Road asset automatic identification and modeling system based on laser point cloud

Abstract

The invention discloses a road asset automatic identification and modeling system based on laser point cloud, which belongs to the technical field of road modeling and comprises an infrastructure layer, a data storage layer, an application support layer, a business application layer, an interaction layer and a user layer. According to the invention, the point cloud processing efficiency and the road asset classification and identification precision are improved by improving and fusing the point cloud processing technology, the classification and identification technology and the automatic hooking technology; adopting a grid thinning algorithm to realize point cloud data thinning; a bilateral filtering algorithm and a Gaussian filtering algorithm are adopted to realize a point cloud denoising function, so that the point cloud processing efficiency is improved; various algorithms are fused, point cloud processing and automatic modeling technologies are improved, and the accuracy of automatic classification modeling is improved to 90% or above; automatic modeling is introduced, point cloud processing, classification identification and automatic modeling technologies are fused, the process is automatically linked, and the construction period is shortened by more than half.

Description

Technical Field

[0001] The present invention relates to the technical field of road modeling, and in particular to a highway asset automatic identification and modeling system based on laser point cloud. Background Art

[0002] At present, the method of individualizing data modeling of road facilities mainly adopts manual methods to individualize road assets and enter attributes. The most advanced point cloud processing software at home and abroad can only achieve point cloud cropping, point cloud thinning, point cloud denoising, point cloud classification, etc., and cannot automatically establish individual three-dimensional models. The cost of manual processing of urban road models is relatively high. If large-scale modeling and individualization of roads are carried out in this way, the existing road property archive information cannot be fully utilized, resulting in a waste of existing resources, high economic costs and low efficiency. Summary of the invention

[0003] In view of the deficiencies in the prior art, the present invention provides a highway asset automatic identification and modeling system based on laser point cloud.

[0004] The technical solution of the present invention to solve the above technical problems is as follows: A highway asset automatic identification and modeling system based on laser point cloud, comprising an infrastructure layer, a data storage layer, an application support layer, a business application layer, an interaction layer and a user layer; The infrastructure layer is the operating environment of the system; The data storage layer includes a spatial database and a business database, which realizes unified and standardized management of spatial data and business data, and provides data support for the business application layer; The application support layer is used to provide basic functional support, including unified authentication, integrated middleware, report components, presentation layer components, workflow engine, data access components, and supports the display and application of the business application layer; The business application layer includes a management module, a data processing module, a data quality inspection module, a road-property matching module, a three-dimensional modeling module and a general function module, wherein: The management module is used to realize the management, query, extraction, statistical analysis and batch production of various point cloud data products; The data processing module includes a data set processing module and a vector editing module; The data quality inspection module includes a data quality inspection tool to implement the inspection of the road property extraction results; The road-asset matching module is used to realize the information connection function between the road asset archive data and the road asset model data, connect various attribute information in the road asset archive to the point cloud model data, and form three-dimensional data information with searchable elements, editable attributes and visualized models, including batch connection submodule, manual connection submodule and classification display submodule functions; The three-dimensional modeling module is used to perform three-dimensional modeling of road assets according to the vector extraction results; The interaction layer is used for point cloud task allocation, point cloud data processing, three-dimensional model establishment, progress display, and data product production, generation and export.

[0005] Furthermore, the management module includes a data management module, a scene management module, a task management module and a layer management module; The data management module includes functions of uploading point cloud data, uploading vector data, uploading street view data, uploading trajectory data, opening data sets, viewing data set details, publishing data sets, downloading data sets, deleting data sets, and data set screening and searching, and realizes the uploading and downloading of point cloud data in las and laz formats, vector data in geojson and shapefile formats, street view data in shapefile and jpg combination formats, and trajectory data in pos and shapefile formats, as well as the publishing and downloading of data; The scene management module includes the functions of creating scenes, saving scenes, sharing scenes, opening\deleting\searching\viewing\collecting\cancelling scenes, realizing the creation, collection, sharing of scenes and setting of scene permissions; The task management module implements point cloud slicing, creation, deletion, cancellation of processing tasks and viewing of task status; The layer management module includes functions for loading point cloud, vector, street view, oblique photography data sets, showing / hiding layers, and setting point cloud style, realizing visualization of point cloud, vector, street view and oblique photography data, and setting the display mode of point cloud data and style of vector data.

[0006] Furthermore, the batch attachment submodule is used to batch attach the attributes of the asset archive data and the model data in the system, based on the fact that the three-dimensional model and the two-dimensional attribute data have the same coordinate system, using the spatial position relationship between the corresponding layer points and surfaces, combined with spatial data query, matching the model coordinate information through the coordinate information of the road asset archive data, and completing the three-dimensional model batch attachment road asset archive attribute function; The manual mounting submodule is used to process the data of the batch mounting submodule that is abnormally mounted due to the offset between the coordinate position in the individual road property file information and the corresponding three-dimensional model coordinates, and provide a mounting abnormality information list. The user can manually mount the abnormal data and record the relevant abnormal description; The classification display submodule is used to display the three-dimensional information of the three-dimensional modeling data, and uses the spatial data attribute table to display the three-dimensional modeling data in different colors.

[0007] Furthermore, the three-dimensional modeling module includes a model library management tool, a vector matching model library tool, and a model editing tool; The model library management tool is used to implement operations of importing established models, viewing models, and deleting existing models; The vector matching model library is used to import the vector files into the required three-dimensional model layers in layers, and the vector files are in .shp format; The model editing tool is used to manually edit and modify the size, position, angle, and direction information of the model as required.

[0008] Furthermore, the data quality inspection module includes algorithm quality inspection and classification result quality inspection.

[0009] Furthermore, the spatial database includes road asset data, road asset vector data, and model data; the business database includes task distribution data, task progress data, system operation logs, and account information data.

[0010] Furthermore, the data set processing module includes a point cloud cropping tool, a point cloud thinning tool, a point cloud denoising tool, a road point cloud adaptive threshold segmentation algorithm, a marking semantic extraction tool, a rod extraction tool, a sign recognition tool, an arrow matching tool, a guardrail fitting tool, a zebra crossing construction tool, and a curb extraction tool.

[0011] Furthermore, the point cloud cutting tool is used to perform segmentation processing on the point cloud data.

[0012] Furthermore, the vector editing module realizes the editing of unrecognized or erroneous parts of road assets identified by algorithms, and can also draw points, broken lines, vertical lines, arcs, polygons, circles, rectangles, edit inflection points, interrupt / move / copy elements, undo, retract, rotate elements, vertical sections, and capture elements.

[0013] In summary, compared with the prior art, the above technical solution has the following beneficial effects: (1) By improving and integrating point cloud processing technology, classification and recognition technology, and automatic attachment technology, the efficiency of point cloud processing and the accuracy of highway asset classification and recognition are improved; the grid thinning algorithm is used to achieve point cloud data thinning; the bilateral filtering algorithm and Gaussian filtering algorithm are used to achieve point cloud denoising, thereby improving the efficiency of point cloud processing; (2) Improve and integrate various algorithms such as point cloud processing and automatic modeling technology to increase the accuracy of automatic classification modeling to more than 90%; (3) Introducing automatic modeling, integrating point cloud processing, classification and recognition, and automatic modeling technologies, and automatically connecting processes, shortening the construction period by more than half. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 is an overall structural block diagram of an embodiment of the present invention; Figure 2 is a schematic diagram of a secondary index architecture in an embodiment of the present invention; Figure 3 A schematic diagram of a model list established by using third-party software in an embodiment of the present invention; Figure 4 is a schematic diagram of the segmentation result after using the road surface point cloud adaptive threshold segmentation algorithm in an embodiment of the present invention; Figure 5 is a schematic diagram of a road surface strength characteristic image in an embodiment of the present invention; Figure 6 is a schematic diagram of the manual marking result of the marking line in an embodiment of the present invention; Figure 7 A schematic diagram of the extraction and recognition results of marking lines in an embodiment of the present invention; Figure 8 is a schematic diagram of a ground intensity grid in an embodiment of the present invention; Fig. 9 is a flow chart of a zebra crossing modeling algorithm in an embodiment of the present invention; Fig.10 A schematic diagram of calculating zebra crossing corner points in an embodiment of the present invention; Fig.11 is a schematic diagram of an original zebra crossing point cloud in an embodiment of the present invention; Fig.12 Schematic diagram of OTSU binary segmentation in an embodiment of the present invention; Fig.13 Schematic diagram of long edge extraction of a zebra crossing in an embodiment of the present invention; Fig.14 Schematic diagram of scanning line extraction at the short edge of a zebra crossing in an embodiment of the present invention; Fig.15 A schematic diagram of zebra crossing corner point extraction in an embodiment of the present invention; Fig.16 It is a schematic diagram of the process of extracting the road bump point cloud in an embodiment of the present invention; Fig.17 is a schematic diagram of the spatial distribution of the absolute elevation and relative elevation of the point cloud in an embodiment of the present invention; Fig.18 Schematic diagram of the distribution of normal vector directions of street point clouds in an embodiment of the present invention; Fig.19 Schematic diagram of the refinement result of the normal vector direction of the intersection point cloud in an embodiment of the present invention; Fig. 20 Schematic diagram of the road bump point cloud extraction result in an embodiment of the present invention. DETAILED DESCRIPTION

[0015] The principles and features of the present invention are described below in conjunction with all the accompanying drawings. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.

[0016] The embodiment of the present invention discloses a highway asset automatic identification and modeling system based on laser point cloud.

[0017] Reference Figure 1-Figure 20 , a highway asset automatic identification and modeling system based on laser point cloud, includes infrastructure layer, data storage layer, application support layer, business application layer, interaction layer and user layer.

[0018] The infrastructure layer is the operating environment of the system.

[0019] The data storage layer includes a spatial database and a business database to achieve unified and standardized management of spatial data and business data, and provide data support for the business application layer; the spatial database includes road asset data, road asset vector data, and model data; the business database includes task distribution data, task progress data, system operation logs, and account information data.

[0020] The application support layer is used to provide basic functional support, including unified authentication, integrated middleware, report components, presentation layer components, workflow engine, data access components, and supports the display and application of the business application layer.

[0021] The business application layer includes a management module, a data processing module, a data quality inspection module, a road-property matching module, a three-dimensional modeling module and a general function module.

[0022] The interaction layer is used for point cloud task allocation, point cloud data processing, three-dimensional model establishment, progress display, and data product production, generation and export.

[0023] The user layer is the main object of the application system service, including three types of users: center leaders, business personnel and operation and maintenance personnel. Different user types are given different user permissions according to business work, security level, etc.

[0024] The management module is used to realize the management, query, extraction, statistical analysis and batch production of various point cloud data products. The management module includes data management module, scene management module, task management module and layer management module; The data management module includes functions of uploading point cloud data, uploading vector data, uploading street view data, uploading trajectory data, opening data sets, viewing data set details, publishing data sets, downloading data sets, deleting data sets, and data set filtering and searching, realizing the uploading and downloading of point cloud data in las and laz formats, vector data in geojson and shapefile formats, street view data in shapefile and jpg combination formats, and trajectory data in pos and shapefile formats, as well as the publishing and downloading of data.

[0025] The scene management module includes the functions of creating scenes, saving scenes, sharing scenes, opening\deleting\searching\viewing\collecting\cancelling scenes, and realizing the creation, collection, sharing of scenes and the setting of scene permissions.

[0026] The task management module implements point cloud slicing, creation, deletion, cancellation of processing tasks and viewing of task status.

[0027] The layer management module includes functions for loading point cloud, vector, street view, oblique photography data sets, showing / hiding layers, and setting point cloud style, realizing visualization of point cloud, vector, street view and oblique photography data, and setting the display mode of point cloud data and style of vector data.

[0028] The batch attachment submodule is used to batch attach the attributes of the asset archive data and model data in the system. Based on the fact that the three-dimensional model and the two-dimensional attribute data have the same coordinate system, the spatial position relationship conditions between the corresponding layer points and surfaces are utilized, combined with spatial data query, and the coordinate information of the road asset archive data is matched with the model coordinate information to complete the three-dimensional model batch attachment road asset archive attribute function.

[0029] The batch attachment submodule includes an attachment query unit, a manual editing unit, an attachment result classification display unit, an attachment progress unit, and an attribute matching unit. Among them: The attachment query unit sets an attribute table for the three-dimensional data, and writes the batch of automatically attached road property archive data into the attribute table. Set a query unit to query the information in the attribute table. If the road property archive data table is empty, manual attachment is required, and the attribute table can be clicked to determine the location of the data that is not automatically attached and visualize it.

[0030] The manual editing unit is used to set the point selection or box selection module, which can select part of the three-dimensional model and write the road property file data into the attribute table.

[0031] The attachment result classification display unit is used to display the execution results after the batch attachment operation is completed, including attachment success, attachment exception and description of the exception, and displays the corresponding information of the archive data and model data of the successful attachment, and the corresponding information of the archive data and model data of the attachment exception in a list format to ensure the completeness, accuracy and consistency of the data attachment.

[0032] The mounting progress unit is used to check whether the mounting is successful. If there is no null value in the 3D model attribute table, it means that the mounting is successful, the mounting success status box is displayed, and the next step can be performed.

[0033] After the automatic mounting is completed, it will show whether all data are mounted successfully. If the data is not successfully mounted, the log of mounting exception will be output. The log will describe the data source where the mounting failed, the reason for the failure and other information.

[0034] After the display is successfully mounted, the attribute table is output, with the 3D modeling data on the left and the road property file data on the right. The attribute table is editable, which is convenient for fine-tuning some data.

[0035] The attribute matching unit includes spatial index and spatial query processing algorithm.

[0036] Introduce a secondary index on value in the key-value database (Key-Value database, referred to as KV database). The architecture of the secondary index is as follows Figure 2 As shown in Figure 1, data in the KV database is organized and stored in the form of Key-Value correspondence. The query of vector spatial data can be divided into the query of single spatial object and the query of multiple spatial objects.

[0037] The manual mounting submodule is used to process data caused by mounting abnormalities in the batch mounting submodule due to the offset between the coordinate position in the individual road property file information and the corresponding three-dimensional model coordinates, and provide a mounting abnormality information list. Users can manually mount the abnormal data and record relevant abnormal descriptions.

[0038] By setting up a reclassification module, users can classify several data they need into one category and set a new attribute to display the modeling data in different colors on the visualization interface according to their needs.

[0039] The color and type of the classification can be changed according to user needs, so that the visual interface can show different artistic colors.

[0040] The classification display submodule is used to display the three-dimensional information of the three-dimensional modeling data, and uses the spatial data attribute table to display the three-dimensional modeling data in different colors.

[0041] The three-dimensional modeling module includes a model library management tool, a vector matching model library tool, and a model editing tool. The model library management tool is used to implement the operations of importing completed models, viewing models, and deleting existing models. According to the needs, the model list that needs to be established using third-party software is as follows: Figure 3 As shown: The vector matching model library is used to import vector files into the required three-dimensional model layer in layers. The vector files are in .shp format. The system will automatically place the required model in the correct position according to the model space position coordinates, angle information, model library file code and other information in the vector layer.

[0042] The model editing tool is used to manually edit and modify the size, position, angle, and direction information of the model as required.

[0043] The data processing module includes a data set processing module and a vector editing module.

[0044] The data set processing module includes a point cloud cropping tool, a point cloud thinning tool, a point cloud denoising tool, a road point cloud adaptive threshold segmentation algorithm, a marking semantic extraction tool, a rod extraction tool, a sign recognition tool, an arrow matching tool, a guardrail fitting tool, a zebra crossing construction tool, and a curb extraction tool.

[0045] The point cloud cutting tool is used to perform segmentation processing on point cloud data.

[0046] Point cloud thinning is to screen the original point cloud data in a regular manner on the basis of ensuring that production accuracy or actual needs are met, so as to minimize the number of point clouds. According to the rules adopted by different point cloud thinning algorithms, they can be summarized as random sampling algorithms and algorithms that take terrain features into account. Random sampling algorithms mainly include regular grid-based thinning algorithms and system-based thinning algorithms.

[0047] The thinning algorithm based on regular grids divides the entire data range into many small grids, and randomly selects a data point or the average value of several data points in the grid to replace all the data in the grid. The system-based thinning algorithm sets the sampling interval N when loading the original point cloud, takes the first of the N point cloud data or randomly extracts a data point from these N point cloud data. Both algorithms are simple and fast, but they cannot preserve the terrain features well, reducing the accuracy of the point cloud data. Theoretically, the purpose of thinning the point cloud is to reduce the amount of data as much as possible without losing the accuracy of the original data, saving more time for data post-processing. Its main purpose is to accurately describe the characteristics of the ground and objects with fewer points, and to achieve a balance between point cloud density and data accuracy.

[0048] In the embodiment of the present invention, a grid thinning algorithm is used to thin the point cloud data. First, the original file is read, a distance threshold is set, and the first point is used as the base point to calculate the distance from other points to the base point. All points are traversed, and if the distance is less than the threshold, the point is discarded, and if the distance is greater than the threshold, the point is retained. Then, the second point is selected from the retained points as the base point, and the distance from all subsequent points to the base point is calculated. The above conditions are repeated to determine whether to retain, until all points are processed in a loop. In this way, the less point data is calculated later, the faster the calculation speed is.

[0049] Affected by the characteristics of the instrument, the surrounding environment, and the scanned target itself, some noise is unavoidable in the point cloud data. There are many sources of noise, such as points that exceed the set scanning range; points caused by the surrounding wind, vibration of surrounding objects, etc.; or the influence of water vapor in the air, etc. The noise points generated will not only increase the amount of point cloud data, but also affect the accuracy of modeling and information extraction, etc., and need to be removed. The embodiment of the present invention adopts bilateral filtering algorithm and Gaussian filtering algorithm to realize the point cloud denoising function.

[0050] The adaptive threshold segmentation algorithm for road point cloud is realized by comparing the gray value of a pixel with the average gray value of the pixels in the surrounding rectangular area. The average gray value reflects the overall gray value of the background. If the gray value of a pixel is greater than the average gray value of its surrounding pixels by a certain amount, it can be considered as a foreground pixel, otherwise it is considered as a background pixel. Let the pixel coordinates be Pi( , ), the corresponding gray value is I( , ), the integral image is intI( , ), define two parameters s and t to represent the background radius and segmentation coefficient respectively. s is the background radius, and its value should be larger than the target radius to be segmented. The larger s is, the closer it is to single threshold image segmentation. The segmentation coefficient is generally around 1. The sum of the grayscale values ​​of the pixels in the rectangular area with Pi as the center and 2s+1 as the side length is It can be expressed as: ; The pixel value of the binary image after segmentation is expressed by the following formula. The point cloud corresponding to the pixel with a pixel value of 1 is the marking point cloud to be extracted:

[0051] The parameters s=12 and t=1.25 are used to perform adaptive threshold segmentation on the generated road strength grid map. The segmentation results are shown in the figure below. Figure 4 shown.

[0052] The extracted marking point cloud after the above processing is still a bunch of discrete points, without marking object information. The embodiment of the present invention uses the Euclidean clustering method to first cluster the marking point cloud to form a point cloud cluster, and then obtains the minimum bounding rectangle of the point cloud cluster. According to the geometric features of its minimum bounding rectangle, the feature attribute screening method is used to perform semantic recognition on the marking point cloud. Assume that the length and width of the minimum bounding rectangle are a and b respectively, and the aspect ratio is , the point cloud area is , the rectangularity (the ratio of the point cloud area to the area of ​​its minimum circumscribed rectangle) is .

[0053] The marking line semantic extraction tool includes a dotted line recognition tool and a solid line recognition tool.

[0054] The same type of dashed line is composed of multiple short solid line segments, each of which is a rectangle with the same size, so identifying a type of dashed line is to identify the corresponding rectangular point cloud cluster. Since each road marking has a fixed specification, let the length and width of the standard line segment corresponding to the dashed line be , , then when the characteristic attribute of the point cloud settlement satisfies the following formula, it is considered to be a dotted line:

[0055] In the above formula , , , are the four limiting coefficients, , It reflects the allowable difference between the length and width of the point cloud settlement and the design size; Indicates the limit on the minimum area of ​​the marking point cloud; It indicates the restriction on the rectangularity of the settlement. The specific value of the coefficient is closely related to the density of the settlement point cloud and the completeness of the marking point cloud. Generally, the coefficient values ​​are 0.1, 0.2, 0.1, and 0.6.

[0056] The solid line recognition method is similar to the dotted line recognition method. Suppose the standard width of the solid line is , when the characteristic attribute of the point cloud settlement satisfies the following formula, it is considered to be a solid line:

[0057] In the above formula, It is a restriction on the aspect ratio of the minimum circumscribed rectangle of the point cloud cluster. The aspect ratio of the linear marking should be set larger. =7; , The coefficient value setting is the same as that of dotted line recognition. The semantic recognition results of dotted and solid lines are shown in Figure 5-Figure 7 .

[0058] The pole extraction tool uses an algorithm based on PCL's Conditional Euclidean Clustering and RANSAC (Random Sample Consensus) to fit straight lines to extract poles from point clouds. The algorithm based on PCL's Conditional Euclidean Clustering to extract poles is an existing technology and will not be elaborated on here. The sign recognition tool is mainly based on shape-based detection. The reflection intensity-based method will miss traffic signs with low reflectivity due to factors such as aging or damage, as well as traffic signs with the back facing the scanner. Therefore, a shape-based method is combined to ensure the integrity of the detection. First, the voxel-based upward growth method is used to filter out the ground points from the original three-dimensional scene; secondly, the Euclidean distance clustering method is used to cluster the non-ground points to separate the spatially independent target clusters, and the rod-shaped targets are identified from these clusters; finally, the PCA method (Principal Component Analysis) is used to analyze the dimensional characteristics of the non-rod-shaped part points of the rod-shaped target, so as to detect the traffic sign target.

[0059] Common pole-shaped objects on both sides of the road can be divided into several categories: trees, electric poles, street lights, and traffic signs. The composition of the non-pole-shaped parts of these types of objects is shown in Table 1: Table 1 Composition of non-rod-shaped parts of the target

[0060] Among the composition categories, only traffic signs contain a large number of plane elements. Therefore, the PCA algorithm can be used to distinguish traffic signs from other non-research targets by analyzing the dimensionality characteristics.

[0061] The arrow matching tool uses a guide arrow recognition algorithm based on binary coding SVM, and realizes the classification of multiple guide arrows by customizing the binary coding of the results. The algorithm does not need to perform inverse perspective transformation on the collected images, and only needs one SVM classifier to realize the recognition of five guide arrows. After testing, the recognition accuracy of the algorithm is better than 96.8%, which effectively improves the efficiency of guide arrow recognition. The five types of guide arrows are: straight, left turn, right turn, straight left turn and straight right turn.

[0062] The guardrail extraction tool extracts guardrails based on the ground raster map, converting the classified point cloud into a Geotiff image, that is, ground rasterization. At this time, the guardrail is projected as a straight line. The straight line extracted from Geotiff is the identified guardrail.

[0063] Guardrail modeling is divided into road surface point cloud rasterization, road surface point cloud adaptive threshold segmentation, and guardrail edge extraction from point cloud.

[0064] Road point cloud rasterization is to convert the three-dimensional space point cloud into a grayscale raster image according to its attribute information, and use the inverse distance weighted (IDW) method to obtain the grayscale value of each pixel in the raster image. Assuming that the road point cloud is , The reflection intensity value of the laser point is generated by using a resolution of Rpixel (the actual size represented by a single pixel) and an interpolation radius of RIDW to generate an intensity grid image, and P is projected onto the plane.

[0065] In order to ensure that there are point clouds in each pixel after interpolation and avoid image gaps, the interpolation radius RIDW should be larger than the interval between points. Generally, the pixel resolution Rpixel can be equal to the point interval, and the interpolation radius is set to 3Rpixel~4Rpixel. The point interval of the actual collected point cloud data is about 0.02 m. Based on this, the sample data b is rasterized with a resolution of 0.05 m and an interpolation radius of 0.2 m to generate an intensity raster image. Figure 8 shown.

[0066] Adaptive threshold segmentation of road point cloud is achieved by comparing the gray value of a pixel with the average gray value of the pixels in the surrounding rectangular area. The average gray value reflects the overall gray value of the background. If the pixel gray value is greater than the average gray value of its surrounding pixels by a certain amount, it can be considered as a foreground pixel, otherwise it is considered as a background pixel.

[0067] In the embodiment of the present invention, the Euclidean clustering method is used to first cluster the marking point cloud to form a point cloud cluster, and the least squares method is used to fit the points after the cluster to extract the guardrail edge line.

[0068] The zebra crossing construction tool includes the zebra crossing corner point fine extraction of point cloud scan lines and the zebra crossing corner point coordinate extraction.

[0069] The algorithm improves the OTSU method and introduces it into the point cloud data processing to calculate the optimal point cloud echo reflection intensity binary segmentation threshold, so as to effectively separate the zebra crossing point cloud from the non-zebra crossing point cloud. At the same time, the point cloud scan line is used as the basic research unit, and the frequency distribution histogram is combined for statistical analysis to eliminate invalid scan lines. Finally, the point cloud located on the long side of the zebra crossing is retained for linear fitting, and the long and short sides of the zebra crossing are successfully located, and the coordinates of the zebra crossing corners are calculated. The flow chart of the zebra crossing modeling algorithm is as follows: Fig. 9 shown.

[0070] like Figure 10-15 As shown, considering that the scanning line is not completely perpendicular to the direction of the zebra crossing, the straight lines L1 and L2 do not completely coincide with the short side of the zebra crossing, and further judgment is required based on the geometric relationship of the following four straight lines. Calculate the two intersection points P1 and P2 of the straight line L1 and the long side straight line SL1 of the zebra crossing, calculate the two intersection points P3 and P4 of the straight line L2 and the long side SL2 of the zebra crossing, and draw straight lines SS1 and SS2 perpendicular to SL1 through P1, P2 and P3, P4 respectively. SS1 and SS2 are both equations of the short side straight line of the zebra crossing, and SL1 and SL2 are both equations of the long side straight line of the zebra crossing. The four intersection points formed by the intersection of the above four straight lines are the four corner points of the zebra crossing.

[0071] like Figure 16-Figure 20 As shown in the figure, the curb extraction tool extracts curb point cloud based on SVM. First, the original vehicle-mounted laser scanning data is thinned and segmented. Then, the spatial distribution characteristics and local geometric characteristics of the point cloud are combined to construct a six-dimensional point cloud feature vector containing relative elevation, normal vector direction, multi-scale elevation difference and multi-scale elevation variance. SVM is used to preliminarily extract curb point cloud. Finally, clustering and denoising are performed on the extracted results to realize the extraction of curb point cloud from vehicle-mounted laser scanning data. The specific steps include the following: S1.1 Data Preprocessing Specifically, using relative elevation as one of the features of road bump point cloud extraction can effectively overcome the influence of terrain undulation in subsequent road bump point cloud extraction, and accurately constrain the extraction results to the height of the road bump. Calculate the height of the point cloud relative to the ground, and filter out the road (ground + curb) according to the set threshold, such as Fig.17 shown.

[0072] S1.2. Constructing point cloud feature vector Specifically, for the extraction of the road bump point cloud, the normal vector direction, multi-scale elevation difference and multi-scale elevation variance in the point cloud neighborhood are constructed as the local geometric features describing the scanned object. The point cloud normal vector direction (Normal Direction) is the angle between the normal vector of the target point cloud and the Z axis of the coordinate system. When the normal vector is facing the positive direction of the Z axis, the angle between the normal vector and the vector (0, 0, 1) is calculated. When the normal vector is facing the negative direction of the Z axis, the angle between the normal vector and the vector (0, 0, -1) is calculated, such as Fig.18 shown.

[0073] S1.3. Extracting road bump point cloud based on SVM Specifically, a few road bump and non-road bump point clouds are manually extracted as training samples, the point cloud feature vector F is used as the input feature of SVM, the most widely used radial basis function (RBF) is selected as the kernel function, and the particle swarm optimization algorithm (PSO) is used to optimize the penalty factor C and the kernel function parameter γ. Then, the preliminary extraction of road bump point clouds in vehicle-mounted laser scanning data is achieved through SVM training and classification.

[0074] S1.4 Clustering denoising Since there are still a lot of scattered noise points in the extraction results, cluster analysis is performed on the extraction results, and the clustering threshold N (number of point clouds) is set. When the number of point clouds in a cluster is less than the threshold N, the point cloud of the cluster is regarded as a noise point and removed. Finally, the road bump point cloud is obtained as follows: Fig.19 and Fig. 20 shown.

[0075] The vector editing module can realize the editing of unrecognized or erroneous parts of road assets identified by algorithms, and can also draw points, broken lines, vertical lines, arcs, polygons, circles, rectangles, edit inflection points, interrupt / move / copy elements, undo, retract, rotate elements, vertical sections, capture elements and other functions.

[0076] The data quality inspection module includes data quality inspection tools to verify the road property extraction results. The data quality inspection module includes algorithm quality inspection and classification result quality inspection. After the data is automatically quality inspected by the algorithm, a comprehensive re-inspection is required to confirm the accuracy and completeness of the post-processing. The quality inspection mainly checks the following issues: a. Extraction error issues, such as incorrect sign content recognition and lane line generation errors; b. Extraction omission problem, which is caused by occlusion, damage, etc.; c. Attribute generation error issues, such as incorrect classification, etc.

[0077] Algorithm quality inspection includes topology inspection and self-setting quality inspection rules. Topology inspection refers to an operation that checks point, line and surface data according to the corresponding topology rules and returns objects that do not meet the rules. It mainly includes no overlap within the surface dataset, no overlap between surface datasets, surface datasets are included by surface datasets, surface datasets are covered by surface datasets, no overlap within line datasets, no overlap between line datasets, and no duplicate points within point datasets.

[0078] Set your own quality inspection rules and conduct data quality inspection based on the actual road conditions. For example, the distance between adjacent roads of the same type must not be less than a certain value.

[0079] Quality inspection of classification results: superimpose the classification results and check them through spatial and attribute analysis.

[0080] The road asset matching module is used to realize the information connection function between road asset archive data and road asset model data, connect various attribute information in the road asset archive to the point cloud model data, and form three-dimensional data information with searchable elements, editable attributes and visualized models, including batch connection sub-module, manual connection sub-module and classification display sub-module functions.

[0081] The three-dimensional modeling module is used to perform three-dimensional modeling of road assets according to the vector extraction results.

[0082] The general function module includes a setting unit, a view tool unit, and a support unit. In the setting unit, users can set according to their personal preferences, including functions such as "mouse setting" and "terrain setting".

[0083] In the view tool unit, a series of operations are performed on the view window to meet the application requirements of different scenarios. The view menu includes functions such as "Create View", "Close Current View", "Close All Views", "Auto Arrange Windows", "Previous View", "Next View", and "View Selection".

[0084] The Support section includes functions such as "User Manual", "Technical Support", "Check for Updates", and "About".

[0085] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A highway asset automatic identification and modeling system based on laser point cloud, characterized in that: It includes infrastructure layer, data storage layer, application support layer, business application layer, interaction layer and user layer; The infrastructure layer is the operating environment of the system; The data storage layer includes a spatial database and a business database, which realizes unified and standardized management of spatial data and business data, and provides data support for the business application layer; The application support layer is used to provide basic functional support, including unified authentication, integrated middleware, report components, presentation layer components, workflow engine, data access components, and supports the display and application of the business application layer; The business application layer includes a management module, a data processing module, a data quality inspection module, a road-property matching module, a three-dimensional modeling module and a general function module, wherein: The management module is used to realize the management, query, extraction, statistical analysis and batch production of various point cloud data products; The data processing module includes a data set processing module and a vector editing module; The data quality inspection module includes a data quality inspection tool; The road-asset matching module is used to realize the information connection function between the road asset archive data and the road asset model data, connect various attribute information in the road asset archive to the point cloud model data, and form three-dimensional data information with searchable elements, editable attributes and visualized models, including batch connection submodule, manual connection submodule and classification display submodule functions; The three-dimensional modeling module is used to perform three-dimensional modeling of road assets according to the vector extraction results; The interaction layer is used for point cloud task allocation, point cloud data processing, three-dimensional model establishment, progress display, and data product production, generation and export.

2. The highway asset automatic identification and modeling system based on laser point cloud according to claim 1 is characterized by: The management module includes a data management module, a scene management module, a task management module and a layer management module; The data management module includes functions of uploading point cloud data, uploading vector data, uploading street view data, uploading trajectory data, opening data sets, viewing data set details, publishing data sets, downloading data sets, deleting data sets, and data set screening and searching, and realizes the uploading and downloading of point cloud data in las and laz formats, vector data in geojson and shapefile formats, street view data in shapefile and jpg combination formats, and trajectory data in pos and shapefile formats, as well as the publishing and downloading of data; The scene management module includes the functions of creating scenes, saving scenes, sharing scenes, opening\deleting\searching\viewing\collecting\cancelling scenes, realizing the creation, collection, sharing of scenes and setting of scene permissions; The task management module implements point cloud slicing, creation, deletion, cancellation of processing tasks and viewing of task status; The layer management module includes functions for loading point cloud, vector, street view, oblique photography data sets, showing / hiding layers, and setting point cloud style, realizing visualization of point cloud, vector, street view and oblique photography data, and setting the display mode of point cloud data and style of vector data.

3. The highway asset automatic identification and modeling system based on laser point cloud according to claim 1 is characterized by: The batch attachment submodule is used to batch attach the attributes of the asset archive data and the model data in the system. Based on the fact that the three-dimensional model and the two-dimensional attribute data have the same coordinate system, the spatial position relationship between the corresponding layer points and surfaces is used, combined with spatial data query, and the coordinate information of the road asset archive data is matched with the model coordinate information to complete the three-dimensional model batch attachment road asset archive attribute function; The manual mounting submodule is used to process the data of the batch mounting submodule that is abnormally mounted due to the offset between the coordinate position in the individual road property file information and the corresponding three-dimensional model coordinates, and provide a mounting abnormality information list. The user can manually mount the abnormal data and record the relevant abnormal description; The classification display submodule is used to display the three-dimensional information of the three-dimensional modeling data, and uses the spatial data attribute table to display the three-dimensional modeling data in different colors.

4. The highway asset automatic identification and modeling system based on laser point cloud according to claim 1, characterized in that: The three-dimensional modeling module includes a model library management tool, a vector matching model library tool and a model editing tool; The model library management tool is used to implement operations of importing established models, viewing models, and deleting existing models; The vector matching model library is used to import the vector files into the required three-dimensional model layers in layers, and the vector files are in .shp format; The model editing tool is used to manually edit and modify the size, position, angle, and direction information of the model as required.

5. The highway asset automatic identification and modeling system based on laser point cloud according to claim 1 is characterized by: The data quality inspection module includes algorithm quality inspection and classification result quality inspection.

6. The highway asset automatic identification and modeling system based on laser point cloud according to claim 1, characterized in that: The spatial database includes road asset data, road asset vector data, and model data; the business database includes task distribution data, task progress data, system operation logs, and account information data.

7. The highway asset automatic identification and modeling system based on laser point cloud according to claim 1, characterized in that: The data set processing module includes a point cloud cropping tool, a point cloud thinning tool, a point cloud denoising tool, a road point cloud adaptive threshold segmentation algorithm, a marking semantic extraction tool, a rod extraction tool, a sign recognition tool, an arrow matching tool, a guardrail fitting tool, a zebra crossing construction tool, and a curb extraction tool.

8. The highway asset automatic identification and modeling system based on laser point cloud according to claim 7 is characterized by: The point cloud cutting tool is used to perform segmentation processing on point cloud data.

9. The highway asset automatic identification and modeling system based on laser point cloud according to claim 1, characterized in that: The vector editing module realizes the editing of unrecognized or erroneous parts of road assets identified by algorithms, and can also draw points, broken lines, vertical lines, arcs, polygons, circles, rectangles, edit inflection points, interrupt / move / copy elements, undo, retract, rotate elements, vertical sections, and capture elements.

Images