model based on data fusion, and this reverse engineering system for a road infrastructure BIM model based on data fusion belongs to a reverse engineering equipment for a road infrastructure BIM model based on data fusion.

[0022] As an example, the first point cloud data can be point cloud data of various road areas and various infrastructures in the road areas collected by data acquisition equipment such as high-precision cameras and 3D laser scanning. The infrastructure can be various road elements, such as lane lines, guardrails, signs, roads, and street lights.

[0023] As an example, the first point cloud data can be collected in the following ways: By installing a lidar on the top of the vehicle and a high-precision GPS device inside the vehicle, the vehicle moves along the road direction and obtains the road route through the high-precision GPS device during the movement. Based on the width of the road where the vehicle is located, the GPS data collected by the lidar is selected within a local area centered on the vehicle and the width of the road, in order to avoid the influence of other interferences outside the road infrastructure on the point cloud data.

[0024] Since road infrastructure needs to be calibrated and semantically segmented, the types of existing road facilities are obtained in advance. In this embodiment of the application, all types of known facilities on the currently collected road, as well as the standard point cloud data of each infrastructure, are also read.

[0025] The above-mentioned first point cloud data are represented in the following way: That is, represents the i-th point cloud data. Its three-dimensional coordinates.

[0026] Step S20: The first point cloud data is segmented to obtain multiple first road segments.

[0027] As an example, different road sections have different facilities installed. For instance, the boundary between urban roads and non-road areas is marked by curbs, while the boundary between urban and non-road areas is marked by fences, and the boundary between bridges and non-road areas is marked by concrete fences. In addition, roads also include signs, traffic lights, etc. The point cloud data collected varies along the direction of the road (i.e., the main direction of the point cloud data) in different road sections. Therefore, based on the main direction of each first point cloud data, the direction of the road is determined, and each road is segmented according to the point cloud density of different road sections to obtain multiple first road segments. Here, a first road segment represents multiple road segments obtained after segmenting based on the first point cloud data, thereby achieving the purpose of accurately dividing each road.

[0028] Step S30: Based on the random sampling consensus algorithm, perform local optimal matching between the point cloud data in the first road segment and the standard point cloud data of the infrastructure to obtain the target point cloud data.

[0029] As an example, the random sampling consensus algorithm is used to randomly sample a small subset of all data to fit an initial model, and then use this model to test all other data points to see how many points support (fit) the model. This process is repeated many times, and finally the model supported by the most data points is selected as the best model.

[0030] In related technologies, a random sampling consensus algorithm is used to select a certain number of point cloud data points locally and then fit these point cloud data points. Normal road infrastructure mainly consists of lines and surfaces with obvious structures, so it can be fitted after random sampling. Points that do not conform to the fitted points are taken as outliers, and point cloud data that conform to the fitted model are taken as inliers. Inliers are the points that conform. However, in this process, roads are not straight, and some structures may appear randomly, such as signs. At the same time, road infrastructure is usually connected to roads or other structures in an embedded manner. For example, streetlights, signs, and fences are all connected by being embedded in the foundation. This leads to different point cloud data at different structures. The random sampling consensus algorithm uses multiple iterations to select the most inliers to obtain the final fitted model. When encountering complex road structures, there are many outliers. This causes outliers to participate in the fitting model during the fitting process, resulting in model errors, thus increasing the number of iterations to obtain inliers.

[0031] In this embodiment, the present application utilizes standard point cloud data of various known infrastructures, and through local optimal matching, combined with the reproducibility of facilities in the road direction, slices the first point cloud data collected in the road direction to obtain multi-layer point data. Then, by combining the multi-layer point cloud data and the standard point cloud data, data denoising is obtained in the reverse modeling process through data fusion.

[0032] As an example, the presence of lane lines and rough road surface structures on roads can lead to a large number of interference points. Analyzing and calculating all interference points would result in a huge amount of computation, which would reduce the efficiency of the random sampling consensus algorithm and Scan-to-BIM. Therefore, in this embodiment, when selecting local points using the random sampling consensus algorithm, the target point cloud data to be analyzed and calculated is selected by judging the local optimal match between the selected first point cloud data and the standard point cloud data.

[0033] Step S30 includes: Step S31: Extract a preset number of second point cloud data from each first point cloud data using a random sampling consensus algorithm.

[0034] As an example, the preset quantity can be hundreds, thousands, etc., and there is no specific limitation. The second point cloud data is a portion of the point cloud data from all the first point cloud data.

[0035] Step S32: Set the distance between the second point cloud data and the nearest point cloud data in the standard point cloud data corresponding to all infrastructures as the shortest matching distance of the second point cloud data.

[0036] As an example, the random sampling consensus algorithm has multiple iterations during the calculation process. Taking the current iteration as the p-th iteration as an example, for the point cloud data of the p-th iteration, the matching degree between it and the standard point cloud data of all infrastructure is calculated. The matching degree is defined as: the set of distances between each second point cloud data in the iterative point cloud data and its nearest neighbor in the standard point cloud data. The set of distances corresponding to each second point cloud data is denoted as the set of the shortest matching distances of each point cloud data in the p-th iteration.

[0037] Step S33: Sort the shortest matching distances corresponding to all second point cloud data in descending order to obtain the point cloud distance sorting sequence, and select the optimal matching sequence segment in the point cloud distance sorting sequence.

[0038] As an example, after calculating the shortest matching distance, the shortest matching distances corresponding to all the second point cloud data are sorted in descending order, that is, from largest to smallest. After sorting, a point cloud distance sorting sequence is obtained. Based on the changes of each shortest matching distance in the point cloud distance sorting sequence, the point cloud distance sorting sequence is divided, and the sequence segment with the smaller shortest matching distance is selected as the optimal matching sequence segment.

[0039] Step S33 further includes: Calculate the distance difference between the two shortest matching distances adjacent to each other in the point cloud distance sorting sequence.

[0040] Based on the location corresponding to the maximum distance difference, the point cloud distance sorting sequence is divided to obtain the first sequence segment and the second sequence segment.

[0041] As an example, the distance difference between each adjacent short matching distance in the point cloud distance sorting sequence can be calculated. The distance difference can be calculated by subtracting the next sequence value from the previous sequence value and determining the maximum value among the various distance differences, thereby locating the position corresponding to the maximum value of the distance difference.

[0042] As an example, after locating the position corresponding to the maximum value of the distance difference, the point cloud distance sorting sequence is divided into two parts at that position to obtain the first sequence segment and the second sequence segment. The position is the middle position of the two adjacent sequence values. After dividing the first sequence segment and the second sequence segment, the two sequence values ​​are also divided into the first sequence segment and the second sequence segment respectively.

[0043] The segment corresponding to the minimum average distance between the first and second sequence segments is taken as the optimal matching sequence segment.

[0044] As an example, the average distance of each sequence value in the first sequence segment and the average distance of each sequence value in the second sequence segment are calculated respectively. The sequence segment corresponding to the minimum of the two average distances is selected as the optimal matching sequence segment. This division method is a two-level division. The purpose of the two-level division is to avoid the influence of interference points. On the other hand, since the facilities and roads are embedded, there is a loss of point cloud data. As a result, there are unmatched points in the process of calculating the distance of the standard point cloud data, which increases the result. Therefore, only the part of the point cloud data with smaller distances is selected first.

[0045] Step S34: Based on the point cloud data corresponding to the optimal matching sequence segment and the shortest matching distance, calculate the reliability of the point cloud data for the current iteration number.

[0046] As an example, the reliability of the point cloud data in the current iteration is determined based on the proportion of the point cloud data corresponding to the optimal matching sequence segment in all the second point cloud data, and the shortest matching distance.

[0047] Step S34 includes: Determine the first proportion of point cloud data corresponding to the optimal matching sequence segment, and the first average value of each shortest matching distance.

[0048] The reliability of the point cloud data at the current iteration number is calculated based on the ratio between the first proportion and the first average value.

[0049] As an example, the first proportion is the proportion of the point cloud data corresponding to the optimal matching sequence segment in all point cloud data, and the first average is the average of the shortest matching distances in the optimal matching sequence segment.

[0050] As an example, the reliability of the point cloud data at the current iteration number. The calculation method can be:

[0051] in, Indicates the first percentage, The first average value is represented by norm(), which represents the normalization calculation. The reliability level is used to indicate the degree to which the current batch of point cloud data can be used for analysis and calculation. When the reliability level meets certain conditions, it means that the point cloud data of this batch is accurate.

[0052] Step S35: If the reliability is greater than the preset threshold, then the point cloud data of the current iteration number is used as the target point cloud data.

[0053] As an example, the preset threshold can be 0.68, 0.7, etc., and there is no specific limitation.

[0054] As an example, when the calculated reliability is greater than a preset threshold, the point cloud data of the current iteration number is selected as the target point cloud data, and the reproducibility of these target point cloud data is calculated. In this way, the target point cloud data is filtered to achieve the purpose of data denoising.

[0055] Step S40: Based on the reproducibility of each target point cloud data at different road locations, the target point cloud data is filtered to perform noise reduction processing on the target point cloud data.

[0056] As an example, reproducibility is used to indicate the regularity of point cloud data appearing at equal intervals at the same relative position on a road. A road may include multiple different road positions, and the distance between each road position may be 1 meter, 2 meters, etc., without any specific limitation.

[0057] As an example, one way to filter target point cloud data is to calculate the reproducibility of each target point cloud data at different road locations, and then use this reproducibility to reduce the spatial distance between each point cloud data and the fitted model constructed in the p-th iteration of the random sampling consensus algorithm. Finally, the reduced distance is used to divide the inner and outer points, thereby achieving the purpose of data filtering.

[0058] Step S40 includes: Step S41: Construct slice planes at multiple road locations in each road region along the road direction corresponding to the road region.

[0059] As an example, the road locations have been divided beforehand. Along the road direction (road centerline) corresponding to each road area, slice planes are constructed at each road location. The slice planes are constructed by taking the plane perpendicular to the road direction as the slice plane at that road location. In this application, the road locations are at preset intervals, taking a 1-meter interval as an example.

[0060] Step S42: Based on the distance between target point cloud data in different slice planes, determine the suspected reproducing point cloud data for each target point cloud data.

[0061] As an example, in different slice planes, based on the distance between each target point cloud data, suspected recurring point cloud data in a preset nearby area of ​​the target point cloud data are selected. The suspected recurring point cloud data indicates that the point cloud data may be the same point cloud data that reappears at the next road location. For example, signs are road infrastructure and will reappear after a certain interval.

[0062] Step S42 includes: For any slice plane, project the target point cloud data between the current slice plane and the next adjacent slice plane onto the current slice plane, until point cloud data exists on all slice planes.

[0063] As an example, after determining the slice plane at each road location, the target point cloud data between two adjacent slice planes (the current slice plane and the next slice plane) is projected onto the current slice plane until point cloud data exists on all slice planes. The last slice plane is not calculated.

[0064] The point cloud data of each slice plane in each road region is projected onto the first slice plane with the road center as the origin.

[0065] As an example, the road centerline is the skeleton of the road. If the target point cloud data belongs to the road infrastructure, then the road infrastructure is repeated, and the corresponding target point cloud data is also repeated. Therefore, after slicing, the target point cloud data corresponding to the repeated road facilities are located in the same plane, and the target point cloud data of the same feature point are located at similar intervals in the plane. Therefore, this application obtains the reproducibility of the target point cloud data by projecting the target point cloud data onto a plane and then by the average number of intervals of the point cloud data in the same plane at similar positions.

[0066] As an example, the point cloud data of each slice plane in each road area is projected onto the same slice plane with the road center as the origin, that is, the first slice plane. At this time, the point cloud data of all slice planes are projected onto one slice plane, so as to determine the suspected reproducing point cloud data of each target point cloud data located on different slice planes on the same slice plane.

[0067] For any target point cloud data in the first slice plane, the point cloud data in the preset neighboring region of the current target point cloud data is taken as the suspected reproduction point cloud data of the current target point cloud data.

[0068] As an example, the preset neighboring region can be a circular region extended with a preset radius, which can be 0.1 meters. Then, for any target point cloud data in the first slice plane, the point cloud data in the preset neighboring region of the current target point cloud data will be used as the suspected reproduction point cloud data of the current target point cloud data.

[0069] It should be noted that since there may be multiple point cloud data in the preset neighboring area, multiple point cloud data may be selected as suspected reproduction point cloud data in the same slice plane. Therefore, it is necessary to filter the suspected reproduction point cloud data. The filtering method is as follows: if there are multiple point cloud data in the same slice plane as suspected reproduction point cloud data, then only the point cloud data that is closest to the current target point cloud data in the same slice plane is selected as the suspected reproduction point cloud data. This ensures that for the same target point cloud data, there is at most one suspected reproduction point cloud data in each slice plane.

[0070] Step S43: Based on the number of suspected reproduced point cloud data and the number of slice planes, calculate the reproduction degree of each target point cloud data at different road locations.

[0071] As an example, for each target point cloud data, there will be multiple suspected reproducible point cloud data. First, based on the number of suspected reproducible point cloud data, the number of slice planes, and the regularity of the slice plane distribution between two adjacent suspected reproducible point cloud data, the degree of reproducibility of each target point cloud data at different road locations is calculated.

[0072] Step S43 includes: All suspected reproducible point cloud data of different target point cloud data are arranged in order of road direction, and the number of planar intervals between any two adjacent suspected reproducible point cloud data of each target point cloud data is determined.

[0073] As an example, all suspected reproducible point cloud data of each target point cloud data located on different slice planes are arranged in the order of road direction, and the number of planar intervals between any two adjacent suspected reproducible point cloud data on the first slice plane is determined, wherein the number of planar intervals is the number of slice planes that separate two adjacent suspected reproducible point cloud data.

[0074] Calculate the average distance between each target point cloud data and its corresponding suspected reproduced point cloud data.

[0075] Based on the ratio between the mean distance and the number of planar intervals, and the ratio between the number of suspected reproducible point cloud data and the number of slice planes, the degree of reproducibility of each target point cloud data at different road locations is calculated.

[0076] As an example, there is a spatial distance between each target point cloud data and its corresponding suspected reproduction point cloud data. The average of these spatial distances is calculated to obtain the mean distance.

[0077] As an example, taking the m-th target point cloud data of the first slice plane as an example, its reproducibility... The calculation method is as follows:

[0078] in, This represents the number of suspected reproducing point cloud data points for the m-th target point cloud data in the first slice plane. This represents the total number of slice planes within this road segment. A larger value indicates higher reproducibility; The acquisition method is as follows: Arrange all suspected reproducible point cloud data in order of road direction, calculate the number of planar intervals between two adjacent suspected reproducible point cloud data, and denot the kurtosis of all planar intervals as . This indicates the clustering of the number of planar intervals; the larger the value, the more obvious the clustering, indicating a regular arrangement. This is the average distance between the suspected reproducible point cloud data and the m-th point cloud data. The smaller the value, the stronger the reproducibility. +1 is used to prevent the denominator from being zero.

[0079] Step S44: Determine the spatial distance between each target point cloud data and the fitted model constructed by the random sampling consensus algorithm in the current iteration number.

[0080] Step S45: Based on the degree of reproducibility, the spatial distance is weakened to obtain the weakened distance.

[0081] As an example, the spatial distance between each target point cloud data and the fitted model constructed by the random sampling consensus algorithm in the current iteration is determined, and this spatial distance can be directly obtained in the current iteration.

[0082] As an example, by calculating the reproducibility of each target point cloud data, the corresponding spatial distance is negatively weighted and calculated to weaken the spatial distance, resulting in a weakened distance. The negatively weighted calculation method can be: weakened distance = spatial distance (1 - reproducibility).

[0083] Step S46: Based on the weakened distance, divide the target point cloud data into interior points and exterior points, and retain the interior points.

[0084] Step S46 includes: The weakened distance is compared with a preset distance threshold.

[0085] If the reduced distance is less than the preset distance threshold, the target point cloud data corresponding to the reduced distance is classified as an interior point; otherwise, it is classified as an exterior point.

[0086] As an example, after obtaining the weakened distance, the weakened distance corresponding to each target point cloud data is compared with a preset distance threshold. If the weakened distance is less than the preset distance threshold, the target point cloud data is classified as an interior point; otherwise, it is classified as an exterior point.

[0087] Determine the iteration number with the largest number of interior points among multiple iterations of the random sampling consensus algorithm, retain the interior points at that iteration number, and filter out all exterior points at that iteration number.

[0088] As an example, in the random sampling consensus algorithm, there are multiple iterations. The fitted model with the most inliers is selected, the inliers are retained, and the outliers are filtered out.

[0089] Step S50: Semantic segmentation and instance extraction are performed on the denoised target point cloud data using deep learning algorithms to construct a target road model containing different infrastructures.

[0090] As an example, after selecting the final denoised target point cloud data, the point cloud data is semantically segmented and instance extracted using a deep learning algorithm (such as a point cloud panoramic segmentation model) to automatically identify road elements, thereby constructing a target road model containing different infrastructure. The processing method of the deep learning algorithm is an existing technology and will not be elaborated here.

[0091] The above method is used to denoise all point cloud data to eliminate external points. Using standard point cloud data as a training set, a deep learning algorithm is trained to perform semantic segmentation on the denoised point cloud data, thereby obtaining various infrastructures in the road and realizing reverse modeling of Scan-to-BIM using point cloud data.

[0092] This application provides a method for reverse engineering a BIM model of road infrastructure based on data fusion. Compared to related technologies where preliminary denoising methods fail to remove noise caused by surface particles after scanning point clouds of existing urban road infrastructure, leading to discrepancies between the road facilities segmented by deep learning algorithms and reality, this application acquires first point cloud data of multiple road areas and their infrastructure. By segmenting each first point cloud data point, multiple first road segments are obtained. A random sampling consensus algorithm is used to perform local optimal matching between the point cloud data in the first road segments and the standard point cloud data of the infrastructure to obtain target point cloud data. Based on the reproducibility of the target point cloud data at different road locations, the target point cloud data is filtered for denoising. Since the infrastructure and roads are embedded, this method allows for filtering of point cloud data within embedded structures, thus denoising the target point cloud data. Finally, a deep learning algorithm is used to perform semantic segmentation and instance extraction on the denoised target point cloud data to reconstruct a target road model containing different infrastructures. This results in road facilities in the segmented target road model that better reflect actual road conditions.

[0093] Furthermore, referring to Figure 2 Based on the first embodiment of this application, another embodiment of this application is provided. In this embodiment, the step S20 of segmenting the first point cloud data to obtain multiple first road segments includes: Step S21: Along the road direction corresponding to the road area, the vertical plane perpendicular to the road direction is taken as the road section of each road area.

[0094] As an example, along the road direction corresponding to the road area, draw a vertical plane corresponding to the road centerline, and use this vertical plane as the road section of the corresponding road area.

[0095] Step S22: Determine the number of point clouds corresponding to the first point cloud data at the data collection location within each road area, and use the number of point clouds as the cross-sectional density of each road section.

[0096] As an example, the method for obtaining cross-sectional density can be: The location of the vehicle on the road is obtained, and the number of point clouds in the first point cloud data in the area is calculated with the location as the center and the road as the width, which is used as the cross-sectional density of the road section corresponding to each road location.

[0097] Step S23: Cluster the cross-sectional density to obtain multiple first road segments.

[0098] As an example, a clustering space is constructed with road location as the horizontal axis and cross-sectional density as the vertical axis. The cross-sectional density is then clustered using a hierarchical clustering algorithm to obtain multiple first road segments.

[0099] Step S23 includes: Calculate the mean density of the cross-sectional density for all road areas, and use the mean density as the clustering threshold.

[0100] If the difference in the mean cross-sectional density between any two clusters is greater than the clustering threshold, then these two clusters are determined to be divided into two first road segments, until all road areas are clustered, resulting in multiple first road segments.

[0101] As an example, during the clustering process, since a threshold is needed as the end limit of clustering, the mean of the cross-sectional density of all road locations in each road area is calculated, and 5% of the mean is used as the clustering threshold. When the mean cross-sectional density of two clusters differs by more than 5% during the clustering process, it is considered that two first road segments have been obtained. After all road areas are clustered, multiple first road segments are obtained.

[0102] In this embodiment, the direction along the road is the main direction of the point cloud data. The point cloud data contained in different road segments are different. By dividing each road, multiple first road segments are obtained. The point cloud data of the first road segments are used as the basis to facilitate subsequent road division and data denoising.

[0103] Reference Figure 3 , Figure 3 This is a schematic diagram of the device structure of the hardware operating environment involved in the embodiments of this application.

[0104] like Figure 3 As shown, the road infrastructure BIM model reverse engineering device based on data fusion may include: a processor 1001, a memory 1003, and a communication bus 1002. The communication bus 1002 is used to realize the connection and communication between the processor 1001 and the memory 1003.

[0105] Optionally, the data fusion-based road infrastructure BIM model reverse engineering device may also include a user interface, network interface, camera, RF (Radio Frequency) circuit, sensors, WiFi module, etc. The user interface may include a display screen and an input submodule such as a keyboard; optional user interfaces may also include standard wired and wireless interfaces. The network interface may include standard wired and wireless interfaces (such as a Wi-Fi interface).

[0106] Those skilled in the art will understand that Figure 3 The structure of the road infrastructure BIM model reverse engineering equipment shown in the figure does not constitute a limitation on the road infrastructure BIM model reverse engineering equipment based on data fusion. It may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0107] like Figure 3 As shown, the memory 1003, serving as a storage medium, may include an operating system, a network communication module, and a road infrastructure BIM model reverse engineering program based on data fusion. The operating system is a program that manages and controls the hardware and software resources of the road infrastructure BIM model reverse engineering device based on data fusion, supporting the operation of the road infrastructure BIM model reverse engineering program based on data fusion and other software and / or programs. The network communication module is used to enable communication between the various components within the memory 1003, as well as communication with other hardware and software in the road infrastructure BIM model reverse engineering system based on data fusion.

[0108] exist Figure 3 In the data fusion-based road infrastructure BIM model reverse construction device shown, the processor 1001 is used to execute the data fusion-based road infrastructure BIM model reverse construction program stored in the memory 1003 to implement the steps of the data fusion-based road infrastructure BIM model reverse construction method described above.

[0109] The specific implementation method of the road infrastructure BIM model reverse construction device based on data fusion in this application is basically the same as the various embodiments of the road infrastructure BIM model reverse construction method based on data fusion described above, and will not be repeated here.

[0110] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0111] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0112] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods of the various embodiments of this application.

[0113] The above are merely preferred embodiments of this application and do not limit the scope of this application. Any equivalent structural or procedural transformations made based on the description and drawings of this application, or direct or indirect applications in other related technical fields, are similarly included within the scope of protection of this application.

[0114] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0115] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for reverse engineering a BIM model of road infrastructure based on data fusion, characterized in that, The method includes: Acquire first-point cloud data of multiple road areas and their infrastructure; The first point cloud data is segmented to obtain multiple first road segments; Based on the random sampling consensus algorithm, the point cloud data in the first road segment is locally optimally matched with the standard point cloud data of the infrastructure to obtain the target point cloud data. Based on the degree of reproducibility of each target point cloud data at different road locations, the target point cloud data is filtered to perform noise reduction processing on the target point cloud data; By using deep learning algorithms to perform semantic segmentation and instance extraction on the denoised target point cloud data, a target road model containing different infrastructure is constructed.

2. The method for reverse construction of road infrastructure BIM model based on data fusion as described in claim 1, characterized in that, The segmentation of the first point cloud data yields multiple first road segments, including: Along the road direction corresponding to the road area, the vertical plane perpendicular to the road direction is taken as the road section of each road area; Determine the number of point clouds corresponding to the first point cloud data at each data collection location within each road area, and use the number of point clouds as the cross-sectional density of each road section; Clustering is performed on the cross-sectional density to obtain multiple first road segments.

3. The method for reverse construction of road infrastructure BIM model based on data fusion as described in claim 2, characterized in that, The clustering process of the cross-sectional density yields multiple first road segments, including: Calculate the mean density of the cross-sectional density corresponding to all road areas, and use the mean density as the clustering threshold; If the difference in the mean cross-sectional density between any two clusters is greater than the clustering threshold, then these two clusters are determined to be divided into two first road segments, until all road areas are clustered, resulting in multiple first road segments.

4. The method for reverse construction of road infrastructure BIM model based on data fusion as described in claim 1, characterized in that, The method based on the random sampling consensus algorithm performs a local optimal match between the point cloud data in the first road segment and the standard point cloud data of the infrastructure to obtain the target point cloud data, including: A predetermined number of second point cloud data are extracted from each of the first point cloud data using a random sampling consensus algorithm; Set the distance between the second point cloud data and the nearest point cloud data in the standard point cloud data corresponding to all infrastructures as the shortest matching distance of the second point cloud data. Sort the shortest matching distances corresponding to all second point cloud data in descending order to obtain a point cloud distance sorting sequence, and select the optimal matching sequence segment in the point cloud distance sorting sequence; Based on the point cloud data corresponding to the optimal matching sequence segment and the shortest matching distance, the reliability of the point cloud data at the current iteration number is calculated. When the reliability level is greater than a preset threshold, the point cloud data of the current iteration number is used as the target point cloud data.

5. The method for reverse construction of road infrastructure BIM model based on data fusion as described in claim 4, characterized in that, The step of selecting the optimal matching sequence segment from the point cloud distance sorting sequence includes: Calculate the distance difference between the two shortest matching distances adjacent to each other in the point cloud distance sorting sequence; Based on the position corresponding to the maximum value of the distance difference, the point cloud distance sorting sequence is divided to obtain a first sequence segment and a second sequence segment. The segment corresponding to the minimum average distance between the first and second sequence segments is taken as the optimal matching sequence segment.

6. The method for reverse construction of road infrastructure BIM model based on data fusion as described in claim 4, characterized in that, The calculation of the reliability of the point cloud data for the current iteration number based on the point cloud data corresponding to the optimal matching sequence segment and the shortest matching distance includes: Determine the first proportion of point cloud data corresponding to the optimal matching sequence segment, and the first average value of each of the shortest matching distances; The reliability of the point cloud data at the current iteration number is calculated based on the ratio between the first proportion and the first average value.

7. The method for reverse construction of road infrastructure BIM model based on data fusion as described in claim 4, characterized in that, The filtering of the target point cloud data based on the reproducibility of each target point cloud data at different road locations includes: Along the road direction corresponding to the road region, construct slice planes at multiple road locations in each road region; Based on the distance between target point cloud data in different slice planes, the suspected reproducing point cloud data of each target point cloud data is determined; Based on the number of suspected reproducible point cloud data and the number of slice planes, the degree of reproducibility of each target point cloud data at different road locations is calculated. Determine the spatial distance between each of the target point cloud data and the fitted model constructed by the random sampling consensus algorithm in the current iteration number; Based on the degree of reproduction, the spatial distance is weakened to obtain the weakened distance; Based on the reduced distance, the target point cloud data is divided into inner points and outer points, and the inner points are retained.

8. The method for reverse construction of road infrastructure BIM model based on data fusion as described in claim 7, characterized in that, The method of determining the suspected reproducing point cloud data for each target point cloud data based on the distance between target point cloud data in different slice planes includes: For any slice plane, project the target point cloud data between the current slice plane and the next adjacent slice plane onto the current slice plane, until point cloud data exists on all slice planes; Project the point cloud data of each slice plane in each road region onto the first slice plane with the road center as the origin; For any target point cloud data in the first slice plane, the point cloud data in the preset neighboring region of the current target point cloud data is taken as the suspected reproduction point cloud data of the current target point cloud data.

9. The method for reverse construction of road infrastructure BIM model based on data fusion as described in claim 7, characterized in that, The calculation of the reproduction degree of each target point cloud data at different road locations based on the number of suspected reproduced point cloud data and the number of slice planes includes: All suspected reproduction point cloud data of different target point cloud data are arranged in the order of road direction, and the number of planar intervals between any two adjacent suspected reproduction point cloud data of each target point cloud data is determined. Calculate the average distance between each target point cloud data and its corresponding suspected reproducing point cloud data; Based on the ratio between the mean distance and the number of planar intervals, and the ratio between the number of suspected reproduced point cloud data and the number of sliced ​​planes, the reproduction degree of each target point cloud data at different road locations is calculated.

10. The method for reverse construction of road infrastructure BIM model based on data fusion as described in claim 7, characterized in that, The step of dividing the target point cloud data into interior points and exterior points based on the weakened distance, and retaining the interior points, includes: The weakened distance is compared with a preset distance threshold. If the weakened distance is less than a preset distance threshold, the target point cloud data corresponding to the weakened distance is divided into inner points; otherwise, it is divided into outer points. Determine the iteration number with the largest number of interior points among multiple iterations of the random sampling consensus algorithm, retain the interior points at that iteration number, and filter out all exterior points at that iteration number.

Images