Acquisition surveying and mapping method and system for high-precision geographic space information

Abstract

The invention relates to the technical field of geographic information measurement, in particular to a high-precision geographic space information acquisition surveying and mapping method and system, and the method specifically comprises the steps: constructing a fusion point cloud data set through RGB images, thermal infrared images and combined point cloud data at all moments, and analyzing the terrain complexity; shielding omission features of the point cloud data are analyzed through the data set, and blind area effective scores are constructed; based on the difference between the vertical displacement of the echo intensity at each moment and a preset reference truth value, constructing the reliability score of the data at each moment, and combining the effective score of the blind area and the terrain complexity to construct the shielding compensation confidence at each moment; triggering a path planning algorithm based on the shielding compensation confidence, and guiding the unmanned aerial vehicle to avoid obstacles to complementarily collect a low-confidence region; the integrity and the micro-feature capture rate of the data in the shielding area are remarkably improved, the blind area of high-precision surveying and mapping is reduced, the integrity of the scene data is improved, and the engineering availability of the complex scene data is ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of geographic information measurement, in particular to a high-precision geographic spatial information acquisition and mapping method and system. BACKGROUND

[0002] The acquisition and mapping of high-precision geographic spatial information refers to a process of obtaining accurate positions, geometric shapes, attributes and spatio-temporal change information of objects on the earth surface by using advanced technology. With the development of science and technology, the precision and efficiency of high-precision geographic spatial information acquisition have been greatly improved.

[0003] However, in the acquisition and mapping of high-precision geographic spatial information in complex scenes, the unmanned aerial vehicle aerial photography is difficult to obtain the real terrain at the bottom of the building and under the tree crown due to the limitation of the visual angle, and the ground laser scanning cannot penetrate the vegetation or capture the back target due to the visual obstruction; at the same time, the physical bottleneck of the sensor leads to a large number of omissions of micro-features such as road markings, small well covers and pipeline nodes, which directly affects the engineering usability of the data. In the hidden area and detail acquisition, the multi-band laser radar with stronger penetration is used for ultra-high resolution oblique photography. Although the penetration technology can partially overcome the shielding of water bodies or sparse vegetation to obtain the riverbed or shallow surface shape, the physical penetration ability is limited by the medium properties and wavelength, and it is difficult to obtain the real structure in the deep and dense shielding area, and cannot completely replace the field verification. Thus, the existing technology cannot obtain relatively complete geographic spatial information. SUMMARY

[0004] In order to solve the above technical problems, the purpose of the present application is to provide a high-precision geographic spatial information acquisition and mapping method and system, and the technical solution is as follows:

[0005] In a first aspect, the present application provides a high-precision geographic spatial information acquisition and mapping method, which comprises the following steps:

[0006] For urban streets, the RGB image and the thermal infrared image collected at each time are used to give the RGB color value and the thermal radiation intensity value to the point cloud data collected by the laser sensor at each time, and a fusion point cloud data set at each time is constructed;

[0007] The point cloud data at each time is grid divided, the terrain parameters of each grid are constructed based on the three-dimensional distribution characteristics of the point cloud data in each grid, and the terrain complexity at each time is determined; the reliability score of the point cloud data at each time is constructed based on the difference between the vertical displacement amount of the laser sensor echo intensity and the preset reference true value;

[0008] Cluster the fused point cloud dataset to obtain feature clusters, obtain the horizontal projection area of ​​the point cloud data in each feature cluster, and construct the number of effective points per unit area for each feature cluster based on the number of point clouds in the feature cluster; based on each feature cluster at each moment, use the random forest algorithm to obtain the label value of each feature cluster; calculate the effective score of the blind spot at each moment based on the number of effective points per unit area and the label value;

[0009] The prediction accuracy of the blind spot effective score at each moment is calculated by using the blind spot effective scores of all moments before each moment and the terrain complexity; based on the prediction accuracy and reliability scores of all moments, combined with the compensation algorithm, the compensation value at each moment is obtained;

[0010] The occlusion compensation confidence level at each moment is constructed based on the compensation value and the prediction accuracy, and the motion path of the data acquisition device is adjusted based on the occlusion compensation confidence level.

[0011] In one embodiment, the process of obtaining the terrain complexity at each moment is as follows:

[0012] The horizontal plane projected by the point cloud data at each moment is divided into grids of preset sizes, and the standard deviation of all point cloud elevation values ​​in each grid is used as the first evaluation value; the coefficient of variation of the point cloud projection point density of each grid and all its adjacent grids is calculated as the second evaluation value; the weighted sum of the first evaluation value and the second evaluation value of each grid is used as the terrain parameter of each grid, and the mean of the terrain parameters of all grids at each moment is used as the terrain complexity at each moment.

[0013] In one embodiment, the process of obtaining the reliability score of the point cloud data at each time is as follows:

[0014] Calculate the relative error between the vertical displacement of the laser sensor echo intensity at the current moment and the preset reference true value, and record the difference between the natural number 1 and the relative error as the reliability score of the point cloud data at the current moment.

[0015] In one embodiment, the process of obtaining the number of effective points per unit area of ​​each feature cluster is as follows:

[0016] The horizontal plane projection results of all point cloud data points in each feature cluster are used as the input of the convex hull algorithm to obtain the corresponding horizontal plane projection area; the ratio of the number of point cloud data points in each feature cluster to the horizontal plane projection area is calculated as the number of effective points per unit area of ​​each feature cluster.

[0017] In one embodiment, the process of obtaining the label value of each feature cluster is as follows:

[0018] The random forest model is trained by all feature clusters at historical moments, where the labels and label values ​​of each feature cluster include: valid features are 0, and masked features are 1;

[0019] All feature clusters at the current moment are used as the input of the trained random forest, and the output is the label value of each feature cluster at the current moment.

[0020] In one embodiment, the process of obtaining the effective blind spot score at each moment is as follows:

[0021] Calculate the product of the number of effective points per unit area of ​​each feature cluster and the label value; and use the product of all feature clusters at the current moment as the blind spot effective score at the current moment.

[0022] In one embodiment, the process of obtaining the prediction accuracy of the blind spot effective score at each moment is as follows:

[0023] The sequence of terrain complexity at all times before the current moment is used as the terrain complexity sequence at the current moment; the sequence of blind spot effective scores at all times before the current moment is used as the blind spot effective sequence at the current moment;

[0024] The blind spot validity sequence and terrain complexity sequence at the current moment are used as the input of the XGBoost algorithm, and the output is the prediction accuracy of the blind spot validity score at the current moment.

[0025] In one embodiment, the process of obtaining the compensation value at each moment is as follows:

[0026] The sequence of the prediction accuracies at all moments before the current moment is used as the prediction accuracy sequence at the current moment; the sequence of the reliability scores at all moments before the current moment is used as the reliability score sequence at the current moment;

[0027] The prediction accuracy sequence and reliability score sequence at the current moment are used as the input of the weighted linear compensation algorithm, and the output is the compensation value at the current moment.

[0028] In one embodiment, the step of constructing the occlusion compensation confidence level at each moment and adjusting the motion path of the data acquisition device based on the occlusion compensation confidence level is as follows:

[0029] Taking the normalized value of the ratio of the prediction accuracy to the compensation value at the current moment as the occlusion compensation confidence at the current moment;

[0030] If the occlusion compensation confidence at the current moment is lower than the preset trigger threshold, the path planning algorithm is immediately triggered to generate the optimal re-mining path for the UAV and re-scan the area along the path.

[0031] In a second aspect, an embodiment of the present application also provides a high-precision geographic spatial information collection and mapping system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor implements the steps of any one of the above methods when executing the computer program.

[0032] The embodiments of the present application have at least the following beneficial effects:

[0033] This application assigns RGB color values ​​and thermal radiation intensity values ​​to the point cloud data collected by the laser sensor at each moment through the RGB images and thermal infrared images collected at each moment, and constructs a fused point cloud dataset at each moment; to address the problem of missing micro-features in the shielded area due to the sensor's viewing angle and penetration limitations, the terrain complexity at each moment is analyzed through the three-dimensional distribution characteristics of the point cloud data, and the fused point cloud dataset is clustered to obtain each feature cluster. Combined with random forest, the shielding and omission characteristics of the point cloud data are analyzed to construct the effective score of the blind spot at each moment, quantify the number of effective points per unit area in the shielded area and the identifiability, accurately locate the shielded hotspot area with missing data, and avoid the blindness of traditional methods to the omission of micro-features; to address the problem of accuracy fluctuations in shielded areas and differences in sensor performance in complex terrain. To solve the problem of decreased data credibility, the reliability score of the point cloud data at each moment is constructed based on the difference between the vertical displacement of the sensor echo intensity at each moment and the preset benchmark true value. The prediction accuracy of the blind spot effective score at each moment is calculated in combination with the blind spot effective score and the terrain complexity, and the shielding compensation confidence at each moment is constructed to comprehensively reflect the penetration ability of multi-source collaboration in the shielded area and the data restoration credibility, and eliminate the interference of the sensor's own performance fluctuation on the data quality in the shielded area; based on the shielding compensation confidence at the acquisition moment, the path planning algorithm is triggered to guide the UAV to avoid obstacles and supplement the low-confidence area, significantly improving the integrity of the data in the shielded area and the micro-feature capture rate, reducing the blind spots of high-precision mapping, improving the integrity of scene data, and ensuring the engineering availability of complex scene data. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present application or the prior art, the following is a brief introduction to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0035] Figure 1 A flowchart of a method for collecting and mapping high-precision geographic spatial information provided in one embodiment of the present application;

[0036] Figure 2 Schematic diagram of the process of obtaining terrain complexity. DETAILED DESCRIPTION

[0037] To further illustrate the technical means and effectiveness of this application to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effectiveness of a high-precision geospatial information acquisition and mapping method and system proposed in this application. In the following description, references to different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics of one or more embodiments may be combined in any suitable manner.

[0038] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0039] The following describes in detail a method and system for collecting and mapping high-precision geographic spatial information provided by the present application with reference to the accompanying drawings.

[0040] See also Figure 1 , which shows a flowchart of a method for collecting and mapping high-precision geographic spatial information provided by an embodiment of the present application, the method comprising the following steps:

[0041] Step S1: For urban streets, the RGB images and thermal infrared images collected at each moment are used to assign RGB color values ​​and thermal radiation intensity values ​​to the point cloud data collected by the laser sensor at each moment, and a fused point cloud dataset at each moment is constructed.

[0042] A multispectral LiDAR (Light Detection And Ranging) sensor and a five-lens oblique photography system are deployed on an aerial work platform to collect point cloud data of urban streets. Specifically, centimeter-level precision point cloud data is directly obtained through the multispectral LiDAR sensor, and RGB images and thermal infrared images are synchronously obtained through the five-lens oblique photography system. The image frame rate is synchronized with the LiDAR scan; the point cloud obtained by the LiDAR sensor is used as a spatial reference, and the image obtained by the oblique photography system is projected into the point cloud coordinate system through the POS (Position and Orientation System) data and the ICP registration algorithm. Each LiDAR point is assigned an RGB color value and a thermal radiation intensity value to generate a fused point cloud dataset with multispectral attributes. In the embodiment of the present application, the aerial work platform is a drone. A fused point cloud dataset is obtained at each acquisition moment.

[0043] Regarding the data acquisition frequency of the LiDAR sensor and the five-lens oblique photography system, preferably, in the embodiment of the present application, the data acquisition frequency is set to 10 Hz. As other embodiments of the present application, the implementer can set the data acquisition frequency according to actual conditions.

[0044] Step S2: Grid the point cloud data at each moment, construct the terrain parameters of each grid based on the three-dimensional distribution characteristics of the point cloud data within each grid, and determine the terrain complexity at each moment; construct the reliability score of the point cloud data at each moment based on the difference between the vertical displacement of the laser sensor echo intensity at each moment and the preset reference true value.

[0045] (1) Project the point cloud data at each acquisition moment onto a horizontal plane, and divide the projected horizontal plane into 5m×5m grid cells. Each grid cell must contain at least 50 point cloud projection points, otherwise it will be marked as an invalid grid. For each grid, calculate the standard deviation of the elevation values ​​of all point clouds in the grid as the first evaluation value to quantify the intensity of terrain undulation; calculate the coefficient of variation of the point cloud projection point density of the grid and all its adjacent grids as the second evaluation value; perform a weighted summation of the first evaluation value and the second evaluation value of each grid, wherein the entropy weight method is used to automatically determine the weights of the first evaluation value and the second evaluation value, and use the weighted summation result as the terrain parameter of each grid. The mean of the terrain parameters of all grids at each acquisition moment is used as the terrain complexity at each acquisition moment to achieve comprehensive quantification of terrain ruggedness and morphological mutation. Among them, the acquisition process of elevation value, point cloud projection point density, coefficient of variation, as well as point cloud plane projection and entropy weight method are all well-known technologies, and the specific process will not be repeated here.

[0046] (2) Install buried displacement sensors at key underground nodes, such as manhole covers and pipeline interfaces, and generate sensor reliability data in combination with LiDAR echo intensity consistency analysis. Specifically, the vertical displacement of the surface is continuously monitored by the buried displacement sensor and used as the reference true value; at the same time, the vertical displacement of the LiDAR sensor echo intensity in continuous scanning is obtained. This is a well-known technology and the specific process will not be repeated here; finally, calculate the relative error between the vertical displacement of the echo intensity at the current moment and the reference true value of the buried sensor at the current moment, and record the difference between the natural number 1 and the relative error as the reliability score R of the overall data credibility at the current moment.

[0047] (3) For all acquisition moments, extract the terrain complexity data and reliability score data obtained at all historical acquisition moments before each acquisition moment, and sort the same data in ascending order of time to obtain the terrain complexity sequence and sensor reliability score sequence at each acquisition moment.

[0048] Step S3: cluster the fused point cloud data set to obtain each feature cluster, obtain the horizontal plane projection area of ​​the point cloud data in each feature cluster, and construct the effective number of points per unit area of ​​each feature cluster in combination with the number of point clouds in the feature cluster; based on each feature cluster at each moment, combine the random forest algorithm to obtain the label value of each feature cluster; calculate the effective score of the blind spot at each moment based on the effective number of points per unit area and the label value.

[0049] In high-precision geospatial data collection in complex scenes, sensor field of view obstructions and physical penetration limits often lead to the omission of critical micro-features such as road markings, manhole covers, and pipeline nodes, severely reducing the engineering usability of the data. The physical penetration limit, as used in this application, refers to the upper limit of a sensor's ability to penetrate obstacles and obtain information about obscured targets, due to its physical properties.

[0050] (1) For each acquisition moment, the fused point cloud data set obtained at the acquisition moment is used as the input of the DBSCAN clustering algorithm. In order to avoid the problem that the point cloud clusters of micro features cannot be identified due to excessively large values, in the embodiment of the present application, the neighborhood radius is set to 0.3 and the minimum number of points is set to 5. As other embodiments of the present application, the implementer can set the values ​​of the neighborhood radius and the minimum number of points according to actual conditions. The algorithm outputs each cluster, and each cluster is used as a feature cluster to identify independent landforms clustered in sub-meter space. For example, a manhole cover appears as a dense point cluster with a diameter of 0.6m, and a wall corner appears as a linear high-density point group, so as to achieve the effect of deconstructing the micro-topographic structure of the shielded area. The DBSCAN clustering algorithm is a well-known technology, and the specific process will not be repeated.

[0051] It should be noted that for the clustering of fused point cloud datasets, only one clustering method is provided in the embodiments of this application. There are many existing clustering methods, and implementers can also use other clustering algorithms to cluster the fused point cloud datasets. This application does not make specific restrictions.

[0052] (2) For each feature cluster, the projection area of ​​the point cloud within the feature cluster on the horizontal plane is obtained. Specifically, the projection of the point cloud data points within the feature cluster on the horizontal plane is obtained. The projection process is a well-known technique and the specific process will not be described in detail. The projection results of all the point cloud data points within the feature cluster are then used as the input of the convex hull algorithm to obtain the horizontal plane projection area of ​​these point cloud data points. Furthermore, the ratio of the number of point cloud data points within the feature cluster to the projection area of ​​all the point cloud data points within the feature cluster on the horizontal plane is calculated as the effective number of points per unit area of ​​the feature cluster to reflect the laser penetration capability.

[0053] (3) For each feature cluster at each collection moment in the historical collection task, manually set the label value of each feature cluster at each moment, with the label type being valid feature and masked omission, and the corresponding label value being set to 0 and 1 respectively; among all the collection moments in the historical collection task, randomly select the feature clusters of 70% of the collection moments as the training set, and the feature clusters of the remaining 30% of the collection moments as the test set, and perform random forest training to obtain a trained random forest model. The random forest model training process is a well-known technology, and the specific process will not be repeated here.

[0054] The effective features mentioned in this application refer to that the feature cluster is not obscured and the sensor successfully captures the target; the obscured omission refers to that the feature cluster is missing data due to obscuration and the key micro-features are not captured.

[0055] Afterwards, all feature clusters at the current moment are used as input to the trained random forest. In this embodiment of the application, the number of decision trees is set to 100, the splitting criterion is the Gini index, and the label value of each feature cluster at the current moment is output, with the valid feature being 0 and the occlusion omission being 1, thereby quantifying the identifiability of each feature cluster. For example, the manhole cover is marked as 0 due to the high metal reflectivity, and the soil under the tree canopy is marked as 1 due to the missing point cloud, thereby establishing a feature validity criterion for the occluded area.

[0056] (4) Based on the above analysis, the effective score of the blind spot at each acquisition moment is calculated, and the expression is:

[0057]

[0058] Where A is the effective score of the blind spot at the current moment; I is the number of feature clusters at the current moment; D i is the number of effective points per unit area of ​​the i-th feature cluster at the current moment; l i is the label value of the i-th feature cluster at the current moment.

[0059] In DBSCAN clustering, feature clusters are independent features clustered in sub-meter space. In high-precision geospatial collection, D i The value of is a measure of the sensor's penetration capability and data acquisition efficiency. The larger the value, the more successful the laser is in penetrating sparse vegetation or avoiding shielding, and high-density point clouds are obtained in blind areas such as under tree canopies and at the bottom of buildings. i is the binary label of the i-th feature cluster, which takes values ​​of 0 or 1, indicating the identifiability of the feature cluster and the identification of the masking effect and feature reliability. When its value is 1, it corresponds to the masking hotspot in the real world, and when its value is 0, it indicates the area where the sensor information is collected more completely. i As a binary switch control D i Is it included in the calculation, that is, only when its value is 1 D iIt will be brought into the calculation, indicating that only the problem area of ​​the masked missed cluster is concerned, and the effective feature area does not participate in the calculation.

[0060] The blind spot effectiveness score A represents the comprehensive score of data integrity in the obscured area, quantifying the laser sensor's penetration ability and data density in the obscured area, thereby evaluating the impact of obscuration on the integrity of key micro-feature acquisition: the higher the A value, the more dense data can still be obtained under obscuration, the better the scene's mappability, and the smaller the data blind spot; conversely, it reveals serious data missing.

[0061] Step S4, calculating the prediction accuracy of the blind spot effective score at each moment by using the blind spot effective scores and terrain complexity of all moments before each moment; based on the prediction accuracy and reliability scores of all moments, combined with the compensation algorithm, obtaining the compensation value at each moment.

[0062] Due to the uncertainty of the surface feature integrity index A in the shielded areas (such as dense forests and building blind spots) in complex terrain, the data credibility fluctuates, which directly leads to the problem of out-of-control surveying and mapping accuracy in the shielded areas.

[0063] Therefore, the blind spot effectiveness scores calculated for all acquisition moments before the current moment are sorted in ascending chronological order to obtain the current blind spot effectiveness sequence. This current blind spot effectiveness sequence and the terrain complexity sequence are used as input to the XGBoost (eXtreme Gradient Boosting) algorithm. To improve the ability to fit complex relationships, the number of decision trees in this algorithm is set to 200; to prevent overfitting, the maximum tree depth is set to 6; and to balance convergence speed and accuracy, the learning rate is set to 0.05. The output is the predicted accuracy of the blind spot effectiveness score for the obscured area at the current moment, with a value ranging from 0 to 1. This quantifies the credibility of the restored real data in the obscured area and reflects the ability of sensor collaboration to penetrate obscuration.

[0064] Subsequently, the prediction accuracy of the shielded area obtained at all acquisition moments before the current moment is sorted in ascending order of time to obtain the prediction accuracy sequence at the current moment; the prediction accuracy sequence at the current moment and the reliability score sequence of the sensor are used as inputs of the weighted linear compensation algorithm to set the dynamic compensation coefficient, wherein, if the prediction accuracy P at the current moment is <0.7, the dynamic compensation coefficient is set to 0.3; if the prediction accuracy P at the current moment is ≥0.7, the dynamic compensation coefficient is set to 0.1 to give priority to low-precision areas. The output is the compensation value G at the current moment, which represents the final credibility score of the fusion device performance. Among them, the higher the prediction accuracy P, the less compensation is needed, and the lower the dynamic compensation coefficient should be. As other embodiments of the present application, the implementer can set the dynamic compensation coefficient according to actual conditions.

[0065] Step S5, constructing a shielding compensation confidence of each time point based on the compensation value and the prediction accuracy, and adjusting the motion path of the data acquisition equipment based on the shielding compensation confidence.

[0066] Based on the above analysis, for all data acquisition time points, the shielding compensation confidence of each acquisition time point is calculated, and the expression is:

[0067]

[0068] In the formula, B is the shielding compensation confidence of the current time mapping information; P is the prediction accuracy of the current time; G is the compensation value output by the weighted linear compensation algorithm of the current time; and Norm() is a normalization function.

[0069] P is the prediction accuracy of the shielding area, which is used to quantify the reliability of the sensor system in restoring the real ground features in the shielding environment, and directly reflects the penetration ability and data restoration reliability of the sensor system to the shielding area. The larger the value is, the more effectively the sensor system can penetrate sparse vegetation or building shadows to obtain high-fidelity point cloud and image data, and key micro features are captured completely; G is the compensation value output by the weighted linear compensation algorithm, which represents the compensation score of the reliability of the sensor system. The smaller the value is, the better the performance of the sensor itself is, and the lower the compensation demand is.

[0070] The shielding compensation confidence B quantifies the final reliability and reliability of the data obtained by multi-source sensor cooperation in the limited sensor view angle or physical shielding area, evaluates the ability to penetrate shielding and capture key micro features during geological space acquisition, and the higher the B value is, the smaller the data blind area is, the stronger the scene is, and the more reliable the engineering decision basis is; otherwise, it reveals serious data loss.

[0071] Due to the inherent line-of-sight shielding (such as dense vegetation, building bottoms, steep terrain) in complex geographical environments and the physical penetration limit of sensors, the traditional single-platform acquisition method causes serious data loss (such as ground micro-deformation, underground pipeline nodes, real topography under tree crowns) and a large number of micro-features (such as fine lines, manhole covers) in shielding areas, which directly affects the integrity of the mapping data and the engineering usability, and makes the subsequent analysis or construction have a significant error risk.

[0072] Therefore, a dynamic optimization strategy needs to be designed according to the shielding compensation confidence index B, and the specific dynamic optimization strategy design is as follows:

[0073] The trigger threshold is set, preferably, in the embodiments of the present application, the trigger threshold is set to 0.4. As other embodiments of the present application, the implementer can set the trigger threshold according to the actual situation. When the real-time calculated shielding compensation confidence is lower than the trigger threshold, it indicates that the confidence of the current area data acquisition is insufficient, there is a high risk of shielding omission or distortion, at this time, the DLite dynamic path planning algorithm is triggered immediately, the algorithm generates an optimal supplementary acquisition path online by comprehensively considering the current device position, the B value distribution of all acquisition time points and the known underground key nodes, guides the unmanned aerial vehicle to bypass the shielding object, and carries out targeted rescan along the optimal supplementary acquisition path, so as to maximize the acquisition of direct observation data of the shielded target. Therefore, the reliability of the shielding area data and the capture rate of the micro features are significantly improved.

[0074] The acquisition process of the terrain complexity is shown in the schematic diagram as Figure 2

[0075] Based on the same inventive concept as the above method, the embodiments of the present application also provide a high-precision geospatial information acquisition and mapping system, which comprises a memory, a processor and a computer program stored in the memory and running on the processor, and the processor implements the steps of any one of the methods in the above-mentioned high-precision geospatial information acquisition and mapping method when executing the computer program.

[0076] ​In summary, the embodiment of the present application provides a method for collecting and mapping geographic spatial information with high precision. Through the RGB images and thermal infrared images collected at each moment, RGB color values ​​and thermal radiation intensity values ​​are assigned to the point cloud data collected by the laser sensor at each moment, and a fused point cloud data set at each moment is constructed. In order to solve the problem of missing micro features in the shielded area due to the sensor viewing angle and penetration limitations, the terrain complexity at each moment is analyzed through the three-dimensional distribution characteristics of the point cloud data, and the fused point cloud data set is clustered to obtain each feature cluster. Combined with random forest, the shielding and missing features of the point cloud data are analyzed, and the effective score of the blind spot at each moment is constructed. The effective number of points per unit area of ​​the shielded area and the recognizability are quantified, and the shielded hot spot area with missing data is accurately located to avoid the blindness of traditional methods to the omission of micro features. To address the problem of decreased data credibility caused by fluctuations in accuracy in obscured areas and differences in sensor performance, a reliability score for point cloud data at each moment is constructed based on the difference between the vertical displacement of the sensor echo intensity at each moment and the preset benchmark true value. The prediction accuracy of the blind spot effective score at each moment is calculated in combination with the blind spot effective score and terrain complexity, and the shielding compensation confidence at each moment is constructed to comprehensively reflect the penetration ability of multi-source collaboration in the obscured area and the data restoration credibility, and eliminate the interference of sensor performance fluctuations on the data quality in the obscured area; based on the shielding compensation confidence at the acquisition moment, the path planning algorithm is triggered to guide the UAV to bypass obstacles and supplement the low-confidence area, significantly improving the integrity of the obscured area data and the micro-feature capture rate, reducing the blind spots of high-precision mapping, improving the integrity of scene data, and ensuring the engineering availability of complex scene data.

[0077] It should be noted that the order in which the embodiments of the present application are presented is for illustrative purposes only and does not necessarily represent the superiority or inferiority of the embodiments. Furthermore, the above descriptions are of specific embodiments of the present application. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order or sequential sequence shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0078] The various embodiments in this application are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments.

[0079] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A method for collecting and mapping high-precision geographic spatial information, characterized in that: The method comprises the following steps: For urban streets, RGB images and thermal infrared images collected at each moment are used to assign RGB color values ​​and thermal radiation intensity values ​​to the point cloud data collected by the laser sensor at each moment, thus constructing a fused point cloud dataset at each moment. The point cloud data at each moment is divided into grids, and the terrain parameters of each grid are constructed based on the three-dimensional distribution characteristics of the point cloud data within each grid to determine the terrain complexity at each moment. The reliability score of the point cloud data at each moment is constructed based on the difference between the vertical displacement of the laser sensor echo intensity at each moment and the preset reference true value. Cluster the fused point cloud dataset to obtain feature clusters, obtain the horizontal projection area of ​​the point cloud data in each feature cluster, and construct the number of effective points per unit area for each feature cluster based on the number of point clouds in the feature cluster; based on each feature cluster at each moment, use the random forest algorithm to obtain the label value of each feature cluster; calculate the effective score of the blind spot at each moment based on the number of effective points per unit area and the label value; The prediction accuracy of the blind spot effective score at each moment is calculated by using the blind spot effective scores of all moments before each moment and the terrain complexity; based on the prediction accuracy and reliability scores of all moments, combined with the compensation algorithm, the compensation value at each moment is obtained; The occlusion compensation confidence level at each moment is constructed based on the compensation value and the prediction accuracy, and the motion path of the data acquisition device is adjusted based on the occlusion compensation confidence level.

2. The method for collecting and mapping high-precision geographic spatial information according to claim 1, wherein: The process of obtaining the terrain complexity at each moment is as follows: The horizontal plane projected by the point cloud data at each moment is divided into grids of preset sizes, and the standard deviation of all point cloud elevation values ​​in each grid is used as the first evaluation value; the coefficient of variation of the point cloud projection point density of each grid and all its adjacent grids is calculated as the second evaluation value; the weighted sum of the first evaluation value and the second evaluation value of each grid is used as the terrain parameter of each grid, and the mean of the terrain parameters of all grids at each moment is used as the terrain complexity at each moment.

3. The method for collecting and mapping high-precision geographic spatial information according to claim 1, wherein: The process of obtaining the reliability score of the point cloud data at each time is as follows: Calculate the relative error between the vertical displacement of the laser sensor echo intensity at the current moment and the preset reference true value, and record the difference between the natural number 1 and the relative error as the reliability score of the point cloud data at the current moment.

4. The method for collecting and mapping high-precision geographic spatial information according to claim 1, wherein: The process of obtaining the number of effective points per unit area of ​​each feature cluster is as follows: The horizontal plane projection results of all point cloud data points in each feature cluster are used as the input of the convex hull algorithm to obtain the corresponding horizontal plane projection area; the ratio of the number of point cloud data points in each feature cluster to the horizontal plane projection area is calculated as the number of effective points per unit area of ​​each feature cluster.

5. The method for collecting and mapping high-precision geographic spatial information according to claim 1, wherein: The process of obtaining the label value of each feature cluster is as follows: The random forest model is trained by all feature clusters at historical moments, where the labels and label values ​​of each feature cluster include: valid features are 0, and masked features are 1; All feature clusters at the current moment are used as the input of the trained random forest, and the output is the label value of each feature cluster at the current moment.

6. The method for collecting and mapping high-precision geographic spatial information according to claim 1, wherein: The process of obtaining the effective score of the blind spot at each moment is as follows: Calculate the product of the number of effective points per unit area of ​​each feature cluster and the label value; and use the product of all feature clusters at the current moment as the effective score of the blind spot at the current moment.

7. The method for collecting and mapping high-precision geographic spatial information according to claim 1, wherein: The process of obtaining the prediction accuracy of the blind spot effective score at each moment is as follows: The sequence of terrain complexity at all times before the current moment is used as the terrain complexity sequence at the current moment; the sequence of blind spot effective scores at all times before the current moment is used as the blind spot effective sequence at the current moment; The blind spot validity sequence and terrain complexity sequence at the current moment are used as the input of the XGBoost algorithm, and the output is the prediction accuracy of the blind spot validity score at the current moment.

8. The method for collecting and mapping high-precision geographic spatial information according to claim 1, wherein: The process of obtaining the compensation value at each moment is as follows: The sequence of the prediction accuracies at all moments before the current moment is used as the prediction accuracy sequence at the current moment; the sequence of the reliability scores at all moments before the current moment is used as the reliability score sequence at the current moment; The prediction accuracy sequence and reliability score sequence at the current moment are used as the input of the weighted linear compensation algorithm, and the output is the compensation value at the current moment.

9. The method for collecting and mapping high-precision geographic spatial information according to claim 1, wherein: The occlusion compensation confidence level at each moment is constructed, and the motion path of the data acquisition device is adjusted based on the occlusion compensation confidence level, specifically by taking the normalized value of the ratio of the prediction accuracy to the compensation value at the current moment as the occlusion compensation confidence level at the current moment; If the occlusion compensation confidence at the current moment is lower than the preset trigger threshold, the path planning algorithm is immediately triggered to generate the optimal re-mining path for the UAV and re-scan the area along the path.

10. A high-precision geographic spatial information acquisition and mapping system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 9 are implemented.

Images