A three-dimensional high-precision map generation method, system and cloud platform

Through drones collecting data and using structural relationship tree and neural matching space technology, data vulnerabilities and noise problems in the generation of three-dimensional maps of complex buildings are solved, and high-precision and accurate three-dimensional map reconstruction is achieved.

CN117078873BActive Publication Date: 2024-11-08BEIJING YINUO HEJU DIGITAL TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202310888809.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-07-19
Publication Date
2024-11-08
Estimated Expiration
2043-07-19

AI Technical Summary

Technical Problem

The prior art is difficult to accurately generate high-precision three-dimensional maps of complex buildings, especially data vulnerabilities and noise caused by occlusion between buildings and complex geometric structures, making it difficult to accurately reconstruct three-dimensional maps.

Method used

By setting up drones to collect tilt image data and pulse radar reflection signals of buildings, establish initial three-dimensional image model and point cloud data, use structural relationship trees and structural correlation forests to repair missing parts, and perform three-dimensional images in neural matching space The model is spliced ​​to realize the generation of regional three-dimensional maps.

Benefits of technology

It improves the accuracy of the three-dimensional map and is close to the actual scene, ensures the completeness and accuracy of data, and solves the problems of data vulnerabilities and noise.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117078873B_ABST
    Figure CN117078873B_ABST
Patent Text Reader

Abstract

The present invention discloses a three-dimensional high-precision map generation method, system and cloud platform, and relates to the technical field of three-dimensional images. The present invention collects oblique image data and pulse radar reflection signals of buildings by using a drone, establishes an initial three-dimensional image model according to the oblique image data, and establishes three-dimensional point cloud data according to the pulse radar reflection signals, overlaps and maps the three-dimensional image model with the three-dimensional point cloud data to establish a quasi-three-dimensional image model, establishes a structural relationship tree according to the correlation between various facilities in the building, establishes a structural association forest according to the correlation between various buildings, and then repairs the missing parts of the quasi-three-dimensional image model to obtain a three-dimensional image model, and according to the corresponding correlation between various buildings in the structural association forest, splices various three-dimensional image models in the form of neuron matching to obtain a regional three-dimensional map, and then obtains a high-precision regional three-dimensional map.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of three-dimensional image technology, and specifically to a three-dimensional high-precision map generation method, system and cloud platform. Background Art

[0002] High-precision three-dimensional maps are widely used in architectural design, itinerary navigation, urban planning and other fields. However, due to the complex topological structure, geometric structure and hierarchical structure of buildings, it is difficult to accurately describe various information of buildings on the map.

[0003] In the prior art, a constrained shape estimation solution space is often used to provide a strong prior, but it is usually limited to the predefined topological structure of the building. It is difficult to process and understand deformations beyond the range, making it difficult to reconstruct buildings with more complex topological structures.

[0004] In actual scenarios, due to the complexity of the environment in which buildings are located and the interference between buildings, it is difficult to accurately collect building information. For example, occlusion between buildings and complex geometric structures may cause holes or noise in the map. How to avoid data loopholes while establishing a high-precision three-dimensional map is the current technical difficulty. For this purpose, a three-dimensional high-precision map generation method, system and cloud platform are provided. Summary of the invention

[0005] In order to solve the above technical problems, the purpose of the present invention is to provide a three-dimensional high-precision map generation method, system and cloud platform.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] A method for generating a three-dimensional high-precision map comprises the following steps:

[0008] Step 1: Set image control points for each building, and set the flight path of the drone according to the location of the image control points, and then use the drone to collect the architectural structure information of the building;

[0009] Step 2: Establish an initial 3D image model and 3D point cloud data of the building according to the building structure information, and overlap and map the 3D image model and the 3D point cloud data to establish a quasi 3D image model;

[0010] Step 3: Establish a structural relationship tree according to the correlation between the facilities in the building, and establish a structural correlation forest according to the correlation between the buildings, and then repair the missing parts of the quasi-3D image model of each building to obtain a 3D image model;

[0011] Step 4: According to the association relationship between each building in the structural association forest, the three-dimensional image models are spliced ​​in the form of neural matching to obtain a three-dimensional map of the region.

[0012] Furthermore, the establishment of the initial three-dimensional image model includes:

[0013] The building structure information includes a building number, a first pulse radar reflection signal, a second pulse radar reflection signal, a first oblique image data, and a second oblique image data;

[0014] Performing image correction and image registration on the first oblique image data and the second oblique image data respectively, and then registering the image parts at different oblique angles of the two images to the same longitude and latitude, so that the geometric qualities of the two images are consistent;

[0015] A three-dimensional model of the building structure is established based on the processed first oblique image data and the second oblique image data, and the three-dimensional models of the building structure corresponding to the two are overlapped and mapped to obtain an initial three-dimensional image model, and a plurality of facility location feature points are set on the initial three-dimensional image model.

[0016] Furthermore, the process of establishing the three-dimensional point cloud data includes: filtering, denoising and registering the first pulse radar reflection signal and the second pulse radar reflection signal, establishing the three-dimensional point cloud data according to each peak point and corresponding signal strength on the first pulse radar reflection signal and the second pulse radar reflection signal, overlapping and mapping the two three-dimensional point cloud data to obtain complete three-dimensional point cloud data,

[0017] The point cloud data in the three-dimensional point cloud data and the facility location feature points in the initial three-dimensional image model are each provided with a corresponding facility image model, and the facility image model is in a virtual state.

[0018] Furthermore, the process of establishing the quasi-3D image model includes:

[0019] The initial three-dimensional image model is mapped and overlapped with the complete three-dimensional point cloud data, and the facility location feature points where the point cloud data are in an overlapping mapping state are marked as confirmed parts. The facility image model corresponding to the confirmed parts is then converted from a virtual shadow state to a real shadow state, thereby obtaining a quasi-three-dimensional image model, and the remaining facility location feature points that have not been successfully matched are marked as missing parts.

[0020] Furthermore, the process of establishing the structural relationship tree includes:

[0021] All facility location feature points are extracted from the initial three-dimensional image model, a three-dimensional space xyz coordinate system is established, and each facility location feature point is mapped to the three-dimensional space xyz coordinate system based on its corresponding position in the initial three-dimensional image model. Then, starting from the origin of the three-dimensional space xyz coordinate system, each facility location feature point is connected in sequence along the z-axis of the three-dimensional space xyz coordinate system. For the setting location feature points at the same position relative to the z-axis, they are connected in sequence along the x-axis. For the setting location feature points at the same position relative to the z-axis and the x-axis, they are connected in sequence along the y-axis, thereby obtaining a structural relationship tree.

[0022] Furthermore, the process of establishing the structural association forest includes:

[0023] According to the adjacent relationship of each building in the actual scene, the tree nodes on the corresponding structural relationship tree are mapped and matched with each other, the tree nodes with successful mapping and matching are connected with each other, and the tree nodes with failed mapping and matching are not connected, so as to obtain the structural association forest;

[0024] In the structural association forest, the tree nodes on the structural relationship tree corresponding to each building are marked with colors, the colors of the tree nodes on the structural relationship trees of adjacent buildings are different, and the connection tree nodes of adjacent structural relationship trees have two colors at the same time.

[0025] Furthermore, the process of repairing the quasi-3D image model includes:

[0026] According to the serial number of each building, the corresponding quasi-3D image model is mapped to the structure association forest, and then the tree nodes on the corresponding structure relationship tree in the structure association forest are mapped and matched with the facility location feature points with missing part marks on the quasi-3D image model;

[0027] The facility location feature point marks that are successfully mapped and matched with the tree nodes are changed to confirmation parts, and the corresponding facility image models are transformed from virtual image states to real image states, thereby obtaining a three-dimensional image model. The remaining facility location feature points that have not been successfully mapped and matched are automatically deleted.

[0028] Furthermore, the establishment of the regional three-dimensional map includes: based on the tree nodes with two color labels on each structural relationship tree in the structural association forest, setting the facility location feature points at the same position in the three-dimensional image model corresponding to the structural relationship tree as connection synapses; establishing a neural matching space, and setting a plurality of signal receptors in the neural matching space;

[0029] The pre-connection synapse or post-connection synapse of each three-dimensional image model sends a request connection signal or a response connection information to the signal receptor. The signal receptor matches and combines the structure association forest stored in it and each three-dimensional image model according to the signal contained in the request connection signal or the response connection information, and then matches and connects the corresponding pre-connection synapse and post-connection synapse to the signal. When the pre-connection synapse or post-connection synapse of all three-dimensional image models are connected, a regional three-dimensional map is obtained.

[0030] Further, a three-dimensional high-precision map generation system includes a cloud computing center, wherein the cloud computing center is communicatively connected to a regional building information acquisition module, a building information analysis module, a three-dimensional image model building module, and a regional three-dimensional map generation module;

[0031] The regional building information collection module is provided with a drone management unit and an information collection unit, which are used to establish a drone flight route and manage the drone to take photos of buildings in the collection area and send pulse radar signals;

[0032] The building information analysis module is used to establish the initial three-dimensional image model three-dimensional point cloud data of each building, and then overlap and map the two to obtain a quasi-three-dimensional image model;

[0033] The three-dimensional image model building module is used to build a structural relationship tree and a structural association forest, and according to the association relationship of the structural relationship tree in the structural association forest, the three-dimensional image model is corrected to obtain a three-dimensional image model;

[0034] The regional three-dimensional map generation module is used to establish a neural matching space, and then set connection synapses for the corresponding three-dimensional image models according to the corresponding relationships between the structural relationship trees in the structural association forest, and add the three-dimensional image models to the neural matching space for connection synapse matching to obtain a regional three-dimensional map.

[0035] Furthermore, a three-dimensional high-precision map generation cloud platform includes a master cloud node, and the master cloud node communicates and manages information collection work cloud nodes, information analysis work cloud nodes, and map creation work cloud nodes;

[0036] The information collection work cloud node is used to collect various information of buildings in the area and send it to the information analysis work cloud node;

[0037] The information analysis work cloud node is used to establish a three-dimensional image model according to various information of the building;

[0038] The map building work cloud node is used to map and match the three-dimensional image models of various buildings to each other, thereby obtaining a regional building map.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] 1. The present invention collects the oblique image data and pulse radar reflection signals of the building, and then establishes the initial three-dimensional image model and three-dimensional point cloud data of the building, and then maps and overlaps the two to obtain a quasi-3D image model. At the same time, according to the mutual relationship between the facilities in the rent-reduced building and the corresponding relationship of the building in the actual scene, a structural relationship tree and a structural association forest are established, and then the missing parts in the quasi-3D image model are repaired, which provides data guarantee for the accuracy of the subsequent regional three-dimensional map;

[0041] 2. The present invention establishes a neural matching space and sets connection synapses for each three-dimensional image model. By adding the three-dimensional image model to the neural matching space and setting signal receptors, each three-dimensional image model is matched and connected to obtain a regional three-dimensional map. The matching of the three-dimensional image model is adjusted according to the adjacent positions of each building in the actual scene, so that the established regional three-dimensional map is closer to the actual scene and has higher accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0043] Figure 1 is a flow chart of the method of the present invention;

[0044] Figure 2 It is a system principle diagram of the present invention. DETAILED DESCRIPTION

[0045] To make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be described in detail below. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other implementation methods obtained by ordinary technicians in this field without creative work belong to the scope of protection of the present invention.

[0046] like Figure 1 As shown, a three-dimensional high-precision map generation method includes the following steps:

[0047] Step 1: Set image control points for each building, and set the flight path of the drone according to the location of the image control points, so as to collect the oblique image data and pulse radar reflection signal of the building through the drone;

[0048] Specifically, a number of image control points are set for each building in the acquisition area, and each image control point is numbered S1, S2, ..., S n , where n is a natural number greater than 0, and n represents the total number of buildings in the acquisition area, and each image control point is equipped with a wireless signal sending device;

[0049] The size of the acquisition area is determined according to the distribution of each image control point in the acquisition area, and two UAV flight routes in opposite directions are set according to the number of each image control point. The first UAV flight route is S1, S2, ..., S n , the second UAV flight route is S n , S n-1 , …, S2, S1;

[0050] It should be noted that the drone is equipped with a wireless signal receiving device, a pulse radar signal device, and an oblique photography camera;

[0051] Further, two UAVs are set to fly simultaneously along the first UAV flight route and the second UAV flight route at the same speed;

[0052] Before the UAV reaches the image control point, it obtains the wireless signal sent by the image control point through the wireless signal receiving device, determines its current location, turns on the pulse radar signal device and the tilt camera, and then obtains the pulse radar reflection signal and image data of the building at the current image control point;

[0053] It should be noted that during the flight of the drone, if the drone receives wireless signals from two image control points at the same time, it will determine the signal strengths of the two wireless signals in real time, and then give priority to collecting the pulse radar reflection signal and oblique image data of the image control point corresponding to the wireless signal with greater signal strength;

[0054] When two UAVs arrive at the end point of their assigned UAV flight routes at the same time, the pulse radar reflection signals and image data collected by the two UAVs at the same numbered image control points are merged to obtain the building structure information of the buildings corresponding to each image control point, where the building structure information includes the building number, the first pulse radar reflection signal, the second pulse radar reflection signal, the first oblique image data and the second oblique image data.

[0055] Step 2: Establish an initial three-dimensional image model of the building according to the oblique image data in the building structure information, and establish point cloud data of the building according to the pulse radar reflection signal in the building structure information, and overlap and map the three-dimensional image model and the point cloud data to establish a quasi-three-dimensional image model;

[0056] Specifically, extracting a first pulse radar reflection signal, a second pulse radar reflection signal, a first oblique image data, and a second oblique image data from the building structure information;

[0057] Firstly, image correction and image registration are performed on the first oblique image data and the second oblique image data respectively, and then the images at different oblique angles in the first oblique image data and the second oblique image data are registered to the same longitude and latitude, so that the geometric quality of the first oblique image data and the second oblique image data is guaranteed to be consistent, thereby improving the data accuracy for the subsequent establishment of a three-dimensional image model of the building;

[0058] Establishing a three-dimensional model of the building structure according to the processed first oblique image data and the second oblique image data, and overlapping and mapping the three-dimensional models of the building structure corresponding to the first oblique image data and the second oblique image data, thereby obtaining an initial three-dimensional image model;

[0059] It should be noted that a number of facility location feature points are set on the initial three-dimensional image model, and each facility location feature point in the initial three-dimensional image model corresponds to a facility image model in the building, but the facility image model is in a virtual state in the initial three-dimensional image model;

[0060] Further, after filtering, denoising and registering the first pulse radar reflection signal and the second pulse radar reflection signal, three-dimensional point cloud data is established according to each peak point and signal strength on the first pulse radar reflection signal and the second pulse radar reflection signal, and the two three-dimensional point cloud data are overlapped and mapped to obtain complete three-dimensional point cloud data. It should be noted that each point cloud data in the three-dimensional point cloud data is provided with a corresponding facility image model in the building and is in a virtual state. The facilities in the building are such as stairs, walls, doors and windows;

[0061] The initial three-dimensional image model is mapped and overlapped with the complete three-dimensional point cloud data, and the facility location feature points where the facility location feature points and the point cloud data are in an overlapping mapping state are marked as confirmed parts, and then the facility image model corresponding to the confirmed parts is converted from a virtual image state to a real image state, thereby obtaining a quasi-three-dimensional image model;

[0062] It should be noted that if there is a facility location feature point on the initial three-dimensional image model that has no point cloud data that overlaps with it, or there is no facility location feature point that overlaps with the point cloud data in the complete three-dimensional point cloud data, then it is judged that the facility location corresponding to the facility location feature point or the point cloud data is missing, and then the missing part is marked at the location of the facility location feature point or point cloud data on the quasi-three-dimensional image model.

[0063] Step 3: Establish a structural relationship tree according to the correlation between the facilities in the building, and establish a structural correlation forest according to the correlation between the buildings, and then repair the missing parts of the quasi-3D image model of each building to obtain a 3D image model;

[0064] Specifically, all facility location feature points are extracted from the initial three-dimensional image model, a three-dimensional space xyz coordinate system is established, and each facility location feature point is mapped to the three-dimensional space xyz coordinate system based on its corresponding position in the initial three-dimensional image model, and then starting from the origin of the three-dimensional space xyz coordinate system, each facility location feature point is sequentially connected along the z-axis of the three-dimensional space xyz coordinate system, and the setting location feature points at the same position relative to the z-axis are sequentially connected along the x-axis, and the setting location feature points at the same position relative to the z-axis and the x-axis are sequentially connected along the y-axis, thereby obtaining a structural relationship tree;

[0065] Furthermore, in the process of the drone collecting the oblique image data and pulse radar reflection signals of the buildings along the flight route of the drone, if the distances between the buildings corresponding to the image control points are relatively close, the oblique image data and pulse radar reflection signals of the buildings collected by the drone may contain relevant information about the adjacent buildings.

[0066] Then, according to the actual adjacent relationship of each building, a structural association forest is established;

[0067] The process of establishing the structural association forest includes: firstly obtaining the position and orientation of the adjacent buildings of each building in the actual scene, and then mapping and matching the tree nodes on the structural relationship tree corresponding to each building according to the adjacent relationship of each building in the actual scene, connecting the tree nodes with successful mapping and matching, and disconnecting the tree nodes with failed mapping and matching;

[0068] It should be noted that in the structural association forest, the tree nodes on the structural relationship tree corresponding to each building are marked with colors, the colors of the tree nodes on the structural relationship trees of adjacent buildings are different, and the connection tree nodes of adjacent structural relationship trees have two colors at the same time;

[0069] For example, number S i and S j In the actual scene, there are two adjacent buildings, one of which is numbered S i The building is located in S j The building numbered S i The building's south-facing facility location feature point is the same as the one numbered S jThe facility location feature points of the building facing due north are mapped and matched. If there is a facility location feature point on each side that corresponds to the same facility image model, it is determined that the above two facility location feature points are mapped and matched with each other, and no operation is performed on the remaining unmapped facility location feature points;

[0070] According to the adjacent relationship of each building in the actual scene, all structural relationship trees are interconnected to obtain the structural relationship forest;

[0071] Furthermore, according to the serial number of each building, the corresponding quasi-3D image model is mapped to the structure association forest, and then the tree nodes on the corresponding structure relationship tree in the structure association forest are mapped and matched with the facility location feature points and point cloud data with missing part marks on the quasi-3D image model;

[0072] The facility location feature points or point cloud data that are successfully mapped and matched with the tree nodes are marked as the confirmation part, and then the facility image model corresponding to the facility location feature points or point cloud data marked with the confirmation part is converted from a virtual image state to a real image state, thereby obtaining a three-dimensional image model.

[0073] Step 4: According to the association relationship of each building in the structural association forest, the three-dimensional image models are spliced ​​in the form of neural matching to obtain a three-dimensional map of the area;

[0074] Specifically, each structural relationship tree in the structural association forest has tree nodes labeled with two colors, and then the facility location feature points at the same position in the three-dimensional image model corresponding to the structural relationship tree are set as connection synapses;

[0075] It should be noted that the connection synapse includes a pre-connection synapse and a post-connection synapse, wherein the pre-connection synapse is used to send a connection request signal, and the post-connection synapse is used to send a connection response information, wherein the connection request signal includes the number of the pre-connection synapse in the three-dimensional image model and its orientation angle in the three-dimensional image model, and the response connection signal includes the number of the post-connection synapse in the three-dimensional image model and its orientation in the three-dimensional image model, and on the three-dimensional image model, the number of the pre-connection synapse and the post-connection synapse is the same, and according to the corresponding relationship between the tree node and the structural relationship tree associated with it in the structural association forest, the corresponding facility location feature point is marked as a pre-connection synapse in one three-dimensional image model, and the facility location feature point corresponding to the facility location feature point on another three-dimensional image model is marked as a post-connection synapse;

[0076] A neural matching space is established, and a plurality of signal receptors are set in the neural matching space, and then all three-dimensional image models are sequentially placed in the neural matching space according to their numbers, wherein the signal receptors store the structural association forest and each three-dimensional image model;

[0077] Further, the pre-connection synapse or post-connection synapse of each three-dimensional image model sends a connection request signal or a connection response information to the signal receptor, and the signal receptor matches and combines the structure association forest stored in it and each three-dimensional image model according to the signal included in the connection request signal or the connection response information, and then matches and connects the corresponding pre-connection synapse and post-connection synapse signals;

[0078] When all pre-connection synapses or post-connection synapses of the three-dimensional image model are connected, a three-dimensional map of the region is obtained;

[0079] It should be noted that when the three-dimensional image model in the regional three-dimensional map is updated or a new three-dimensional image model is added, the corresponding structural relationship tree and structural association forest must be updated first, or a new structural relationship tree must be established and added to the structural association forest, and then the regional three-dimensional map will be updated.

[0080] The present invention also discloses a three-dimensional high-precision map generation system, comprising a cloud computing center, wherein the cloud computing center is communicatively connected with a regional building information acquisition module, a building information analysis module, a three-dimensional image model building module, and a regional three-dimensional map generation module;

[0081] The regional building information collection module is provided with a drone management unit and an information collection unit;

[0082] The drone acquisition unit is used to establish a drone flight route according to the locations of the image control points in the acquisition area, and manage the drone to take photos of buildings in the acquisition area and send pulse radar signals during the flight;

[0083] The information acquisition module is used to pre-process the information collected by the drone, thereby obtaining the oblique image data and pulse radar reflection signals of each building in the acquisition area;

[0084] The building information analysis module is used to establish an initial three-dimensional image model based on the tilted image data of each building, and to establish three-dimensional point cloud data based on the pulse radar reflection signal, and then overlap and map the initial three-dimensional image model with the three-dimensional point cloud data to obtain a quasi-three-dimensional image model;

[0085] The three-dimensional image model building module is used to establish a structural relationship tree and a structural association forest according to the initial three-dimensional image and the positional relationship of each building in the actual scene, and to correct the three-dimensional image model according to the association relationship of the structural relationship tree in the structural association forest to obtain a three-dimensional image model;

[0086] The regional three-dimensional map generation module is used to establish a neural matching space, and then set connection synapses for the corresponding three-dimensional image models according to the corresponding relationships between the structural relationship trees in the structural association forest, and add the three-dimensional image models to the neural matching space for connection synapse matching, so as to obtain a regional three-dimensional map.

[0087] The present invention discloses a three-dimensional high-precision map generation cloud platform including a master cloud node, wherein the master cloud node communicates and manages information collection work cloud nodes, information analysis work cloud nodes, and map creation work cloud nodes;

[0088] The information collection work cloud node is used to collect various information of buildings in the area and send it to the information analysis work cloud node;

[0089] The information analysis work cloud node is used to establish a three-dimensional image model according to various information of the building;

[0090] The map building work cloud node is used to map and match the three-dimensional image models of various buildings to each other, thereby obtaining a regional building map.

[0091] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A method for generating a three-dimensional high-precision map, characterized in that: The following steps are involved: Step 1: Set image control points for each building, and set the flight path of the drone according to the location of the image control points, and then use the drone to collect the architectural structure information of the building; Step 2: Establish an initial 3D image model and 3D point cloud data of the building according to the building structure information, and overlap and map the 3D image model and the 3D point cloud data to establish a quasi 3D image model; Step 3: Establish a structural relationship tree according to the correlation between the facilities in the building, and establish a structural correlation forest according to the correlation between the buildings, and then repair the missing parts of the quasi-3D image model of each building to obtain a 3D image model; Step 4: According to the association relationship between each building in the structural association forest, the three-dimensional image models are spliced ​​in the form of neural matching to obtain a three-dimensional map of the area; The establishment of the regional three-dimensional map includes: Based on the tree nodes with two color labels on each structural relationship tree in the structural association forest, the facility location feature points at the same position in the three-dimensional image model corresponding to the structural relationship tree are set as connection synapses, a neural matching space is established, and a number of signal receptors are set in the neural matching space, and then all the three-dimensional image models are placed in the neural matching space in sequence according to the numbers; The connection synapse includes a pre-connection synapse and a post-connection synapse, wherein the pre-connection synapse is used to send a connection request signal, and the post-connection synapse is used to send a connection response information, wherein the connection request signal includes the number of the pre-connection synapse in the three-dimensional image model and its orientation angle in the three-dimensional image model, and the response connection signal includes the number of the post-connection synapse in the three-dimensional image model and its orientation in the three-dimensional image model, and on the three-dimensional image model, the number of the pre-connection synapse and the post-connection synapse is the same, and according to the corresponding relationship between the tree node and the structural relationship tree associated with it in the structural association forest, the corresponding facility location feature point is marked as a pre-connection synapse in one three-dimensional image model, and the facility location feature point corresponding to the facility location feature point on another three-dimensional image model is marked as a post-connection synapse; The pre-connection synapse or post-connection synapse of each three-dimensional image model sends a request connection signal or a response connection information to the signal receptor. The signal receptor matches and combines the structure association forest stored in it and each three-dimensional image model according to the signal contained in the request connection signal or the response connection information, and then matches and connects the corresponding pre-connection synapse and post-connection synapse to the signal. When the pre-connection synapse or post-connection synapse of all three-dimensional image models are connected, a regional three-dimensional map is obtained.

2. A three-dimensional high-precision map generation method according to claim 1, characterized in that: The establishment of the initial three-dimensional image model includes: The building structure information includes a building number, a first pulse radar reflection signal, a second pulse radar reflection signal, a first oblique image data, and a second oblique image data; Performing image correction and image registration on the first oblique image data and the second oblique image data respectively, and then registering the image parts at different oblique angles of the two images to the same longitude and latitude, so that the geometric qualities of the two images are consistent; A three-dimensional model of the building structure is established based on the processed first oblique image data and the second oblique image data, and the three-dimensional models of the building structure corresponding to the two are overlapped and mapped to obtain an initial three-dimensional image model, and a plurality of facility location feature points are set on the initial three-dimensional image model.

3. A three-dimensional high-precision map generation method according to claim 2, characterized in that: The process of establishing the three-dimensional point cloud data includes: filtering, denoising and registering the first pulse radar reflection signal and the second pulse radar reflection signal, establishing the three-dimensional point cloud data according to the peak points and corresponding signal strengths on the first pulse radar reflection signal and the second pulse radar reflection signal, and overlapping and mapping the two three-dimensional point cloud data to obtain complete three-dimensional point cloud data. The point cloud data in the three-dimensional point cloud data and the facility location feature points in the initial three-dimensional image model are each provided with a corresponding facility image model, and the facility image model is in a virtual state.

4. A three-dimensional high-precision map generation method according to claim 3, characterized in that: The process of establishing the quasi-3D image model includes: The initial three-dimensional image model is mapped and overlapped with the complete three-dimensional point cloud data, and the facility location feature points where the point cloud data are in an overlapping mapping state are marked as confirmed parts. The facility image model corresponding to the confirmed parts is then converted from a virtual shadow state to a real shadow state, thereby obtaining a quasi-three-dimensional image model, and the remaining facility location feature points that have not been successfully matched are marked as missing parts.

5. A three-dimensional high-precision map generation method according to claim 2, characterized in that: The process of establishing the structural relationship tree includes: All facility location feature points are extracted from the initial three-dimensional image model, a three-dimensional space xyz coordinate system is established, and each facility location feature point is mapped to the three-dimensional space xyz coordinate system based on its corresponding position in the initial three-dimensional image model. Then, starting from the origin of the three-dimensional space xyz coordinate system, each facility location feature point is connected in sequence along the z-axis of the three-dimensional space xyz coordinate system. For the setting location feature points at the same position relative to the z-axis, they are connected in sequence along the x-axis. For the setting location feature points at the same position relative to the z-axis and the x-axis, they are connected in sequence along the y-axis, thereby obtaining a structural relationship tree.

6. A three-dimensional high-precision map generation method according to claim 5, characterized in that: The process of establishing the structural association forest includes: According to the adjacent relationship of each building in the actual scene, the tree nodes on the corresponding structural relationship tree are mapped and matched with each other, the tree nodes with successful mapping and matching are connected with each other, and the tree nodes with failed mapping and matching are not connected, so as to obtain the structural association forest; In the structural association forest, the tree nodes on the structural relationship tree corresponding to each building are marked with colors, the colors of the tree nodes on the structural relationship trees of adjacent buildings are different, and the connection tree nodes of adjacent structural relationship trees have two colors at the same time.

7. A three-dimensional high-precision map generation method according to claim 6, characterized in that: The process of repairing the quasi-3D image model includes: According to the serial number of each building, the corresponding quasi-3D image model is mapped to the structure association forest, and then the tree nodes on the corresponding structure relationship tree in the structure association forest are mapped and matched with the facility location feature points with missing part marks on the quasi-3D image model; The facility location feature point marks that are successfully mapped and matched with the tree nodes are changed to confirmation parts, and the corresponding facility image models are transformed from virtual image states to real image states, thereby obtaining a three-dimensional image model. The remaining facility location feature points that have not been successfully mapped and matched are automatically deleted.

8. A three-dimensional high-precision map generation system, used to implement a three-dimensional high-precision map generation method according to any one of claims 1 to 7, comprising a cloud computing center, characterized in that: The cloud computing center is communicatively connected to a regional building information acquisition module, a building information analysis module, a three-dimensional image model building module, and a regional three-dimensional map generation module; The regional building information collection module is provided with a drone management unit and an information collection unit; The drone management unit is used to establish a drone flight route according to the location of the image control points in the acquisition area, and manage the drone to take photos of buildings in the acquisition area and send pulse radar signals during the flight; The information acquisition unit is used to pre-process the information collected by the drone, thereby obtaining the oblique image data and pulse radar reflection signals of each building in the acquisition area; The building information analysis module is used to establish an initial three-dimensional image model based on the tilted image data of each building, and to establish three-dimensional point cloud data based on the pulse radar reflection signal, and then overlap and map the initial three-dimensional image model with the three-dimensional point cloud data to obtain a quasi-three-dimensional image model; The three-dimensional image model building module is used to build a structural relationship tree and a structural association forest, and according to the association relationship of the structural relationship tree in the structural association forest, the three-dimensional image model is corrected to obtain a three-dimensional image model; The regional three-dimensional map generation module is used to establish a neural matching space, and then set connection synapses for the corresponding three-dimensional image models according to the corresponding relationships between the structural relationship trees in the structural association forest, and add the three-dimensional image models to the neural matching space for connection synapse matching to obtain a regional three-dimensional map.

9. A three-dimensional high-precision map generation cloud platform, used to implement a three-dimensional high-precision map generation method according to any one of claims 1 to 7, comprising a master control cloud node, characterized in that: The master cloud node communication management includes information collection work cloud node, information analysis work cloud node, and map creation work cloud node; The information collection work cloud node is used to collect various information of buildings in the area and send it to the information analysis work cloud node; The information analysis work cloud node is used to establish a three-dimensional image model according to various information of the building; The map building work cloud node is used to map and match the three-dimensional image models of various buildings to each other, thereby obtaining a regional building map.