Old building digital surveying and mapping system and method based on air-ground collaborative point cloud fusion

Abstract

The invention relates to the technical field of building surveying and mapping, and provides an old building digital surveying and mapping system and method based on air-ground collaborative point cloud fusion, and the method comprises the steps: carrying out the blind area detection of space-time fusion data scanned by unmanned aerial vehicle equipment and a ground robot, and carrying out the blind area detection when the top point cloud data / bottom point cloud data have a blind area visual angle; starting a dynamic master-slave mechanism; according to the dynamic master-slave mechanism, equipment corresponding to the point cloud data with the blind area serves as slave equipment, equipment corresponding to the point cloud data without the blind area serves as host equipment, the slave equipment is controlled to execute point cloud filling, blind area compensation data are determined, and after the blind area compensation data are fused into time-space fusion data, the time-space fusion data are subjected to time-space fusion. And generating target space-time fusion data loaded to the visual interface. A closed-loop workflow from data acquisition to quality verification is constructed through a dynamic master-slave control mechanism, a multi-dimensional synchronization strategy and an intelligent compensation algorithm driven by architectural features.

Description

Technical Field

[0001] The present invention relates to the technical field of building surveying and mapping, and in particular to a digital surveying and mapping system and method for old buildings based on air-ground collaborative point cloud fusion. Background Art

[0002] The surveying and mapping of historic buildings, due to their complex structural features, delicate decorative components, and irregular geometric forms, places extremely high demands on the accuracy, integrity, and adaptability of surveying and mapping technology. Traditional surveying and mapping technologies have significant limitations in multi-device collaboration, dynamic environmental adaptability, and intelligent compensation mechanisms, making them difficult to meet the high standards required for the preservation and restoration of historic buildings.

[0003] In existing technologies, single-platform scanning solutions are limited by physical viewing angle constraints and cannot effectively cover blind spots at the top and bottom of buildings. Drone scanning easily misses hidden areas such as eaves and the back of carvings, while ground-based equipment has difficulty obtaining high-level details such as roofs. This type of method relies on manual experience to identify blind spots, which can easily lead to missed detections or misjudgments. In addition, the generation of compensation data often uses simple interpolation algorithms, resulting in geometric distortion and broken texture features. When heterogeneous data is fused, the spatiotemporal asynchrony between devices causes coordinate offsets and motion blur. Traditional registration algorithms are not adaptable enough to complex structures and are prone to error accumulation. In addition, the fixed master-slave control mode lacks flexibility in dynamic environments, equipment collaboration is inefficient, and the ability to recover after data loss is weak.

[0004] Existing compensation technologies often rely on manual modeling or general generation algorithms, failing to fully consider architectural constraints. This results in a poor match between the generated data and the original structure's morphology, texture, and mechanical properties. The lack of quality verification makes error tracing difficult, and data quality relies on manual screening at a later stage, which is inefficient and unreliable. The industry has long faced core challenges, including a mismatch between high-precision requirements and equipment capabilities, low efficiency in processing massive amounts of data, and poor adaptability to dynamic environments.

[0005] In recent years, technological improvements have focused on hardware performance enhancements or local algorithm optimizations, such as increasing the power of scanning devices, improving the efficiency of registration algorithms, or introducing additional mobile platforms. However, these solutions have failed to overcome the inherent limitations of single-device data acquisition, nor have they established a cross-platform intelligent collaborative framework. Consequently, they have been unable to systematically address the coordination issues of blind spot compensation, multi-source data fusion, and dynamic control. Summary of the Invention

[0006] This paper proposes a system and method for digital mapping of old buildings based on collaborative air-ground point cloud fusion. By leveraging a dynamic master-slave control mechanism, a multi-dimensional synchronization strategy, and an intelligent compensation algorithm driven by building features, this system establishes a closed-loop workflow from data acquisition to quality verification. This solution effectively leverages the spatial coverage advantages of drones and ground robots, combining real-time fusion with adaptive compensation technology to provide a high-precision, high-efficiency, and highly robust full-process solution for the digitization of old buildings.

[0007] In a first aspect, the embodiments of the present application provide a method for digital mapping of old buildings based on air-ground collaborative point cloud fusion, comprising:

[0008] Perform blind spot detection on the spatiotemporal fusion data scanned by UAV equipment and ground robots, and activate the dynamic master-slave mechanism when there is a blind spot view in the top point cloud data / bottom point cloud data;

[0009] Among them, the dynamic master-slave mechanism regards the device corresponding to the point cloud data with blind spots as the slave device, and the device corresponding to the point cloud data without blind spots as the host device, and controls the slave device to perform point cloud filling, determine the blind spot compensation data, and fuse the blind spot compensation data into the spatiotemporal fusion data to generate the target spatiotemporal fusion data loaded into the visualization interface.

[0010] In an embodiment of the present application, when performing digital mapping of old buildings, the present application dynamically switches between master and slave devices when blind spots exist to fill in data for the blind spots of the old buildings from different perspectives. During data filling, because synchronized master and slave control terminals are always present, master-slave verification and master-slave synchronization are being implemented at all times, thereby achieving high-precision spatiotemporal synchronization compensation and digital visualization output. Because of spatiotemporal synchronization, there will be no mapping deviations caused by a single device or because the devices are not related when compensating for blind spots.

[0011] In conjunction with the first aspect, before performing blind spot detection on the spatiotemporal fusion data scanned by the UAV device and the ground robot, the method further includes:

[0012] A synchronous data transmission mechanism is pre-configured to receive the top point cloud data of the old building scanned by the drone equipment and the bottom point cloud data scanned by the ground robot; among them, the synchronous transmission mechanism includes a time-space synchronization mechanism, a master-slave dynamic synchronization mechanism and an interface perspective synchronization mechanism.

[0013] In an embodiment of the present application, for data synchronization before blind spot monitoring, the synchronous transmission mechanism configured in the present application can achieve three-layer synchronization in time and space acquisition, master-slave control and interface perspective, thereby ensuring data accuracy.

[0014] In combination with the first aspect, the spatiotemporal synchronization mechanism is configured with a unified coordinate system based on a joint calibration target; wherein, the ground robot is given real-time timing in a sliding window of the unified coordinate system through the drone equipment to generate spatiotemporal fusion data of the top point cloud and the bottom point cloud.

[0015] In an embodiment of the present application, the space-time synchronization mechanism of the present application is configured with a unified coordinate system. By setting a sliding window for space-time synchronization, real-time time data synchronization is performed. The synchronized data exists in a unified coordinate system and the same data window, so there will be no space-time deviation during fusion.

[0016] In combination with the first aspect, the interface perspective synchronization mechanism performs multi-perspective segmentation on the target spatiotemporal fusion data based on a sliding timeline according to the unified coordinate system of the spatiotemporal synchronization mechanism; wherein the segmented perspectives include drone views, ground robot views and fusion views, and the sliding timeline is used for perspective synchronization backtracking, and the curvature similarity of the target spatiotemporal fusion data point cloud after perspective synchronization backtracking is the same.

[0017] In an embodiment of the present application, the interface perspective synchronization mechanism can perform three-way segmentation of drones, ground robots and fused views during multi-perspective segmentation. Because there is a sliding timeline during the segmentation process, synchronous backtracking of the perspective can be achieved based on the sliding timeline. In the process of synchronous backtracking of the perspective, the similarity of the point cloud curvature is integrated, and the master-slave device scanning data can be fused with extremely low or even no deviation. It can also realize backtracking judgment of any event to determine whether there is a point cloud data error.

[0018] In combination with the first aspect, fusing the blind spot compensation data into the spatiotemporal fusion data includes:

[0019] Determine the overlapping area of the blind zone edge based on the master device and the slave device;

[0020] Extracting the overlapping boundary coordinates of the top point cloud data and the bottom point cloud data that are temporally and spatially aligned with the edge of the blind zone in the unified coordinate system;

[0021] The blind spot compensation data and the spatiotemporal fusion data of the blind spot edge overlapping area are bidirectionally projected according to the overlapping boundary coordinates to determine whether there is an anomaly in the compensation data. When the compensation data is anomaly, the texture feature parameters of the bidirectional projections are different.

[0022] In an embodiment of the present application, when the blind spot compensation data is fused into the spatiotemporal fusion data, that is, during the process of blind spot data compensation, the spatiotemporal alignment of the blind spot compensation data in the unified coordinate system will be performed through the overlapping area of the blind spot edge. During compensation, the blind spot data and the spatiotemporal fusion data can be projected bidirectionally to determine whether there are abnormalities in the edge overlapping area, thereby determining whether there are differences in texture features.

[0023] In combination with the first aspect, fusing the blind spot compensation data into the spatiotemporal fusion data further includes:

[0024] Construct an architectural feature library of old buildings and identify functional areas with different architectural features;

[0025] When there is a characteristic blind spot for which blind spot compensation data cannot be identified, the functional area in the building feature library is matched based on the target area where the characteristic blind spot is located, and active prediction compensation is performed when the match is consistent.

[0026] In an embodiment of the present application, when performing blind spot compensation, the present application will actively predict the blind spot compensation data, based on the active prediction of the blind spot functionality, thereby actively predicting the blind spot compensation data to achieve blind spot compensation.

[0027] In combination with the first aspect, after fusing the blind spot compensation data into the spatiotemporal fusion data, the method further includes:

[0028] Match the geometric topological features of the adjacent areas of the blind spot to generate a virtual point cloud with functional area shape rules based on architectural features;

[0029] The virtual point cloud is verified for curvature consistency with the point cloud data of the host device to determine whether the curvature error exceeds a preset threshold, and the virtual point cloud that does not exceed the preset threshold is used as the target data for blind spot compensation data.

[0030] In an embodiment of the present application, when compensating for blind spots, the present application will determine the curvature consistency based on the shape rules of the old building and the characteristics of the virtual point cloud. Through the curvature consistency, the matching shape features and the characteristics of the blind spots of the old building will be determined for compatibility, thereby determining the data for blind spot compensation.

[0031] In combination with the first aspect, when the curvature error does not exceed a preset threshold, the method includes:

[0032] The target area is divided into three sub-areas according to the curvature distribution and the differential error threshold is set;

[0033] Infrared thermal imaging and historical drawing data are integrated to generate candidate point clouds through graph neural networks;

[0034] Screen candidate data based on dynamic scoring of curvature continuity, texture consistency and structural compliance;

[0035] Generative adversarial network is used for iterative optimization until the discriminator confidence difference is less than the preset threshold.

[0036] In an embodiment of the present application, in order to minimize errors in the compensation process during blind spot compensation to address the curvature error problem, the present application uses an iterative optimization method of an adversarial network generator to calculate the confidence level to determine the most appropriate compensation data in the blind spot compensation data.

[0037] In conjunction with the first aspect, before performing blind spot detection on the spatiotemporal fusion data scanned by the UAV device and the ground robot, the method further includes:

[0038] The point cloud data collected by the drone is divided into spatial grids, the change rate of the point cloud density in each grid is detected, and only the grids with a change rate exceeding a threshold are incrementally encoded;

[0039] Extract high curvature feature points from the point cloud data collected by the ground robot as key frames, and give priority to encoding and transmitting key frame data;

[0040] Establish the spatiotemporal mapping relationship between the drone's incremental data blocks and the robot's key frames;

[0041] When data loss is detected for a certain device, related data is extracted from the data stream of another device based on the spatiotemporal mapping relationship for interpolation and completion.

[0042] In the embodiment of the present application, before blind spot monitoring, because the drone and ground robot devices are performing real-time scanning, the data collected by the drone device can be spatially gridded, and incremental encoding is achieved based on the point cloud density within the grid, that is, incremental data. For the data collected by the ground robot, key frame determination of high curvature feature points is required. The two are mapped through time and space. If the mapping is consistent, data compensation can be performed based on the time and space mapping relationship to determine if there is data loss in either the master or slave device.

[0043] Secondly, this application proposes a digital mapping system for old buildings based on air-ground collaborative point cloud fusion, including:

[0044] Blind spot dynamic master-slave control module: used to detect blind spots in the spatiotemporal fusion data scanned by UAV equipment and ground robots, and activate the dynamic master-slave mechanism when blind spots exist in the top point cloud data / bottom point cloud data;

[0045] Terminal loading module: used for the dynamic master-slave mechanism to use the device corresponding to the point cloud data with blind spots as the slave device, and the device corresponding to the point cloud data without blind spots as the host device, and control the slave device to perform point cloud filling, determine the blind spot compensation data, and fuse the blind spot compensation data into the spatiotemporal fusion data to generate the target spatiotemporal fusion data loaded into the visualization interface.

[0046] In an embodiment of the present application, when performing digital mapping of old buildings, the present application dynamically switches between master and slave devices when blind spots exist to fill in data for the old building blind spots from different perspectives. During data filling, because synchronized master and slave control terminals exist at all times, master-slave verification and master-slave synchronization are implemented at all times, thereby achieving high-precision spatiotemporal synchronization compensation and digital visualization output. Because of spatiotemporal synchronization, there will be no mapping deviations caused by a single device or because the devices are not related when compensating for blind spots.

[0047] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.

[0048] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0050] Figure 1 This is a method flow chart of a digital mapping method for old buildings based on air-ground collaborative point cloud fusion in an embodiment of the present invention;

[0051] Figure 2 Flowchart of an implementation of a synchronous data transmission mechanism according to an embodiment of the present invention;

[0052] Figure 3 1 is a flow chart of the implementation of the space-time synchronization mechanism in an embodiment of the present invention;

[0053] Figure 4 This is a flowchart of the implementation of the interface perspective synchronization mechanism in an embodiment of the present invention;

[0054] Figure 5 Detection execution diagram for compensating for data anomalies in an embodiment of the present invention;

[0055] Figure 6 This is a flowchart of the implementation of the feature library-driven compensation mechanism in an embodiment of the present invention;

[0056] Figure 7 This is an implementation diagram of the curvature error determination process in an embodiment of the present invention;

[0057] Figure 8 This is a process implementation diagram of a multi-source data fusion path in an embodiment of the present invention;

[0058] Figure 9 Implementation diagram of the data compensation mechanism in an embodiment of the present invention;

[0059] Figure 10 This is a structural diagram of the time-space mapping relationship in an embodiment of the present invention;

[0060] Figure 11 This is a system composition diagram of an old building digital mapping system based on air-ground collaborative point cloud fusion in an embodiment of the present invention. DETAILED DESCRIPTION

[0061] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0062] In terms of surveying and mapping of old buildings, traditional surveying and mapping of old buildings mainly relies on a single platform. For example, it only uses ground mobile equipment to scan in stages, and then manually judges the blind spots that appear during the scanning process. Then, the blind spot positions are mainly predicted based on manual experience, and then the scanning path is planned.

[0063] In blind spot compensation, missing data is filled using spatial interpolation algorithms from neighboring point clouds, or detailed information is supplemented through later manual modeling. Collection devices coordinate operations using a fixed master-slave model. The master device always performs core data collection tasks, such as a ground mobile device. Based on missing data, another ground mobile device or drone is manually assigned to assist in collection. Auxiliary slave devices only perform supplementary scans, which can easily lead to data gaps.

[0064] Traditional solutions, most of which rely on single-device scanning, are limited by the physical viewing angle and cannot eliminate inherent blind spots at the top and bottom. Manual judgment of blind spots is highly subjective and inefficient, easily leading to the omission of key component data. The compensated data generated by simple interpolation algorithms does not adequately match the original structural geometry, resulting in frequent texture fractures and surface distortion. The fixed master-slave model lacks dynamic responsiveness. When the master device encounters environmental interference, the system requires a long time to reconstruct the operation process, resulting in data gaps and a surge in time costs.

[0065] Traditional compensation methods primarily use rigid registration algorithms (such as ICP) to stitch multi-source data, rely on manually calibrated feature points to achieve coordinate system unity, and use offline batch processing to complete data fusion. Quality verification uses only curvature continuity as a single evaluation metric, and abnormal data is screened through manual visual inspection, lacking a systematic verification mechanism.

[0066] Furthermore, rigid registration algorithms cannot adapt to the morphological changes of irregular structures in older buildings, easily leading to error accumulation and coordinate shifts. The accuracy of manually calibrated feature points is significantly affected by the operator's experience, and differences in the density of heterogeneous data lead to significant transitional faults in the fusion interface. A single curvature metric cannot comprehensively assess texture alignment and structural compliance, and manual screening is inefficient and has a high rate of missed detections. Consequently, the lack of a closed-loop verification system makes it difficult to trace the causes of errors, and repairing abnormal data relies on repeated scanning, severely restricting surveying and mapping efficiency.

[0067] In view of this, in this application, by constructing an air-ground collaborative intelligent mapping system, the core defects of traditional technologies such as incomplete blind spot processing, low data fusion accuracy, and poor dynamic adaptability are solved.

[0068] This application is mainly based on a dynamic master-slave control mechanism. The surveying and mapping equipment autonomously switches the master-slave role according to the real-time blind spot detection results, realizing intelligent optimization of the scanning path and efficient allocation of resources, and preventing the data fault problem in the traditional fixed mode. This application combines the synchronous transmission of multi-source data and joint calibration technology to establish a unified spatiotemporal reference framework for the entire domain, eliminating the coordinate offset and motion blur caused by device heterogeneity, thereby improving the accuracy of multi-view data fusion. By introducing a compensation mechanism driven by architectural features, a virtual point cloud that conforms to the shape rules is generated through a pre-set component feature library and a deep learning model. This can meet the aesthetic characteristics of historical buildings while solving the geometric distortion and texture fracture problems caused by traditional interpolation compensation. In the closed-loop verification process, the curvature, texture, and structure multi-dimensional evaluation indicators are integrated, combined with the iterative optimization of the generative adversarial network, to achieve a full-process control link from data generation to quality feedback, and comprehensively guarantee the morphological compliance and feature consistency of the compensation data. The collaborative work of the real-time visualization engine and adaptive processing algorithm of this application can significantly shorten the processing delay from data collection to output of results, and then provide high-precision digital mapping through dynamic adjustment and on-site automatic or proactive decision-making in complex environments, solving the collection accuracy of blind spots.

[0069] Example 1:

[0070] The present application provides a method for digital mapping of old buildings based on air-ground collaborative point cloud fusion. Figure 1 , in the specific real-time process:

[0071] First, blind spot detection is performed on the spatiotemporal fusion data scanned by the UAV equipment and the ground robot. When there is a blind spot view in the top point cloud data / bottom point cloud data, a dynamic master-slave mechanism is activated.

[0072] The scanning equipment of the present application is a combination of ground equipment and drone equipment. The drone scanning 100 obtains the features of the outline and appearance of the old building by real-time aerial viewing of the old building, thereby realizing the first surveying and mapping. The drone scanning of the present application is not limited to high-altitude drone scanning. For the special texture of the old building, when the present application performs scanning, both the drone and the ground robot can be indoors and perform synchronous space-time scanning. The drone has the advantage of drone aerial scanning, and the ground robot has the advantage of ground robot scanning of areas that are inaccessible to the drone, such as gaps. The ground robot scanning 101 realizes synchronous space-time scanning by combining with the ground robot.

[0073] The first data of the drone scanning 100 and the second data of the ground robot scanning 101 are scanned by the old building in the same time domain and space domain based on the air-ground collaborative spatiotemporal data acquisition 103 in a spatiotemporal collaborative manner, and the two are adaptively integrated.

[0074] During this process, blind spot detection 104 of spatiotemporal fusion data is performed. The specific process is to compare the scanning coverage areas of the drone and the ground robot with different patrol scanning angles, and identify the top / bottom blind spot boundaries in the old building and bottom areas through point cloud density gradient analysis and curvature continuity detection algorithm. As for the area between the bottom and the top, the two scans are cross-referenced through data to realize feature collection in the middle area.

[0075] If a missing perspective is detected, that is, a blind spot exists, the dynamic master-slave switching protocol is triggered and the dynamic master-slave mechanism 105 is started; if no blind spot is detected, the collected data can directly generate spatiotemporal fusion data 111, that is, the data collected by the drone equipment and the data scanned by the ground robot are fused, the data of the two are spatiotemporally fused, duplicate data is eliminated, and the combined accurate mapping data is generated.

[0076] The following example illustrates the perspective loss compensation process: a Qing Dynasty hip-and-gable rooftop building features multiple layers of carved eaves and brackets. Drones are unable to scan the blind area beneath the eaves. While scanning the bottom data, a ground-based robot detected the blind area at the top, triggering a dynamic master-slave switch. The drone then becomes the slave, generating a spiral path along the blind area's boundary and performing high-density compensation scans. Fusion utilizes feature-constrained interpolation, matching the bracket-arch rule base to generate a virtual point cloud. Through the master-slave data verification mechanism, the curvature difference between the compensated data and the original structure is reduced to an imperceptible level.

[0077] During the master-slave device determination 106 process, the dynamic master-slave mechanism in the present application marks the device corresponding to the point cloud data with a blind spot as a slave device 108, and marks the device corresponding to the point cloud data without a blind spot as a host device 107, and controls the slave device to perform point cloud filling compensation 109, generates blind spot compensation data 110, and fuses the blind spot compensation data 112 with the spatiotemporal fusion data to generate target spatiotemporal fusion data 113 loaded into the visualization interface 114.

[0078] During implementation, the point cloud data from the current device without blind spots is used as the master reference system, and devices with blind spot data in the master reference system are converted to slave devices. The master device maintains the baseline data stream, and the slave device generates a compensation path based on the master data characteristics, such as adjusting the laser scanning angle and increasing the scan overlap rate.

[0079] During blind spot compensation, a two-way projection verification method is used, such as orthogonal projection combined with perspective projection. The compensated data and the data collected by the main device are placed in the same coordinate system for texture feature matching and curvature continuity verification to ensure geometric consistency of the fusion interface, such as the correct texture alignment of carved components in ancient buildings. This is because sudden movement of personnel on site can interrupt drone scanning, creating dynamic blind spots.

[0080] The device determination 106 mechanism in the master-slave controller switches the ground robot to the master in 0.5 seconds or even shorter, generates a predicted point cloud based on historical scanning data, and after the UAV is restored, quickly aligns the data streams before and after the interruption through the space-time mapping table, and uses the sliding window verification method to automatically repair the discontinuous point cloud in the time domain.

[0081] In this process, multi-dimensional verification indicators are automatically introduced, such as curvature, texture, structure, and automatic rejection rate of abnormal compensation data to avoid secondary errors caused by human intervention.

[0082] Example 2:

[0083] Before performing blind spot detection on the spatiotemporal fusion data scanned by drone equipment and ground robots, this application will also set up a synchronization data mechanism, see Figure 2 :

[0084] When this application is implemented, during the process of air-ground collaborative spatiotemporal data collection 103, a pre-configured synchronous data transmission mechanism will be used to receive the top point cloud data of the old building scanned by the drone equipment and the bottom point cloud data scanned by the ground robot; wherein the synchronous transmission mechanism includes a spatiotemporal synchronization mechanism, a master-slave dynamic synchronization mechanism and an interface perspective synchronization mechanism.

[0085] The synchronous data transmission mechanism configuration 1031 is to achieve hierarchical synchronous transmission of the top point cloud data of the drone scanning 100 and the bottom point cloud data of the ground robot scanning 101, generate a highly consistent data basis during blind spot detection, fuse the point clouds of the top data and the bottom data, and establish a unified coordinate system based on physical calibration objects in the process of spatiotemporal synchronous transmission to achieve coordinate system alignment. By jointly calibrating the spatiotemporal parameters of the sensors of the drone and the ground robot, timestamp synchronization is achieved, and the posture deviation between devices is eliminated, mainly to monitor the difference in accuracy and achieve mutual matching of scanning frequencies.

[0086] Hardware-level clock synchronization is used in the time dimension to ensure strict alignment of scan timing. During the master-slave device switching process, the data dependency between devices is updated in real time. In other words, the master-slave synchronization mechanism 1051 maintains the stability of the collaborative link through heartbeat detection and status broadcast synchronization.

[0087] The master device continuously pushes a baseline data stream, while the slave devices adaptively adjust the scanning frequency and data format. Multi-source data is mapped into a unified visualization space. Dynamic binding of viewing angle parameters automatically corrects the perspective relationship between the drone's downward-looking view and the ground's upward-looking view, ensuring spatial consistency within the user interface.

[0088] For the spatiotemporal synchronization of joint scanning, a spatiotemporal synchronization mechanism is deployed: infrared calibration targets are set at the four corners of the building. After the UAV and ground robot are turned on, they automatically complete the joint calibration, and the master and slave are dynamically synchronized.

[0089] The ground robot recognizes the top scanning requirement and actively requests to switch to the slave role to receive the scanning path sent by the drone.

[0090] Interface perspective synchronization display: Based on the interface perspective synchronization mechanism 1032, the control operating platform renders the fusion view of the drone's bird's-eye view point cloud and the ground's upward view point cloud in real time, automatically hiding overlapping redundant data. In this process, the scanning perspectives of the drone and the ground robot are synchronized, and then mapped on a unified coordinate system to generate a visual coordinate system. Rendering technology is used for real-time synchronous rendering, highlighting the current overall scanning and mapping of the old building, and improving the clarity and accuracy of the scan.

[0091] Example 3:

[0092] The spatiotemporal synchronization mechanism of this application is configured with a unified coordinate system based on a joint calibration target. Figure 3 Among them, the UAV equipment provides real-time timing to the ground robot under the sliding window of the unified coordinate system, generating spatiotemporal fusion data of the top point cloud and the bottom point cloud.

[0093] This application deploys uniquely coded physical targets, such as QR codes / reflective balls, at key locations on a building. UAVs and ground robots use multiple sensors, such as lidar and visual monitoring, to form a joint recognition target, establish a unified spatial reference across devices, and form a unified coordinate system 1042. The joint recognition target is based on the world coordinate system and measures from three angles: horizontally, vertically, and on the main facade of the building, to ensure the accuracy of the surveying and mapping data during the measurement process.

[0094] The target's three-dimensional coordinates are absolute reference coordinates, eliminating the cumulative error of the device's local coordinate system. The drone scanning system 1001 and the ground robot 1011 form the scanning system. They use the BeiDou satellite's PPS signal as a real-time authorization reference signal, determine the spatiotemporal reference node, and broadcast timing signals to the ground robot and drone's lidars to align timestamps.

[0095] That is, the ground equipment dynamically adjusts the scanning timing within the sliding time window to ensure that the timestamps of the top and bottom point clouds are strictly aligned.

[0096] The window size is adaptively adjusted based on environmental complexity, balancing real-time performance and accuracy requirements. Within a unified coordinate system, the drone's pitch scan data and the ground robot's elevation data are spatially interpolated and fused in a time series to generate a continuous 3D point cloud model. The fusion process preserves the characteristics of the device's original data, avoiding information loss.

[0097] Example 4:

[0098] The interface perspective synchronization mechanism 1032 of the present application performs multi-perspective segmentation of the target spatiotemporal fusion data based on a sliding time axis according to the unified coordinate system of the spatiotemporal synchronization mechanism; Figure 4 The perspectives divided by the multi-perspective segmentation module 10321 include drone views, ground robot views, and fusion views. In this process, the sliding timeline control 10322 is used to perform perspective synchronization backtracking. After the perspective synchronization backtracking 10323, the curvature similarity of the target spatiotemporal fusion data point cloud is the same. Through the verification method of curvature similarity, the drone views and ground robot views with different curvatures can also be fine-tuned based on the timeline, and controlled during multi-perspective segmentation.

[0099] During implementation, based on the time-space mapping relationship of a unified coordinate system, the fused data is segmented according to the acquisition perspective into a drone-based bird's-eye view, a ground-based view, and a fused view. Each perspective retains the original data characteristics while establishing perspective associations. Through a timestamp-driven data slicing approach, this process allows for synchronous playback of perspectives from any time period, ensuring strict alignment of the spatiotemporal evolution of the point cloud from different perspectives. During the backtracking process, the curvature distribution similarity of the multi-perspective point clouds is calculated in real time. When local differences are detected, a curvature compensation algorithm is automatically triggered to eliminate morphological deviations caused by differences in device perspectives.

[0100] In actual implementation:

[0101] Segment the drone's top view in a unified coordinate system, for example, to show the roof's folding outline and the beam structure presented by the ground robot's side view;

[0102] Slide the timeline to review data from key construction stages and compare structural deformation trends under different processes;

[0103] Automatically align the curvature distribution of top-view and side-view data to generate a continuous transition 3D roof model;

[0104] The technical effects that can be achieved are: the degree of restoration of the folding curves is significantly improved, the spatial relationship of the mortise and tenon joints of the wooden components is clearly discernible, and accurate mapping and display of old buildings are achieved.

[0105] Example 5:

[0106] This application integrates blind spot compensation data into spatiotemporal fusion data to establish an intelligent perception and compensation verification mechanism for blind spot edges. Figure 5 ,include:

[0107] This application determines the overlapping area of the blind zone edge based on the master device and the slave device;

[0108] Then, the overlapping boundary coordinates of the top point cloud data and the bottom point cloud data that are temporally and spatially aligned with the edge of the blind zone in the unified coordinate system are extracted;

[0109] Finally, the blind spot compensation data and the spatiotemporal fusion data of the blind spot edge overlapping area are bidirectionally projected according to the overlapping boundary coordinates to determine whether there is any compensation data anomaly. When the compensation data is abnormal, the texture feature parameters of the bidirectional projection are different.

[0110] This application is based on the spatial relative posture and scanning path historical data of the master and slave devices. Through feature point density gradient analysis, it dynamically identifies the transition edges between blind spots and effective areas. The boundary recognition process integrates the device motion trajectory prediction to avoid misjudgment caused by dynamic changes in the environment.

[0111] In this application, in a unified coordinate system, namely unified coordinate system 1042, an improved iterative closest point algorithm is used to align the edge point clouds of the master and slave devices, namely the edge point clouds of the drone device scanning data and the edge point clouds of the ground robot scanning data in this application, and then extract a set of boundary feature points with spatiotemporal continuity. Subsequently, stable overlapping coordinates are selected through curvature similarity constraints as the reference anchor points for data fusion.

[0112] The compensation data is projected from the master device's perspective (e.g., orthogonal projection) and the slave device's perspective (e.g., perspective projection) to achieve bidirectional mapping. Multi-scale texture feature comparisons, such as those based on SIFT keypoint distribution and local binary pattern (LBP), are used to determine if abnormal areas exist. When the difference in texture features between the bidirectional projections exceeds a tolerance threshold, the compensation data regeneration process is triggered.

[0113] In the actual implementation process, for example: the mortise and tenon joints of Song Dynasty temples are blocked by the top beams, forming a scanning blind spot, and the traditional compensation data does not match the original structural texture. The host of this application, that is, the ground robot, scans along the column base to identify the complete outline of the bottom of the mortise and tenon; the slave, that is, the drone, performs the compensation scan. This application extracts the characteristic points of the brackets at the intersection of the beams and columns as the overlapping boundary, and aligns the spatial posture of the mortise and tenon of the master and slave device point clouds. The two-way projection verification of this application finds that the texture direction of the compensation data is abnormal. The texture direction abnormality includes the difference between the traditional carving texture and the modern processing traces generated by the compensation; triggering the compensation regeneration based on the historical form library to generate a virtual point cloud of the mortise and tenon that conforms to the "Construction Code" of the Song Dynasty. The texture of the repaired mortise and tenon nodes is coherent, and the historical craft characteristics are fully preserved.

[0114] Example 6:

[0115] This application integrates blind spot compensation data into spatiotemporal fusion data, and builds a collaborative system of historical building feature knowledge base and intelligent compensation to construct blind spot restoration based on cultural heritage drive. Figure 6 :

[0116] This application first constructs a database of architectural features 200 for old buildings and identifies functional areas associated with different architectural features. Based on historical documents, construction methods, and existing surveying and mapping results, a multi-dimensional feature library encompassing component form, material characteristics, and craftsmanship techniques is established. Each functional area, such as brackets, tile ends, and carvings, is associated with morphological rules, such as mortise and tenon joint proportions and pattern symmetry; physical properties, such as wood grain direction and stone weathering patterns; and cultural semantics, such as official rank and regional style. These functional areas can be functionally matched with data from feature-directed blind areas. If the functional match is successful, texture features in the blind areas can be proactively predicted and compensated 203.

[0117] That is, when there is a characteristic blind spot for which blind spot compensation data cannot be identified, the functional area in the building feature library is matched based on the target area where the characteristic blind spot is located, and active prediction compensation is performed when the match is consistent.

[0118] During this process, point cloud semantic segmentation technology is used to extract residual features around the blind spot, such as residual mortises on brackets and broken edges on tiles. Spatial topological relationships, such as eaves position and beam hierarchy, are then combined to infer the functional area category to which the blind spot belongs. When the blind spot features match the feature library, the generation rules for the corresponding area are invoked, such as the proportional constraints on brackets in official Qing Dynasty architecture. This is combined with the attention-based point cloud completion network within the deep learning model to generate virtual data that conforms to historical authenticity. Blind spot type monitoring 104201 transmits characteristic data of the blind spot to the building feature library for matching. This primarily involves multi-source data surrounding the blind spot, namely scanned point cloud data. By extracting contour features from this data, the blind spot type is identified. Blind spot types include characteristic blind spots, enabling active prediction and compensation. There are also conventional blind spots, that is, blind spots that cannot be identified through features, which are specific blind spots that have never appeared before or temporary blind spots caused by interference. They are compensated through a standard compensation process. After the compensation data is verified, compensation is performed if the verification is passed to achieve spatiotemporal data fusion and visualize the fused data.

[0119] Example 7:

[0120] This application integrates blind spot compensation data into spatiotemporal fusion data and also builds an intelligent compensation verification system under shape constraints. Figure 7 :

[0121] This application uses collected point cloud data to construct host point cloud data. The application then extracts geometric features or form features of adjacent areas of the blind area through feature matching based on the geometric topological features of adjacent areas. Combined with the form rule library 3011, a virtual point cloud with form rules for functional areas based on architectural features is generated. In this process, based on the component topological rules of the architectural feature library 200, such as the spatial relationships between the warping, raising, and shuatou of brackets, geometric features of adjacent areas of the blind area, such as the angles of mortise and tenon joints and the direction of carved textures, are extracted to form the form rule library 3011, i.e., a rule library for different architectural form features in old buildings. This generates a virtual point cloud that conforms to historical forms. This process is executed using a virtual point cloud generator 3012. The generation process incorporates parameters of traditional craftsmanship techniques, such as the roof curvature gradient of the lifting and folding method, to ensure the historical authenticity of the virtual data.

[0122] At the fusion interface of virtual point cloud and real data, the joint analysis method of Gaussian curvature and mean curvature is used to detect the continuity and naturalness of the surface transition, and the micro-curvature jump area is identified through the sliding window scanning mechanism.

[0123] The virtual point cloud is then verified for curvature consistency 3021 with the point cloud data from the host device to determine whether the curvature error exceeds a preset threshold. This is known as error threshold determination 3021. Virtual point clouds that do not exceed the preset threshold are used as target data for blind spot compensation. If the curvature difference exceeds the threshold, a feature reconstruction algorithm based on form rules is triggered to adjust the virtual point cloud generation parameters, such as the bracket projection ratio and tile spacing, until the curvature consistency verification is passed.

[0124] Example 8:

[0125] When the curvature error of this application does not exceed the preset threshold, a multimodal data-driven intelligent optimization system will be constructed to achieve refined control of cultural heritage restoration. Figure 8 :

[0126] First, the target area is divided into three sub-areas based on curvature distribution and a differentiated error threshold is set. Data fusion is performed when the curvature error does not exceed the threshold. For example, 400 data fusion and generation generate candidate point clouds. For example, when the curvature of the target area does not exceed the threshold, curvature sensitivity levels are divided based on the functional properties of building components, such as load-bearing structure, decorative components, and texture style, and differentiated verification standards are set.

[0127] Highly sensitive areas, such as bracket nodes, use micro-curvature continuity detection;

[0128] Low-sensitivity areas, such as walls, focus on macroscopic morphology matching.

[0129] Then, infrared thermal imaging and historical drawing data are fused to generate a candidate point cloud set through a graph neural network. The internal structural information revealed by infrared thermal imaging, such as the decayed cavities in wood, is combined with the shape rules recorded in historical drawings. A cross-temporal and spatial feature association model is established through a graph neural network to generate candidate data that takes into account both historical authenticity and structural safety.

[0130] Then, the dynamic optimization system 401 built into the server that controls the host and slave machines of this application implements scoring sorting, and dynamically screens candidate data based on curvature continuity, texture consistency and structural compliance; constructs a three-dimensional scoring matrix of curvature, texture and structure, and dynamically adjusts the indicator weights based on the attention mechanism. For example: the decorative area focuses on texture, and the load-bearing area focuses on structure, and the optimal candidate set is screened. The optimal candidate set is generally compensable data with high consistency and a score of more than 85 on a percentage basis.

[0131] Finally, the generative adversarial network is iteratively optimized 402 until the discriminator confidence difference is less than a preset threshold. A dual-discriminator architecture, for example, a combination of a form discriminator and an engineering discriminator, is employed to iteratively optimize the generated data until both historical style fidelity and engineering reliability are achieved.

[0132] Example 9:

[0133] Before performing blind spot detection on the spatiotemporal fusion data scanned by drone equipment and ground robots, this application also constructs an intelligent hierarchical data processing system to perform drone preprocessing and ground robot preprocessing on the scanned data, and achieves efficient data collaboration in complex scenarios. Figure 9 and Figure 10 :

[0134] During the drone preprocessing 1001 process, the point cloud data collected by the drone is divided into spatial grids, and the point cloud density change rate in each grid is detected. Only the grids with a change rate exceeding the threshold are incrementally encoded; the drone point cloud is divided into spatial grids, and active areas of building deformation are identified based on the density change rate, such as the eaves bending area and the crack expansion area. Only the density change rate in the spatial grid is judged, and incremental encoding is implemented for areas with significant changes to reduce redundant data transmission.

[0135] During the ground robot preprocessing 1003, high curvature feature points are extracted from the point cloud data collected by the ground robot as key frames, and the key frame data is preferentially encoded and transmitted. For the ground robot data, high curvature points reflecting the key features of the building are extracted, such as carved turning edges and mortise and tenon joints, and a key frame priority transmission channel is established to ensure the integrity and real-time performance of the core feature data.

[0136] Finally, a spatiotemporal mapping relationship 1002 between the drone's incremental data blocks and the robot's key frames is established; through the bidirectional binding of timestamps and spatial coordinates, an association relationship between the drone's incremental data blocks and the robot's key frames is constructed, forming a resilient network in which data between devices are backed up for each other.

[0137] When data loss is detected for a device, related data is extracted from another device's data stream based on the spatiotemporal mapping relationship for interpolation and completion. When a single device's data stream is abnormal, the related data segment is quickly located based on the spatiotemporal mapping table, and a feature-driven interpolation algorithm is used to generate temporary compensation data to maintain the continuity of the surveying and mapping process.

[0138] Example 10:

[0139] This application proposes a digital mapping system for old buildings based on air-ground collaborative point cloud fusion. Figure 11 include:

[0140] Blind Spot Dynamic Master-Slave Control Module: This module detects blind spots in spatiotemporal fusion data from drone and ground robot scans. It activates a dynamic master-slave mechanism when blind spots exist in the top / bottom point cloud data. Based on real-time point cloud density and feature continuity analysis, it dynamically determines the master and slave device roles. The master device maintains a high-confidence data stream, while the slave device generates a compensated scanning path based on the master data's features, creating a bidirectional verification relationship from baseline to compensation.

[0141] Terminal loading module: Used in a dynamic master-slave mechanism, the device corresponding to the point cloud data with blind spots is designated as a slave device, and the device corresponding to the point cloud data without blind spots is designated as the master device. The slave device is controlled to perform point cloud filling, determine blind spot compensation data, and fuse the blind spot compensation data with the spatiotemporal fusion data to generate the target spatiotemporal fusion data that is loaded into the visualization interface. Guided by the master data, the slave device performs progressive scanning along the blind spot boundary and generates a compensated point cloud based on shape rule constraints. The compensated data is verified through bidirectional projection and, after forward / reverse projection, is fused with the master data under a unified spatiotemporal reference. The visualization interface renders the fusion results in real time and supports manual annotation of key areas of concern, such as vulnerable parts of cultural relics. The system dynamically adjusts the compensation strategy accordingly to achieve localized encrypted scanning.

[0142] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A digital mapping method for old buildings based on air-ground collaborative point cloud fusion, characterized by: include: Perform blind spot detection on the spatiotemporal fusion data scanned by UAV equipment and ground robots, and activate the dynamic master-slave mechanism when there is a blind spot view in the top point cloud data / bottom point cloud data; Among them, the dynamic master-slave mechanism regards the device corresponding to the point cloud data with blind spots as the slave device, and the device corresponding to the point cloud data without blind spots as the host device, and controls the slave device to perform point cloud filling, determine the blind spot compensation data, and fuse the blind spot compensation data into the spatiotemporal fusion data to generate the target spatiotemporal fusion data loaded into the visualization interface.

2. The digital mapping method for old buildings based on air-ground collaborative point cloud fusion according to claim 1 is characterized in that: Before performing blind spot detection on the spatiotemporal fusion data scanned by the UAV equipment and the ground robot, the method further includes: A synchronous data transmission mechanism is pre-configured to receive the top point cloud data of the old building scanned by the drone equipment and the bottom point cloud data scanned by the ground robot; among them, the synchronous transmission mechanism includes a time-space synchronization mechanism, a master-slave dynamic synchronization mechanism and an interface perspective synchronization mechanism.

3. The digital mapping method for old buildings based on air-ground collaborative point cloud fusion according to claim 2 is characterized in that: The spatiotemporal synchronization mechanism is configured with a unified coordinate system based on a joint calibration target; wherein, the UAV equipment provides real-time timing to the ground robot under the sliding window of the unified coordinate system to generate spatiotemporal fusion data of the top point cloud and the bottom point cloud.

4. The digital mapping method for old buildings based on air-ground collaborative point cloud fusion according to claim 3 is characterized in that: The interface perspective synchronization mechanism performs multi-perspective segmentation of the target spatiotemporal fusion data based on a sliding timeline according to the unified coordinate system of the spatiotemporal synchronization mechanism; wherein the segmented perspectives include drone views, ground robot views, and fusion views, and the sliding timeline is used for perspective synchronization backtracking. After the perspective synchronization backtracking, the curvature similarity of the target spatiotemporal fusion data point clouds is the same.

5. The digital mapping method for old buildings based on air-ground collaborative point cloud fusion according to claim 3 is characterized in that: The step of fusing the blind spot compensation data into the spatiotemporal fusion data includes: Determine the overlapping area of the blind zone edge based on the master device and the slave device; Extracting the overlapping boundary coordinates of the top point cloud data and the bottom point cloud data that are temporally and spatially aligned with the edge of the blind zone in the unified coordinate system; The blind spot compensation data and the spatiotemporal fusion data of the blind spot edge overlapping area are bidirectionally projected according to the overlapping boundary coordinates to determine whether there is an anomaly in the compensation data. When the compensation data is anomaly, the texture feature parameters of the bidirectional projections are different.

6. The digital mapping method for old buildings based on air-ground collaborative point cloud fusion according to claim 1 is characterized in that: The fusing of the blind spot compensation data into the spatiotemporal fusion data further includes: Construct an architectural feature library of old buildings and identify functional areas with different architectural features; When there is a characteristic blind spot for which blind spot compensation data cannot be identified, the functional area in the building feature library is matched based on the target area where the characteristic blind spot is located, and active prediction compensation is performed when the match is consistent.

7. The digital mapping method for old buildings based on air-ground collaborative point cloud fusion according to claim 6 is characterized in that: After fusing the blind spot compensation data into the spatiotemporal fusion data, the method further includes: Match the geometric topological features of the adjacent areas of the blind spot to generate a virtual point cloud with functional area shape rules based on architectural features; The virtual point cloud is verified for curvature consistency with the point cloud data of the host device to determine whether the curvature error exceeds a preset threshold, and the virtual point cloud that does not exceed the preset threshold is used as the target data for blind spot compensation data.

8. The digital mapping method for old buildings based on air-ground collaborative point cloud fusion according to claim 7 is characterized in that: When the curvature error does not exceed a preset threshold, it includes: The target area is divided into three sub-areas according to the curvature distribution and the differential error threshold is set; Fusing infrared thermal imaging with historical drawing data, a candidate point cloud set is generated through a graph neural network; Screen candidate data based on dynamic scoring of curvature continuity, texture consistency and structural compliance; Generative adversarial network is used for iterative optimization until the discriminator confidence difference is less than the preset threshold.

9. The digital mapping method for old buildings based on air-ground collaborative point cloud fusion according to claim 1, characterized in that: Before performing blind spot detection on the spatiotemporal fusion data scanned by the UAV equipment and the ground robot, the method further includes: The point cloud data collected by the drone is divided into spatial grids, the change rate of the point cloud density in each grid is detected, and only the grids with a change rate exceeding a threshold are incrementally encoded; Extract high curvature feature points from the point cloud data collected by the ground robot as key frames, and give priority to encoding and transmitting key frame data; Establish the spatiotemporal mapping relationship between the drone's incremental data blocks and the robot's key frames; When data loss is detected for a certain device, related data is extracted from the data stream of another device based on the spatiotemporal mapping relationship for interpolation and completion.

10. A digital mapping system for old buildings based on air-ground collaborative point cloud fusion, characterized by: include: Blind spot dynamic master-slave control module: used to detect blind spots in the spatiotemporal fusion data scanned by UAV equipment and ground robots, and activate the dynamic master-slave mechanism when blind spots exist in the top point cloud data / bottom point cloud data; Terminal loading module: used for the dynamic master-slave mechanism to use the device corresponding to the point cloud data with blind spots as the slave device, and the device corresponding to the point cloud data without blind spots as the host device, and control the slave device to perform point cloud filling, determine the blind spot compensation data, and fuse the blind spot compensation data into the spatiotemporal fusion data to generate the target spatiotemporal fusion data loaded into the visualization interface.

Images