Channel tree barrier real-time modeling system and method based on point cloud data

Abstract

This invention relates to the field of data modeling technology and discloses a real-time channel tree obstacle modeling system and method based on point cloud data. The method includes: partitioning the channel for inspection based on a 3D channel model and partitioning rules; configuring a first channel tree obstacle inspection strategy and a second channel tree obstacle inspection strategy to obtain first inspection data and second inspection data; updating the channel 3D model based on channel service security requirement data, historical standard inspection datasets, and standard inspection datasets; determining the target area based on the updated channel 3D model and inspection upgrade rules, and performing a secondary inspection on the target area. This solves the problem that related channel tree obstacle inspection modeling technologies are difficult to adapt to the high-precision, real-time, and refined tree obstacle inspection and modeling maintenance requirements of large-scale linear infrastructure channels.

Description

Technical Field

[0001] This invention relates to the field of data modeling technology, specifically to a real-time modeling system and method for channel tree barriers based on point cloud data. Background Technology

[0002] With the continuous expansion of linear infrastructure networks such as power transmission channels, rail transit, and oil and gas transmission corridors in China, tree obstructions caused by vegetation growth along these channels have become a major source of risk affecting the safe and stable operation of infrastructure. Vegetation grows continuously with the seasons, easily encroaching on the safety clearance of facilities, causing safety accidents such as short circuits, power outages, and facility scraping. Regular inspections and 3D modeling of tree obstructions along these channels are a core requirement for operation and maintenance management. The use of UAV-borne LiDAR and aerial photography equipment for channel inspections has gradually replaced outdated methods such as traditional manual foot inspections and simple visual estimations.

[0003] In practical engineering operation and maintenance scenarios, linear corridors are characterized by long spans and complex and variable environments along their routes. Uneven distribution of vegetation density and spatial distances between trees and facilities naturally leads to varying tree obstacle risks in different sections. Existing routine inspection operations often utilize fixed routes and standardized configurations for comprehensive data collection. While this operational model is regular and easy to deploy in batches, it struggles to balance the precision required for key sections with the efficiency of operations in ordinary sections when faced with the varying characteristics of local environments within the corridor. This often results in insufficient detailed representation of key areas and redundant resource consumption in routine areas, indicating room for improvement in refined operation and maintenance management. Multi-UAV collaborative parallel inspections are already a common method for large-scale corridor operations. However, the objective differences in payload parameters, sensing accuracy, and flight attitude benchmarks among different devices result in significant heterogeneity in the generated point clouds, images, and positioning attitude data. Existing general data processing methods can complete basic noise reduction and simple stitching integration, but when faced with heterogeneous inspection data from multiple models and batches, there are still engineering pain points in areas such as unified spatial benchmarks, refined noise filtering, and smooth integration of fragmented data. This can easily lead to issues like misaligned local points, unnatural data transitions, and loss of local details, affecting the overall consistency and accuracy stability of subsequent 3D model reconstruction. At the 3D model operation and maintenance iteration level, the industry currently mostly adopts a periodic full-scale reconstruction approach, re-collecting data and updating the overall model at fixed time intervals. This operation and maintenance mode is rather static and lagging. It is difficult to accurately capture the subtle morphological changes and spatial distance evolution caused by slow vegetation growth, and it is also difficult to achieve synchronous and linked updates of model entity parameters and hazard attribute annotations. Furthermore, existing operation and maintenance processes mostly remain at the level of independent operations for single inspections and single analyses, lacking a closed-loop operation and maintenance system encompassing inspection discovery, model updates, adaptive strategy adjustments, and secondary verification. They rely heavily on manual experience to judge subsequent inspection plans, and the intelligent, dynamic, and adaptive operation and maintenance capabilities need further improvement.

[0004] In summary, the relevant tree obstacle inspection and modeling technologies still have significant shortcomings in terms of the rationality of inspection area planning, adaptability of data collection operation modes, accuracy of multi-source data processing, timeliness of updating the three-dimensional model of the channel, and the continuity and dynamic adjustment capability of the inspection and maintenance process. They are difficult to adapt to the high-precision, real-time and refined tree obstacle inspection and modeling maintenance needs of large-scale linear infrastructure channels. Summary of the Invention

[0005] In view of this, the present invention provides a real-time tree obstacle modeling system and method based on point cloud data to solve the problem that related tree obstacle inspection and modeling technologies still have significant shortcomings in terms of the rationality of inspection area planning, adaptability of data acquisition operation mode, accuracy of multi-source data acquisition processing, timeliness of channel 3D model update, and continuity and dynamic adjustment capability of inspection and maintenance process. They are difficult to adapt to the high-precision, real-time and refined tree obstacle inspection and modeling maintenance needs of large-scale linear infrastructure channels.

[0006] In a first aspect, the present invention provides a method for real-time modeling of channel tree barriers based on point cloud data. The method includes: partitioning the inspection channel based on a three-dimensional channel model and partitioning rules to obtain a first region set and a second region set; configuring a first channel tree barrier inspection strategy and a second channel tree barrier inspection strategy for the first region set and the second region set respectively, wherein the first channel tree barrier inspection strategy is an encrypted acquisition strategy for the first region set, and the second channel tree barrier inspection strategy is a standard acquisition strategy for the second region set; executing the first channel tree barrier inspection strategy and the second channel tree barrier inspection strategy to obtain first inspection data and second inspection data; performing cross-machine registration, adaptive point cloud denoising, multi-machine point cloud incremental fusion, and data format standardization processing on the first inspection data and the second inspection data to generate a standard inspection dataset; acquiring a historical standard inspection dataset; updating the channel three-dimensional model based on channel business security requirement data, the historical standard inspection dataset, and the standard inspection dataset; determining a target region based on the updated channel three-dimensional model and inspection upgrade rules; updating the tree barrier inspection strategy of the target region from the second channel tree barrier strategy to the first channel tree barrier strategy; and performing a second inspection on the target region.

[0007] The real-time channel tree obstacle modeling method based on point cloud data provided in this embodiment firstly divides the inspection channel into regions based on the channel's 3D model and partitioning rules, and matches corresponding tree obstacle inspection strategies to different region sets, improving the precision of inspection region division, optimizing the matching logic of inspection tasks, and enhancing the adaptability of inspection operations to the actual channel scenario. Secondly, by executing the corresponding configured channel tree obstacle inspection strategies and acquiring corresponding inspection data, the quality of inspection data acquisition under different scenarios is improved, the completeness of scene information in key areas is enhanced, and the basic acquisition level of regular areas is maintained stably, optimizing the overall data acquisition balance. Then, by sequentially performing cross-machine registration, adaptive denoising, multi-machine point cloud incremental fusion, and data format standardization processing on multiple types of inspection data to form a standard inspection dataset, the baseline deviation between multi-source heterogeneous data is reduced, the interference of irrelevant noise on effective point cloud information is weakened, the fusion and normalization capability of multiple batches of inspection data is improved, and the compatibility and adaptability between various types of inspection data are enhanced. By introducing historical standard inspection datasets and combining them with current standard inspection datasets to update the overall 3D model of the passageway, the sensitivity of the 3D model's status synchronization is improved, enhancing the model's ability to perceive changes in vegetation and facility morphology, and improving the synchronization level of model morphological parameters and spatial attributes. Then, based on the updated 3D model and inspection upgrade rules, target areas are identified, the original inspection strategy is adapted and adjusted, and secondary inspections are conducted in the target areas. This increases the depth of verification in sensitive areas, reduces the probability of missed detections of tree obstructions, enhances the closed-loop linkage capability of the inspection and maintenance process, improves the flexibility of dynamic strategy adaptation, and strengthens the stability of long-term safe operation and maintenance of the passageway. Implementing this solution addresses the significant shortcomings of existing tree obstruction inspection and modeling technologies in terms of the rationality of inspection area planning, adaptability of data collection modes, accuracy of multi-source data processing, timeliness of 3D model updates, and the continuity and dynamic adjustment capability of the inspection and maintenance process. These shortcomings make it difficult to meet the high-precision, real-time, and refined tree obstruction inspection and modeling maintenance needs of large-scale linear infrastructure passageways.

[0008] In one optional implementation, the 3D model of the passage includes infrastructure data, 3D tree data, and hazard labeling data attached to the 3D tree data. The hazard labeling data includes the location, height, crown width, distance from passage facilities, growth trend prediction results, and risk level of each tree. Based on the 3D model of the passage and zoning rules, the inspection passage is divided into a first set of regions and a second set of regions. This includes: parsing the hazard labeling data in the 3D model of the passage and marking trees with a risk level higher than a preset level as target trees; dividing the inspection passage into several inspection areas based on the coverage constraints of the UAV inspection operation, the location, height, crown width, distance from passage facilities, and growth trend prediction results of each target tree, with overlapping areas between adjacent inspection areas; and, based on zoning rules, assigning inspection areas containing target trees to the first set of regions and assigning inspection areas not containing target trees to the second set of regions.

[0009] In one optional implementation, a first-channel tree obstacle inspection strategy and a second-channel tree obstacle inspection strategy are configured for the first set of regions and the second set of regions, respectively. This includes: configuring encrypted data acquisition parameters and corresponding inspection drones for each inspection area in the first set of regions based on the tree location, height, crown width, distance from the passage facilities, and growth trend prediction results of each target tree, thereby generating a first-channel tree obstacle inspection strategy. The encrypted data acquisition parameters include flight altitude, flight path overlap, point cloud sampling frequency, and number of acquisition viewpoints. For each inspection area in the second set of regions, uniform conventional data acquisition parameters and corresponding inspection drones are configured to generate a second-channel tree obstacle inspection strategy. The conventional data acquisition parameters include flight altitude, flight path overlap, point cloud sampling frequency, and number of acquisition viewpoints. The point cloud sampling frequency and number of acquisition viewpoints in the conventional data acquisition parameters are lower than those in the encrypted data acquisition parameters.

[0010] In one optional implementation, both the first and second inspection data include 3D point clouds of passage facilities, 3D point clouds of vegetation, aerial imagery data, GNSS positioning data, data acquisition time sequence timestamps and corresponding acquisition operation parameters, IMU attitude data, raw LiDAR scan data, and UAV model identifiers. The first and second inspection data undergo cross-machine registration, adaptive point cloud denoising, multi-machine point cloud incremental fusion, and data format standardization to generate a standard inspection dataset. This includes: using a registration algorithm based on the GNSS positioning data and IMU attitude data of each inspection UAV, the first inspection data... The first and second inspection data are uniformly registered to the channel's global coordinate system; the denoising thresholds corresponding to each inspection drone are obtained, and density clustering algorithms are used to denoise the registered first and second inspection data based on the denoising thresholds corresponding to each inspection drone. The denoising thresholds corresponding to each inspection drone are set based on the point cloud acquisition accuracy; the denoised first and second inspection data are incrementally fused in a layered manner based on the geographical location information of each inspection area to obtain a fused inspection dataset; the fused inspection dataset is converted into a standard format adapted to the channel's 3D model update to generate a standard inspection dataset.

[0011] In one optional implementation, a historical standard inspection dataset is acquired, and the 3D model of the passage is updated based on the passage business security requirement data, the historical standard inspection dataset, and the standard inspection dataset. This includes: extracting the passage facility point cloud accuracy threshold, tree obstacle spatial distance constraints, and model update trigger accuracy conditions based on the passage business security requirement data; performing point cloud registration and alignment and voxel-based difference calculation on the standard inspection dataset and the historical standard inspection dataset to generate a passage spatial point cloud difference feature set; filtering out a set of difference regions based on the passage spatial point cloud difference feature set and the model update trigger accuracy conditions; and, based on the standard inspection dataset, under the passage facility point cloud accuracy threshold constraints... Local vertex correction and topology reconstruction are performed on each differential region, and the changes in point cloud geometric parameters before and after reconstruction are recorded. Based on the changes in point cloud geometric parameters, infrastructure data and tree 3D data are updated, and the position offset, height increment, crown change rate, and spatial distance change value of each tree with respect to access facilities are calculated. Based on the position offset, height increment, crown change rate, spatial distance change value of each tree with respect to access facilities, and preset tree obstacle risk judgment rules, the growth trend prediction results and risk levels in the hazard labeling data are updated. The accuracy of the updated model is verified based on the point cloud accuracy threshold of access facilities, and spatial compliance is verified based on the spatial distance constraints of tree obstacles.

[0012] Secondly, this invention provides a real-time channel tree obstacle modeling system based on point cloud data. The system includes: an inspection strategy planning module, a collaborative acquisition module, a heterogeneous data preprocessing module, a modeling and identification update module, an operation and maintenance closed-loop module, and a data service layer. Specifically: the inspection strategy planning module calls the channel 3D model stored in the data service layer, partitions the inspection channel based on the channel 3D model and partitioning rules, obtains a first region set and a second region set, and configures a first channel tree obstacle inspection strategy and a second channel tree obstacle inspection strategy for the first region set and the second region set respectively. The first channel tree obstacle inspection strategy is an encrypted acquisition strategy for the first region set, and the second channel tree obstacle inspection strategy is a standard acquisition strategy for the second region set. The collaborative acquisition module includes an inspection drone cluster and a collaborative scheduling terminal, used to receive and execute the first channel tree obstacle inspection strategy and the second channel tree obstacle inspection strategy, obtain first inspection data and second inspection data, and then... The data is transmitted in real time to the heterogeneous data preprocessing module. This module performs cross-machine registration, adaptive point cloud denoising, multi-machine point cloud incremental fusion, and data format standardization on the first and second inspection data to generate a standard inspection dataset. This standard inspection dataset is then transmitted to the modeling and recognition update module. The modeling and recognition update module acquires historical standard inspection datasets and updates the channel's 3D model based on channel business security requirements data, historical standard inspection datasets, and the standard inspection dataset. The operation and maintenance closed-loop module determines the target area based on the updated channel 3D model and inspection upgrade rules, updates the tree obstacle inspection strategy for the target area from the second channel tree obstacle inspection strategy to the first channel tree obstacle inspection strategy, and performs a secondary inspection on the target area. The data service layer stores the channel 3D model, historical standard inspection datasets, channel business security requirements data, partitioning rules, target tree determination conditions, inspection area overlap parameters, inspection upgrade rules, and model update logs.

[0013] The real-time channel obstacle modeling system based on point cloud data provided in this embodiment first establishes a fixed interaction link between the inspection strategy planning module and the underlying data service layer, streamlining the internal business scheduling logic of the system, improving the overall smoothness of task planning, and reducing the obstruction of instruction interaction between modules. Secondly, the collaborative acquisition module, based on a terminal scheduling architecture and UAV cluster onboard hardware, coordinates the operation sequence and task allocation of multiple devices, mitigating the impact of inconsistent operation paces between devices and ensuring continuous and orderly on-site acquisition. Next, the heterogeneous data preprocessing module, based on its built-in processing unit, handles multi-source inspection data, relying on its own hardware processing capabilities to complete the entire process of registration, denoising, and fusion, alleviating the computational pressure on the main system business and improving the throughput of large-volume point cloud image data processing. Furthermore, the modeling and recognition update module, based on its own computing hardware, handles time-series comparison and 3D reconstruction, independently completing model difference analysis and local reconstruction calculations without consuming resources from other system modules, improving the operational efficiency of channel 3D model iteration and refresh, and making the model state update process more stable and reliable. Then, a business linkage path is formed through the operation and maintenance closed-loop module and the front-end acquisition and back-end modeling modules. Based on the system's inherent transmission link, area determination, strategy adjustment, and secondary inspection task distribution are completed, improving the tightness of the entire operation and maintenance business link. Finally, the data service layer adopts a layered storage architecture to carry all kinds of business data of the entire system, providing unified data reading and writing support for each functional module, standardizing the data access order of the entire system, improving the adaptability and capacity when multiple modules retrieve data simultaneously, and avoiding operational delays caused by centralized data access. By implementing this solution, the significant shortcomings of related channel tree obstacle inspection and modeling technologies in terms of the rationality of inspection area planning, adaptability of acquisition operation mode, accuracy of multi-source acquisition data processing, timeliness of channel 3D model updates, and the continuity and dynamic adjustment capability of inspection and maintenance processes have been resolved. These shortcomings make it difficult to adapt to the high-precision, real-time, and refined tree obstacle inspection and modeling operation and maintenance needs of large-scale linear infrastructure channels.

[0014] In one optional implementation, the inspection strategy planning module is configured with a channel 3D model data interface and a strategy distribution bus; wherein: the channel 3D model data interface is used to establish a bidirectional data interaction link with the data service layer, supporting the calling and writing back of the channel 3D model; the strategy distribution bus is used to synchronously distribute the first channel tree obstacle inspection strategy and the second channel tree obstacle inspection strategy to the collaborative acquisition module in the form of an instruction stream.

[0015] In one optional implementation, the collaborative acquisition module is configured with a multi-machine collaborative scheduling bus, an airborne acquisition unit, and a real-time data feedback link; wherein: the multi-machine collaborative scheduling bus is used to receive the instruction stream issued by the inspection strategy planning module, and to perform multi-UAV timing coordination, area division coordination, and payload working status coordination management; the airborne acquisition unit integrates a lidar payload, a GNSS positioning module, and an IMU attitude module, and is used to synchronously acquire and generate 3D point clouds of channel facilities, 3D point clouds of vegetation, aerial image data, GNSS positioning data, IMU attitude data, and raw lidar scanning data; the real-time data feedback link is used to encapsulate the acquired first inspection data and second inspection data into a business data stream and push it to the heterogeneous data preprocessing module in real time.

[0016] In one optional implementation, the heterogeneous data preprocessing module is configured with a cross-machine registration interface, an adaptive denoising engine, an incremental fusion storage unit, and a format conversion unit. Specifically: the cross-machine registration interface is used to access GNSS positioning data and IMU attitude data from various inspection drones, uniformly registering the first and second inspection data to the channel's global coordinate system; the adaptive denoising engine has a built-in denoising threshold mapping table corresponding to different drone models, and uses a density clustering algorithm to perform differentiated point cloud denoising processing based on the point cloud acquisition accuracy and the appropriate threshold; the incremental fusion storage unit is used to partition and cache the denoised inspection data, performing hierarchical incremental fusion and data version tracing based on the geographical location information constraints of the inspection area; and the format conversion unit is used to convert the fused inspection dataset into a standard format that conforms to the access specifications of the modeling and recognition update module.

[0017] In one optional implementation, the modeling and recognition update module is configured with a model update control bus, a differential comparison engine, a 3D topology reconstruction unit, and a multi-dimensional verification unit. Specifically: the model update control bus establishes hardware-level linkage links with the format conversion unit of the heterogeneous data preprocessing module, the data service layer, and the operation and maintenance closed-loop module, respectively, for timing synchronization and priority management of data interaction; the differential comparison engine has a built-in point cloud data cache and differential feature extraction circuit, used to cache the standard inspection dataset output by the format conversion unit and the historical standard inspection dataset called by the data service layer, and supports the parallel processing of the standard inspection dataset and the historical standard inspection dataset through hardware logic circuits. The system includes: a row differential comparison unit; a 3D topology reconstruction unit, which integrates local reconstruction control circuitry and parameter storage registers, used to correct vertices and reconstruct models in different regions based on the accuracy threshold constraints in the channel service security requirement data, and synchronously store the changes in point cloud geometric parameters during the reconstruction process in the registers; and a multi-dimensional verification unit, which is equipped with accuracy verification circuitry and spatial compliance verification circuitry, used to read the accuracy threshold of the channel facility point cloud and the spatial distance constraints of tree obstacles, and perform parallel dual verification on the reconstructed local model. If the verification passes, a model write-back permission signal is generated; if the verification fails, a graded anomaly signal is generated, and the anomaly type and anomaly region are marked synchronously. Attached Figure Description

[0018] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating a specific example of a real-time channel tree obstacle modeling method based on point cloud data according to an embodiment of the present invention;

[0020] Figure 2 This is a principle block diagram of a specific example of a real-time channel tree obstacle modeling system based on point cloud data according to an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] With the continuous expansion of linear infrastructure networks such as power transmission channels, rail transit, and oil and gas transmission corridors in China, tree obstructions caused by vegetation growth along these channels have become a major source of risk affecting the safe and stable operation of infrastructure. Vegetation grows continuously with the seasons, easily encroaching on the safety clearance of facilities, causing safety accidents such as short circuits, power outages, and facility scraping. Regular inspections and 3D modeling of tree obstructions along these channels are a core requirement for operation and maintenance management. The use of UAV-borne LiDAR and aerial photography equipment for channel inspections has gradually replaced outdated methods such as traditional manual foot inspections and simple visual estimations.

[0023] In practical engineering operation and maintenance scenarios, linear corridors are characterized by long spans and complex and variable environments along their routes. Uneven distribution of vegetation density and spatial distances between trees and facilities naturally leads to varying tree obstacle risks in different sections. Existing routine inspection operations often utilize fixed routes and standardized configurations for comprehensive data collection. While this operational model is regular and easy to deploy in batches, it struggles to balance the precision required for key sections with the efficiency of operations in ordinary sections when faced with the varying characteristics of local environments within the corridor. This often results in insufficient detailed representation of key areas and redundant resource consumption in routine areas, indicating room for improvement in refined operation and maintenance management. Multi-UAV collaborative parallel inspections are already a common method for large-scale corridor operations. However, the objective differences in payload parameters, sensing accuracy, and flight attitude benchmarks among different devices result in significant heterogeneity in the generated point clouds, images, and positioning attitude data. Existing general data processing methods can complete basic noise reduction and simple stitching integration, but when faced with heterogeneous inspection data from multiple models and batches, there are still engineering pain points in areas such as unified spatial benchmarks, refined noise filtering, and smooth integration of fragmented data. This can easily lead to issues like misaligned local points, unnatural data transitions, and loss of local details, affecting the overall consistency and accuracy stability of subsequent 3D model reconstruction. At the 3D model operation and maintenance iteration level, the industry currently mostly adopts a periodic full-scale reconstruction approach, re-collecting data and updating the overall model at fixed time intervals. This operation and maintenance mode is rather static and lagging. It is difficult to accurately capture the subtle morphological changes and spatial distance evolution caused by slow vegetation growth, and it is also difficult to achieve synchronous and linked updates of model entity parameters and hazard attribute annotations. Furthermore, existing operation and maintenance processes mostly remain at the level of independent operations for single inspections and single analyses, lacking a closed-loop operation and maintenance system encompassing inspection discovery, model updates, adaptive strategy adjustments, and secondary verification. They rely heavily on manual experience to judge subsequent inspection plans, and the intelligent, dynamic, and adaptive operation and maintenance capabilities need further improvement.

[0024] In summary, the relevant tree obstacle inspection and modeling technologies still have significant shortcomings in terms of the rationality of inspection area planning, adaptability of data collection operation modes, accuracy of multi-source data processing, timeliness of updating the three-dimensional model of the channel, and the continuity and dynamic adjustment capability of the inspection and maintenance process. They are difficult to adapt to the high-precision, real-time and refined tree obstacle inspection and modeling maintenance needs of large-scale linear infrastructure channels.

[0025] To address the technical problems mentioned in the background section, this application provides a method for real-time modeling of channel tree obstacles based on point cloud data, as detailed in the following embodiments. Figure 1 As shown, the method includes:

[0026] Step S101: Based on the channel 3D model and partitioning rules, the inspection channel is partitioned to obtain a first region set and a second region set. A first channel tree obstacle inspection strategy and a second channel tree obstacle inspection strategy are configured for the first region set and the second region set, respectively. The first channel tree obstacle inspection strategy is an encrypted acquisition strategy for the first region set, and the second channel tree obstacle inspection strategy is a standard acquisition strategy for the second region set.

[0027] Specifically, the 3D model of the passageway includes infrastructure data, 3D tree data, and hazard annotation data attached to the 3D tree data. The hazard annotation data includes the location, height, crown width, distance from the passageway facilities, growth trend prediction results, and risk level of each tree. Step S101 includes:

[0028] Step a1: Analyze the hazard labeling data in the 3D model of the passage and mark trees with a risk level higher than the preset level as target trees.

[0029] Furthermore, the 3D model of the passageway is a 3D digital model covering the entire passageway scene, containing three core types of data: infrastructure data, namely the 3D shape and spatial location data of various fixed facilities within the passageway; tree 3D data, namely the 3D shape and spatial distribution of all vegetation trees along the passageway; and hazard labeling data, namely the attribute data attached to the 3D data of each tree, specifically including the tree's location, height, crown width, distance from passageway facilities, growth trend prediction results, and risk level. These data are interconnected. By retrieving the 3D model of the passageway, the hazard labeling data bound to the 3D data of all trees in the model is extracted, clarifying the specific attribute characteristics of each tree. Among these, tree location refers to the specific spatial orientation of the tree; height refers to the vertical dimension from the ground to the top of the crown; crown width refers to the horizontal extension range of the crown; distance from passageway facilities refers to the shortest spatial interval between the tree and various infrastructure facilities within the passageway; growth trend prediction results refer to the predicted conclusions on the tree's subsequent growth rate and direction; and risk level refers to the degree of safety risk assessed based on the tree's various attributes. Preset risk levels are permanently stored in the rule configuration library of the data service layer. This library independently categorizes risk level classification entries, with each entry containing four levels from low to high: ordinary, attention, warning, and high-risk. Each level corresponds to fixed judgment field thresholds and attribute constraint logic, which can be directly read and called by the machine. Each tree's inherent risk level field is matched against the preset level benchmark. After traversal, all tree entries with a risk level higher than the preset level are selected. Tree entries meeting the selection criteria are written to a unique identifier field, completing the batch labeling and data collection of target trees. For example, along a power transmission line, trees close to the conductor, with large canopies and a growth trend towards the conductor, will be marked as target trees if their risk level is higher than the preset level after analyzing their hazard labeling data.

[0030] Step a2: Based on the constraints of the coverage area of ​​the UAV inspection operation, the tree location, height, crown width, distance from the passage facilities, and growth trend prediction results of each target tree, the inspection passage is divided into several inspection areas, with overlapping areas between adjacent inspection areas.

[0031] Furthermore, the coverage constraints of UAV inspection operations are pre-stored in the operation constraint library of the data service layer in the form of spatial boundary templates. The templates limit the spatial extension range and terrain adaptation boundaries that a single inspection operation can completely cover. The full dataset of target trees already labeled in the data service layer is read, and structured fields such as tree location, height, crown width, distance from passage facilities, and growth trend prediction results are extracted for each entry, integrating them to form a dataset of target tree spatial distribution and morphological characteristics. The spatial boundary templates within the operation constraint library are loaded, and the template boundaries are used as the unit segmentation scale. Combined with the density, shape, spatial spacing, and growth trend characteristics of the target trees, spatial units are segmented segment by segment along the natural direction of the passage, forming several inspection areas with continuous boundaries. During the segmentation process, at the connection boundary positions of every two adjacent inspection areas, continuous spatial bandwidth is reserved according to the connection overlap templates built into the operation constraint library to generate a fixed-shape overlapping area, allowing adjacent inspection areas to spatially overlap. This provides an executable spatial division basis for subsequent multi-drone data acquisition flight path transitions and seamless point cloud data stitching. For example, when dividing power transmission channels in mountainous areas, the overlapping area of ​​two adjacent inspection areas can cover parts of the mountain with large slope changes, ensuring that no data is missed in the collection of data in that area.

[0032] Step a3: Based on the partitioning rules, the inspection areas containing the target trees are assigned to the first region set, and the inspection areas that do not contain the target trees are assigned to the second region set.

[0033] Furthermore, the pre-defined zoning rules are the criteria for defining the category of inspection areas, with the core basis being whether the inspection area contains target trees. According to these zoning rules, each divided inspection area is checked one by one to confirm whether the previously marked target trees exist. Inspection areas confirmed to contain target trees are uniformly classified into the first zone set. These areas are often sections with concentrated target trees and high risk levels, such as sections of power transmission corridors near conductors with dense target trees. Inspection areas confirmed not to contain any target trees are uniformly classified into the second zone set. These areas are often sections with sparse vegetation, low tree risk levels, and far from corridor facilities, such as power transmission corridor sections in open suburban areas with no target trees.

[0034] Step a4: Based on the tree location, height, crown width, distance from the passage facilities, and growth trend prediction results of each target tree, configure encrypted data collection operation parameters and corresponding inspection drones for each inspection area in the first area set, and generate the first passage tree obstacle inspection strategy. The encrypted data collection operation parameters include flight altitude, flight path overlap, point cloud sampling frequency, and number of collection viewpoints.

[0035] Furthermore, attribute data for all target trees within each inspection area of ​​the first region set are extracted, including tree location, height, crown width, distance from access facilities, and growth trend prediction results. This clarifies the distribution density, growth status, proximity to access facilities, and potential risk level of target trees within each inspection area. Based on the analysis results, encrypted data collection parameters are configured individually for each inspection area in the first region set. These parameters are control parameters used to improve data collection accuracy and ensure the accuracy of hazard identification. Specifically, they include flight altitude, flight path overlap, point cloud sampling frequency, and number of collection angles. Flight altitude refers to the altitude at which the UAV collects data; flight path overlap refers to the overlap ratio between two adjacent inspection flight paths; point cloud sampling frequency refers to the density of point cloud data collection; and the number of collection angles refers to the number of data collected by the UAV from different angles. Simultaneously, based on the scene characteristics of each inspection area, a corresponding inspection UAV model is selected to ensure that the UAV's payload performance and flight stability are suitable for the area's data collection needs. For example, in inspection areas with tall, dense target trees and complex terrain, a UAV with a strong payload and adaptability to complex airspace is selected. By combining the encrypted data collection parameters corresponding to each inspection area with the selected inspection drone model, a first-channel tree obstacle inspection strategy for the first area set is formed.

[0036] Step a5: Configure unified routine data collection parameters and corresponding inspection drone models for each inspection area in the second area set, and generate a second-channel tree obstacle inspection strategy. The routine data collection parameters include flight altitude, flight path overlap, point cloud sampling frequency, and number of collection viewpoints. The point cloud sampling frequency and number of collection viewpoints in the routine data collection parameters are lower than those in the encrypted data collection parameters.

[0037] Furthermore, the standard data acquisition parameters are standardized parameters that meet the needs of basic inspection, data collection, and model updates. Their core dimensions are consistent with those of the encrypted data acquisition parameters, namely flight altitude, flight path overlap, point cloud sampling frequency, and number of acquisition viewpoints. Among these, the point cloud sampling frequency and number of acquisition viewpoints in the standard data acquisition parameters are generally set to lower standards than those in the encrypted data acquisition parameters, only needing to meet basic data collection and daily inspection requirements. Simultaneously, a corresponding model of inspection drone is uniformly selected for all inspection areas in the second region set. This drone model must be adapted to the needs of standard data acquisition operations, possess stable flight performance and basic data collection capabilities, and does not require a high-performance payload. For example, in open, sparsely vegetated inspection areas in suburban areas, a general-purpose inspection drone is uniformly selected. Combining the standardized standard data acquisition parameters with the selected inspection drone model forms a second-channel tree obstacle inspection strategy for the second region set.

[0038] Step S102: Execute the first channel tree obstacle inspection strategy and the second channel tree obstacle inspection strategy to obtain the first inspection data and the second inspection data.

[0039] Specifically, both the first and second inspection data include 3D point clouds of passage facilities, 3D point clouds of vegetation, aerial imagery data, GNSS positioning data, data acquisition time stamps and corresponding acquisition operation parameters, IMU attitude data, raw LiDAR scan data, and UAV model identification. Following the first and second passage tree obstacle inspection strategies, the acquisition operation parameters and the selected inspection UAV model information are loaded sequentially within the strategies. Based on the operation sequence and spatial flight path layout logic arranged by the strategies, automated aerial surveying and data acquisition operations are carried out across the entire passage area. Based on the regional affiliation range corresponding to the strategy, encrypted full-element data acquisition operations are performed on the first regional set coverage area, while standard normal data acquisition operations are performed on the second regional set coverage area. Throughout the process, various airborne sensor acquisition units are synchronously activated according to the predetermined flight trajectory and payload working logic. After completing the aerial surveying and data acquisition operations in the corresponding areas, all raw data generated during the operations in both types of areas are collected and organized to form the corresponding first and second inspection data. Both types of inspection data are packaged and archived according to a unified data structure specification. The 3D point cloud of the corridor facilities is a spatially discrete point set generated by LiDAR scanning. It fully depicts the external morphology and spatial distribution of fixed infrastructure such as tower structures, line hardware, and foundation structures along the corridor, restoring the overall three-dimensional outline and local structural details of the facilities. It serves as the basic geometric data source for constructing the 3D model of the corridor. The 3D point cloud of vegetation is generated synchronously by the same LiDAR sensing device, recording the crown shape, branch distribution, and vertical growth morphology of various trees and shrubs along the corridor. It reproduces the three-dimensional structural characteristics of natural vegetation growth and is used for subsequent determination of spatial distance to tree obstacles and tracking of morphological changes. Aerial imagery data is generated synchronously by the airborne visual acquisition unit, preserving real-world images of the corridor's surface topography, vegetation community distribution, and the surrounding environment of the facilities. This data can complement the point cloud data in terms of texture. GNSS positioning data records the spatial position change trajectory of the inspection drones throughout the entire aerial survey process, continuously marking spatial point information during operations. This provides a positional reference for the homing of multiple drone operation trajectories and cross-regional data spatial benchmark alignment. IMU attitude data is collected in real time to capture attitude changes such as pitch and yaw during UAV flight, recording the continuous dynamic process of flight attitude. This data is used to correct lidar scanning viewpoint deviations and compensate for point cloud spatial misalignment caused by flight attitude fluctuations. Raw lidar scanning data retains the original echo information and scanning time sequence records of the entire laser detection and transmission process, without pre-processing simplification, preserving complete low-level detection information for subsequent depth calculations and fine point cloud reconstruction. Acquisition time sequence timestamps are used to mark the time information corresponding to each scan and image acquisition action frame by frame, forming a time sequence according to the order of operations. This is used to distinguish the acquisition content from different aerial survey periods and different areas.Corresponding data acquisition parameters are synchronously associated and bound to each inspection data entry, fully recording configuration details such as flight altitude, flight path overlap, point cloud sampling frequency, and number of acquisition angles used in the operation, thus achieving traceability and binding between acquisition parameters and raw data. The UAV model identifier is permanently written into the header of each data entry, marking the equipment model used in the current acquisition operation. For example, different UAV models have different LiDAR sensing performance, and specific processing rules can be matched based on the model identifier.

[0040] Step S103: Perform cross-machine registration, adaptive point cloud denoising, multi-machine point cloud incremental fusion, and data format standardization on the first and second inspection data to generate a standard inspection dataset.

[0041] Specifically, step S103 includes:

[0042] Step b1: Using a registration algorithm, the first inspection data and the second inspection data are uniformly registered to the channel global coordinate system based on the GNSS positioning data and IMU attitude data of each inspection UAV.

[0043] Furthermore, the global coordinate system of the channel is a pre-established and fixed global spatial reference framework, covering the entire geographic space of the inspection channel, serving as a reference system for the unified alignment of all inspection data. GNSS positioning data and IMU attitude data attached to the first and second inspection data are retrieved. The GNSS positioning data records the spatial position and temporal trajectory during operation, while the IMU attitude data records the real-time attitude deflection during flight. A preset registration algorithm is invoked to continuously calculate the point-time information and attitude change information corresponding to each individual inspection UAV. Based on the temporal position and attitude deflection, the lidar scanning line-of-sight angle and spatial projection azimuth are corrected. The point cloud and image information collected by different UAVs in their respective local coordinate systems are mapped and transformed frame by frame to the global coordinate system of the channel for coordinate normalization. Corridor point clouds and vegetation images collected by multiple devices in different areas are aligned according to the same reference, enabling the data collected in blocks to be arranged within the same spatial framework, avoiding spatial misalignment and azimuth shift in data from different devices.

[0044] Step b2: Obtain the denoising threshold corresponding to each inspection drone. Use density clustering algorithm to denoise the registered first and second inspection data based on the denoising threshold corresponding to each inspection drone. The denoising threshold corresponding to each inspection drone is set based on the point cloud acquisition accuracy.

[0045] Furthermore, the denoising threshold is a pre-defined criterion based on the hardware acquisition characteristics of the equipment. It is categorized and set according to the point cloud acquisition accuracy of different inspection drones equipped with LiDAR. Different devices have their own independent denoising thresholds, which are then centrally archived and stored. The density clustering algorithm is a processing logic that classifies point locations according to the density of spatial point clusters. It can distinguish between valid entity points and discrete interference points based on local point distribution characteristics. The corresponding denoising threshold is retrieved one by one according to the drone model identifier. The point cloud points in the first and second inspection data, which have already completed global coordinate registration, are traversed and retrieved. Based on the density clustering algorithm, the point cloud is divided into clusters of units. The denoising thresholds of the corresponding devices are used to identify isolated and scattered points from clusters of valid points. Dense point cloud units that fit the shape of infrastructure and the outline of vegetation are retained, while floating noise in the air, scattered noise on the ground, and invalid discrete points formed by environmental reflections are removed, completing the adaptive hierarchical denoising of the entire point cloud data.

[0046] Step b3 involves performing hierarchical incremental fusion of the denoised first and second inspection data based on the geographical location information of each inspection area to obtain a fused inspection dataset.

[0047] Furthermore, the geographical location information of the inspection area is the spatial boundary range information retained and archived in the previous partitioning stage, recording the spatial extension direction and connection range of each inspection area, and also including the spatial range of overlapping areas between adjacent inspection areas. The first and second inspection data, after registration and denoising processing, are retrieved, and the geographical location information of the inspection area to which each data belongs is associated. The data segments are arranged sequentially according to the channel extension direction. For points and image information falling within the overlapping range of adjacent inspection areas, a fixed data acceptance order is established, prioritizing the point cloud and image content corresponding to the first inspection data as the baseline, with the second inspection data only serving as a supplementary reference for edge details. Within the overlapping area, point connection and texture transition processing are performed according to predetermined fusion rules. Using the contour shape and spatial points of the first inspection data as the fitting standard, the second data is adapted for convergence and smoothing fitting, eliminating point discrepancies and texture breaks at the connection points of the segmented data. Based on preserving the collected data in non-overlapping areas, a smooth stitching process dominated by high-priority data is performed in overlapping areas. Data content from each inspection area is incorporated layer by layer and segment by segment. The data collection and integration of the entire domain is completed in an incremental manner, ultimately generating a fused inspection dataset that is seamlessly connected, without spatial fragmentation, and without the need to copy redundant information.

[0048] Step b4: Convert the fused inspection dataset into a standard format adapted to the channel 3D model update to generate a standard inspection dataset.

[0049] Furthermore, the channel 3D model update has fixed data access standards and parsing format specifications, setting unified requirements for point cloud layout structure, image encapsulation method, and storage format of auxiliary attribute fields. The integrated inspection dataset is read and parsed, containing 3D point clouds of channel facilities, 3D point clouds of vegetation, aerial image data, and various auxiliary time-series and parameter identification information. The internal data organization structure is reconstructed according to the access specifications of the model update stage, uniformly adjusting the point cloud storage arrangement, image compression and encapsulation format, and the mounting position and association relationships of various auxiliary fields. The field naming rules, hierarchical directory structure, and index association logic of data from different sources are standardized, enabling the processed overall data to be directly recognized, read, and parsed by the modeling stage. After format conversion, the data is neatly archived to form a standard inspection dataset, maintaining the integrity of all information without loss, while also meeting the access requirements for subsequent time-series comparison, topology reconstruction, and model iteration updates.

[0050] Step S104: Obtain historical standard inspection dataset, and update the channel 3D model based on channel business security requirement data, historical standard inspection dataset, and standard inspection dataset.

[0051] Specifically, step S104 includes:

[0052] Step c1: Based on the channel service security requirement data, extract the channel facility point cloud accuracy threshold, tree obstacle spatial distance constraint, and model update trigger accuracy condition.

[0053] Furthermore, the channel service security requirement data is a pre-organized and archived set of industry operation and maintenance specifications and scenario control constraints, with built-in precision control standards and spatial constraint criteria adapted to channel operation and maintenance scenarios. The complete channel service security requirement data is read, and core constraint items are extracted layer by layer according to the business field classification dimension. Channel facility point cloud precision thresholds, used to control the level of detail in facility modeling, are extracted to define the allowable level of detail and morphological fit in infrastructure point cloud modeling. Simultaneously, tree barrier spatial distance constraints, used to regulate the spacing relationship between vegetation and the structure, are extracted to define the spatial separation benchmarks that must be adhered to between trees and channel infrastructure. Finally, model update trigger precision conditions, used to control the model update initiation conditions, are extracted as the benchmark for determining whether to initiate local model reconstruction.

[0054] Step c2 involves performing point cloud registration and alignment, and voxelization difference calculation on the standard inspection dataset and the historical standard inspection dataset to generate a channel spatial point cloud difference feature set.

[0055] Furthermore, the currently generated standard inspection dataset and the historical standard inspection datasets from previous archives are retrieved and uniformly incorporated into the same spatial reference framework for time-series data normalization. Global point cloud registration and alignment are performed on the two sets of time-series datasets to ensure that point cloud points collected at different times converge to the same spatial coordinate reference, eliminating spatial offsets caused by differences in time-series acquisition perspectives and flight paths. After registration and alignment, voxel-based differential calculations are performed on the two sets of point cloud datasets. Voxel-based differential calculations divide the continuous space into uniformly distributed spatial units. Using the unit as the smallest comparison unit, the point cloud distribution, density, and spatial occupancy of historical and current time periods are compared one by one. The differences in point state and shape contours are compared unit by unit, and information on all units with changes in shape, offset, or spatial occupancy is collected. This information is then integrated according to spatial location associations to form a structured channel spatial point cloud difference feature set, comprehensively recording the changes in entity shape and spatial distribution across the entire temporal domain.

[0056] Step c3: Based on the channel spatial point cloud difference feature set and the model update trigger accuracy condition, a set of difference regions is obtained by filtering.

[0057] Furthermore, the generated channel spatial point cloud difference feature set is retrieved, and the model update trigger accuracy conditions extracted from the channel business security requirement data are loaded. Each spatial difference unit in the difference feature set is then individually compared and verified against the model update trigger accuracy conditions. Based on the spatial offset degree and contour modification range of the point cloud morphology changes recorded within the feature set, each difference unit is judged to meet the preset trigger accuracy level. Spatial difference units that meet the trigger accuracy level requirements are aggregated into connected components according to their geographical adjacency, and spatially connected difference units with similar attributes are grouped and integrated into independent regional units. The scope of each aggregated unit is defined according to the spatial boundary, and these are sequentially aggregated to form a set of contiguous difference regions, defining a clear operational boundary for local model correction and topology reconstruction.

[0058] Step c4: Based on the standard inspection dataset, under the constraint of the accuracy threshold of the point cloud of the channel facilities, perform local vertex correction and topology reconstruction for each differential region, and record the changes in point cloud geometric parameters before and after reconstruction.

[0059] Furthermore, each difference region covered by the difference region set is traversed. Based on the standard inspection dataset as the data source for morphological reconstruction, local reconstruction is carried out using a modeling refinement standard limited by the accuracy threshold of the channel facility point cloud. Within the defined boundary of the difference region, the facility and vegetation point cloud locations corresponding to the standard inspection dataset are extracted. The vertices of the original model mesh are corrected and adjusted point by point, and the surface vertex distribution is refitted according to the actual point cloud contour trend. Simultaneously, local topology reconstruction is carried out to reorganize the mesh connection relationships and spatial topology structure within the difference region, adapting to the shape deformation, branch growth, or subtle structural displacement of the actual entity on site. After completing the local correction and topology reconstruction, the spatial position, shape contour, and extension range of the model point cloud before and after reconstruction are compared. The changes in various point cloud geometric parameters are recorded one by one, providing the original comparison basis for subsequent shape parameter iterations.

[0060] Step c5: Based on the changes in point cloud geometric parameters, update the infrastructure data and tree 3D data, and calculate the position offset, height increment, crown change rate, and spatial distance change value of each tree from the access facilities.

[0061] Furthermore, based on the changes in point cloud geometric parameters recorded in each differential region, these changes are mapped to the corresponding infrastructure data and tree 3D data within the 3D model of the passageway. The model's shape fields are then synchronously updated according to the magnitude of parameter changes. For the infrastructure data, local structural outlines and spatial point distributions are corrected to adapt to actual working conditions such as foundation settlement and minor structural changes. For the tree 3D data, the canopy boundary branch extension and vertical height morphology are refitted, and the positional offset, height increment, crown width change rate, and spatial distance change value of each tree are simultaneously calculated. The positional offset represents the horizontal variation of the tree's overall spatial position, the height increment represents the height difference caused by the tree's vertical growth, the crown width change rate represents the proportion of change in the horizontal extension range of the canopy, and the spatial distance change value represents the increase or decrease in the distance between the tree and surrounding facilities. After all parameters are calculated, they are uniformly written into the corresponding attribute fields of the model to complete the iteration.

[0062] Step c6: Based on the location offset, height increment, crown width change rate, spatial distance change value between each tree and the passage facility, and the preset tree obstacle risk judgment rules, update the growth trend prediction results and risk level in the hazard labeling data.

[0063] Furthermore, the preset tree obstacle risk assessment rules are pre-fixed and stored hierarchical assessment logic, which divides different risk levels based on tree shape parameters and spatial interval status. The system retrieves the updated position offset, height increment, crown width change rate, and spatial distance change value between each tree and the access facilities, and substitutes these values ​​into the preset tree obstacle risk assessment rules for hierarchical matching and analysis. Combining the changing trends of various parameters, the system infers the subsequent growth extension direction and invasion status of the trees, updating the growth trend prediction results within the hazard labeling data. Simultaneously, based on the current crown width expansion degree and growth invasion trend, the system reassesses the safety risk level corresponding to each tree, updating the risk level field within the hazard labeling data. After completing the analysis and updating of each individual tree, the data is synchronously bound to the corresponding tree's 3D data entry, achieving synchronous updates of hazard labeling data and temporal entity status.

[0064] Step c7: Verify the accuracy of the updated model based on the point cloud accuracy threshold of the access facility, and verify the spatial compliance based on the spatial distance constraint of the tree barrier.

[0065] Furthermore, the accuracy threshold of the point cloud of the passage facilities is used as the verification benchmark to conduct a full-domain accuracy verification of the passage 3D model after local correction and topology reconstruction. The fitting degree of the reconstructed point cloud, the contour restoration accuracy, and the completeness of detail expression are compared region by region to verify whether the model's shape replication meets the preset accuracy control standards, ensuring that the restoration of infrastructure and vegetation shapes meets the requirements of operation and maintenance modeling. After completing the accuracy verification, the spatial distance constraint of tree barriers is used as the interval judgment benchmark to verify the 3D data of each tree and the actual spatial interval of the surrounding passage infrastructure, comparing whether the preset safety separation range is adhered to. The proximity of the tree crown extension branches to the framework lines is checked to identify whether there are situations where the spatial spacing approaches the constraint boundary. The full-domain spatial compliance verification is completed sequentially, and the model that passes the verification is officially established as the updated passage 3D model.

[0066] Step S105: Based on the updated channel 3D model and inspection upgrade rules, determine the target area, update the tree obstacle inspection strategy of the target area from the second channel tree obstacle inspection strategy to the first channel tree obstacle inspection strategy, and perform a second inspection on the target area.

[0067] Specifically, the process retrieves the 3D model of the channel after accuracy verification and parameter synchronization updates, and loads pre-stored inspection upgrade rules. These rules are judgment criteria specifically designed for the second region set and are only used to upgrade areas that originally implemented the standard data collection strategy, without performing duplicate verification on the first region set. The inspection upgrade rules clearly define three triggering conditions: first, a significant increase in tree height increment; second, an outward expansion trend in the rate of change of tree crown width; and third, a continuous reduction in the change value of the spatial distance between trees and channel facilities. Meeting any one of these conditions constitutes a condition for strategy upgrade. The process iterates through all inspection areas included in the second region set, extracting the updated position offset, height increment, crown width change rate, and change value of the spatial distance between trees and channel facilities for each tree within the area. Simultaneously, the updated growth trend prediction results and risk levels are matched, and each condition is compared against the inspection upgrade rules. Inspection areas in the second region set that meet any of the triggering conditions are selected and uniformly designated as target areas. After identifying the target area from the second set of regions, the previously bound and effective second-channel tree obstacle inspection strategy for the target area is cancelled, and the original association configuration between the conventional data acquisition parameters and the corresponding inspection drone model is removed. The fixed template of the first-channel tree obstacle inspection strategy is directly applied, and the encrypted data acquisition parameters are replaced with those for the target area. Simultaneously, an inspection drone model suitable for complex vegetation environments and close-range fine scanning conditions is matched. Based on the encrypted data acquisition parameters, the inspection flight path and coverage area within the target area are reconfigured. Strictly following the corresponding encrypted data acquisition operation mode and payload working mechanism, a secondary inspection process is independently executed within the target area boundary. This process fully acquires all information, including 3D point clouds of channel facilities, 3D point clouds of vegetation, aerial imagery data, and supporting GNSS positioning data, IMU attitude data, and raw LiDAR scan data. This completes the strategy switch from conventional standard acquisition to encrypted fine acquisition and a targeted closed-loop verification.

[0068] The real-time channel tree obstacle modeling method based on point cloud data provided in this embodiment firstly divides the inspection channel into regions based on the channel's 3D model and partitioning rules, and matches corresponding tree obstacle inspection strategies to different region sets, improving the precision of inspection region division, optimizing the matching logic of inspection tasks, and enhancing the adaptability of inspection operations to the actual channel scenario. Secondly, by executing the corresponding configured channel tree obstacle inspection strategies and acquiring corresponding inspection data, the quality of inspection data acquisition under different scenarios is improved, the completeness of scene information in key areas is enhanced, and the basic acquisition level of regular areas is maintained stably, optimizing the overall data acquisition balance. Then, by sequentially performing cross-machine registration, adaptive denoising, multi-machine point cloud incremental fusion, and data format standardization processing on multiple types of inspection data to form a standard inspection dataset, the baseline deviation between multi-source heterogeneous data is reduced, the interference of irrelevant noise on effective point cloud information is weakened, the fusion and normalization capability of multiple batches of inspection data is improved, and the compatibility and adaptability between various types of inspection data are enhanced. By introducing historical standard inspection datasets and combining them with current standard inspection datasets to update the overall 3D model of the passageway, the sensitivity of the 3D model's status synchronization is improved, enhancing the model's ability to perceive changes in vegetation and facility morphology, and improving the synchronization level of model morphological parameters and spatial attributes. Then, based on the updated 3D model and inspection upgrade rules, target areas are identified, the original inspection strategy is adapted and adjusted, and secondary inspections are conducted in the target areas. This increases the depth of verification in sensitive areas, reduces the probability of missed detections of tree obstructions, enhances the closed-loop linkage capability of the inspection and maintenance process, improves the flexibility of dynamic strategy adaptation, and strengthens the stability of long-term safe operation and maintenance of the passageway. Implementing this solution addresses the significant shortcomings of existing tree obstruction inspection and modeling technologies in terms of the rationality of inspection area planning, adaptability of data collection modes, accuracy of multi-source data processing, timeliness of 3D model updates, and the continuity and dynamic adjustment capability of the inspection and maintenance process. These shortcomings make it difficult to meet the high-precision, real-time, and refined tree obstruction inspection and modeling maintenance needs of large-scale linear infrastructure passageways.

[0069] The above are embodiments of the real-time channel tree obstacle modeling method based on point cloud data provided in this application. Other embodiments of the real-time channel tree obstacle modeling method based on point cloud data provided in this application are described below.

[0070] This invention also discloses a real-time channel tree obstacle modeling system based on point cloud data, such as... Figure 2 As shown, the system includes: an inspection strategy planning module, a collaborative data acquisition module, a heterogeneous data preprocessing module, a modeling and identification update module, an operation and maintenance closed-loop module, and a data service layer; among which:

[0071] The inspection strategy planning module is used to call the channel 3D model stored in the data service layer, and perform inspection channel partitioning based on the channel 3D model and partitioning rules to obtain a first region set and a second region set. The module then configures a first channel tree obstacle inspection strategy and a second channel tree obstacle inspection strategy for the first region set and the second region set, respectively. The first channel tree obstacle inspection strategy is an encrypted collection strategy for the first region set, and the second channel tree obstacle inspection strategy is a standard collection strategy for the second region set.

[0072] Specifically, the inspection strategy planning module is configured with a channel 3D model data interface and a strategy distribution bus; among which:

[0073] The channel 3D model data interface is used to establish a two-way data interaction link with the data service layer, supporting the calling and writing back of the channel 3D model.

[0074] Furthermore, the channel 3D model data interface serves as a dedicated business interaction path between the inspection strategy planning module and the data service layer. It establishes a stable and reliable bidirectional data transmission link between modules, supporting two basic business interaction behaviors: model data retrieval and result write-back. Following a preset data interaction protocol format, it initiates a data retrieval request, and according to the standardized storage directory structure of the channel 3D model, it reads infrastructure data, tree 3D data, and attached hazard annotation data layer by layer, completely retrieving relevant structured fields and spatial topology information from the entire model. After completing the business calculations related to inspection channel partitioning and strategy configuration, it transmits intermediate business data, such as the region set partitioning results and strategy configuration relationships, back to the designated storage directory of the data service layer via the interaction link, according to a unified write-back encapsulation format. The entire process adheres to fixed communication timing and data encapsulation specifications, ensuring complete and lossless model data retrieval while supporting the orderly write-back of intermediate business processing results.

[0075] The strategy distribution bus is used to synchronously distribute the first channel tree obstacle inspection strategy and the second channel tree obstacle inspection strategy to the collaborative acquisition module in the form of a command stream.

[0076] Furthermore, the strategy distribution bus serves as a dedicated flow path for the inspection strategy planning module to transmit business instructions downwards, undertaking the functions of encapsulating and converting completed inspection strategies and pushing them in a targeted manner. The configured first-channel and second-channel tree obstacle inspection strategies are structurally reorganized according to a preset instruction stream encapsulation protocol, arranging strategy elements such as regional affiliation, data collection parameters, compatible UAV models, and flight path layout logic into a standardized continuous instruction stream. Following the transmission routing rules between business modules, the instruction streams corresponding to the two types of strategies are synchronously pushed to the collaborative acquisition module, maintaining a regular instruction distribution sequence and complete content items. A fixed transmission verification mechanism is followed to confirm the delivery of the instruction streams, ensuring that downstream modules can fully parse the configuration information within the strategies and directly perform multi-UAV collaborative acquisition operations according to the instruction stream content.

[0077] The collaborative data acquisition module includes a cluster of inspection drones and a collaborative scheduling terminal, which are used to receive and execute the first channel tree obstacle inspection strategy and the second channel tree obstacle inspection strategy, obtain the first inspection data and the second inspection data, and transmit the first inspection data and the second inspection data to the heterogeneous data preprocessing module in real time.

[0078] Specifically, the collaborative acquisition module is equipped with a multi-machine collaborative scheduling bus, an airborne acquisition unit, and a real-time data feedback link; among which:

[0079] The multi-drone collaborative scheduling bus is used to receive the instruction stream issued by the inspection strategy planning module and to perform time-series coordination, area division coordination, and payload working status coordination management of multiple drones.

[0080] Furthermore, the multi-machine collaborative scheduling bus undertakes the transmission and scheduling functions of cross-module command reception and internal cluster collaborative management. It continuously monitors the standardized command stream pushed by the bus under the policy, and fully analyzes the regional boundary range, data acquisition configuration, UAV allocation relationship, and operation timing information carried within the command stream. After decomposing the command stream, multi-UAV timing collaborative management is carried out sequentially, uniformly calibrating the operation start time and flight timing rhythm of all UAVs within the cluster to maintain synchronized progress across the entire domain. Simultaneously, regional division collaborative management is carried out, assigning a dedicated operation area to each UAV according to the inspection area boundaries defined by the command stream, ensuring that the boundaries of each operation area are connected in an orderly manner and do not overlap or conflict. Simultaneously, payload working status collaborative management is completed, uniformly setting the start time, working mode, and standby switching logic of the LiDAR payload, visual acquisition unit, and positioning attitude module, ensuring that all devices in the cluster maintain consistent payload start-stop timing and synchronously carry out full-domain data acquisition operations according to unified operating specifications.

[0081] The airborne acquisition unit integrates a lidar payload, a GNSS positioning module, and an IMU attitude module to simultaneously acquire and generate 3D point clouds of passage facilities, 3D point clouds of vegetation, aerial image data, GNSS positioning data, IMU attitude data, and raw lidar scan data.

[0082] Furthermore, the airborne acquisition unit integrates a lidar payload, a GNSS positioning module, and an IMU attitude module into a unified airborne sensing carrier, mounted inside the fuselage of each inspection UAV. It synchronously enters the startup state according to the work sequence issued by the collaborative scheduling bus. The lidar payload continuously emits detection signals and receives echo information, continuously scanning the structure and vegetation contours along the flight path, generating 3D point clouds of the facilities and vegetation point clouds point by point, completely replicating the external structure of the infrastructure and the three-dimensional morphological details of tree crowns and branches. The GNSS positioning module continuously records the spatial position change trajectory during flight, forming a continuous and uninterrupted position time sequence record. The IMU attitude module captures pitch, roll, and other attitude changes in real time during flight, generating a continuous attitude time sequence record. The three types of hardware units maintain synchronized sampling timing, synchronously outputting aerial image data, GNSS positioning data, IMU attitude data, and unprocessed raw lidar scan data, completely constituting a single device's full-dimensional raw acquisition information set.

[0083] The real-time data feedback link is used to encapsulate the collected first and second inspection data into a business data stream and push it to the heterogeneous data preprocessing module in real time.

[0084] Furthermore, a dedicated business transmission channel is established between the collaborative acquisition module and the heterogeneous data preprocessing module for the real-time data backhaul link. This channel undertakes the task of collecting and forwarding all raw data generated by the UAV swarm operation. Various types of raw sensing data generated by each airborne acquisition unit are collected and categorized according to the regional classification of the first and second inspection data. The internal structure, field relationships, and transmission frame format of the data are uniformly arranged according to a preset business data encapsulation protocol. The categorized and organized information is then packaged into a standard business data stream. Maintaining a continuous online real-time transmission status, the data stream is pushed frame by frame during the UAV inspection operation, without waiting for the entire area to finish before centralized transmission. The formed business data stream is continuously pushed to the heterogeneous data preprocessing module in real time, ensuring that the backend can receive data segment by segment and perform continuous registration, denoising, and fusion processing, maintaining the continuity and timing of data flow throughout the entire business link.

[0085] The heterogeneous data preprocessing module is used to perform cross-machine registration, adaptive point cloud denoising, multi-machine point cloud incremental fusion, and data format standardization on the first and second inspection data to generate a standard inspection dataset, and then transmit the standard inspection dataset to the modeling and recognition update module.

[0086] Specifically, the heterogeneous data preprocessing module is configured with a cross-machine registration interface, an adaptive denoising engine, an incremental fusion storage unit, and a format conversion unit; among which:

[0087] The cross-machine registration interface is used to access the GNSS positioning data and IMU attitude data of each inspection drone, and to uniformly register the first inspection data and the second inspection data to the channel global coordinate system.

[0088] Furthermore, a cross-machine registration interface establishes a dedicated data access channel for the heterogeneous data preprocessing module to connect with the front-end collaborative acquisition module. This channel specifically receives GNSS positioning data and IMU attitude data transmitted back by each inspection UAV, while also receiving the synchronously pushed first and second inspection data. Following a pre-defined data access communication protocol, various positioning and attitude time-series information and raw point cloud image data are parsed line by line, extracting the position trajectory sequence and attitude change sequence of each UAV throughout its flight. Built-in spatial calculation logic is invoked to correct the spatial reference deviation of the UAV's flight path based on the positioning time-series trajectory, and to compensate for the deflection error of the lidar scanning angle based on the attitude change sequence. The coordinate normalization and conversion of the point cloud and image information collected by different UAVs are performed frame by frame. All the first and second inspection data collected by different devices and by different areas are mapped and converged into a pre-established channel global coordinate system to achieve unified reference alignment, thus standardizing the spatial positional relationship of data collected by different devices and eliminating spatial misalignment and azimuth offset caused by differences in flight attitude and flight path.

[0089] The adaptive denoising engine has a built-in denoising threshold mapping table corresponding to different drone models. It uses a density clustering algorithm to adapt the threshold based on the point cloud acquisition accuracy and performs differentiated point cloud denoising processing.

[0090] Furthermore, the adaptive denoising engine incorporates a built-in denoising threshold mapping table categorized by device type. Within this table, entries are divided according to different inspection drone models, with each model's entry associated with a matching point cloud acquisition accuracy threshold corresponding to its own LiDAR hardware performance. It receives inspection point cloud data after global coordinate alignment via a cross-drone registration interface. Based on the drone model identifier field attached to the inspection data, it retrieves the corresponding matching threshold entry from the denoising threshold mapping table. The built-in density clustering algorithm is then loaded, using spatial point clustering characteristics as the criterion to perform unit clustering on the entire point cloud, distinguishing between dense points with solid outlines and floating, discrete interference points. According to the matched model-specific matching thresholds, the clustered point cloud units are further stratified, retaining effective dense point cloud units that conform to the shape of the infrastructure and the outline of tree branches and crowns, while eliminating invalid discrete units formed by atmospheric background reflection, floating noise in the air, and scattered irrelevant points on the ground. Differentiated contained denoising logic is then executed based on the drone model.

[0091] The incremental fusion storage unit is used to partition and cache the denoised inspection data. Based on the geographical location information constraints of the inspection area, it performs hierarchical incremental fusion and data version traceability.

[0092] Furthermore, the incremental fusion storage unit undertakes three core functions: partitioned caching of denoised inspection data, hierarchical incremental fusion, and data version traceability. Using the geographical location information formed by the initial inspection area division as boundary constraints, it caches and archives the registered and denoised inspection data by region. Independent partitioned cache spaces are defined, respectively containing point clouds and image content belonging to the first and second inspection data according to the inspection area boundaries, maintaining the isolation and orderly storage of data in each region. It retrieves the geographical location information corresponding to each inspection area and the overlapping range information of adjacent areas, prioritizing the first inspection data as the baseline data source for overlapping areas, with the second inspection data only used to supplement edge details. A smooth transition method is used to complete the stitching of point locations and image textures in overlapping areas. Progressive incremental merging is performed region by region along the overall channel direction, gradually integrating and generating a complete and continuous fused inspection dataset. Simultaneously, time and device identifiers are bound to each fusion process, establishing a complete data version traceability link, recording the source composition and processing history of each batch of fused data.

[0093] The format conversion unit is used to convert the fused inspection dataset into a standard format that conforms to the access specifications of the modeling and recognition update module.

[0094] Furthermore, adhering to the data access specifications and parsing standards preset by the modeling and identification update module, a fixed point cloud layout structure, image encapsulation format, rules for mounting auxiliary attribute fields, and directory hierarchy organization are defined. The fused inspection dataset output by the incremental fusion storage unit is read, and the dataset is disassembled to include 3D point clouds of channel facilities, 3D point clouds of vegetation, aerial image data, and various auxiliary time-series parameter identifiers. The internal data organization structure is reconstructed according to the parsing requirements of the backend modeling business, unifying and standardizing the point cloud storage arrangement and image compression encapsulation format, and simultaneously standardizing the associated mounting positions of auxiliary fields such as GNSS positioning, IMU attitude, acquisition parameters, and device identifiers. The field naming system, index association logic, and hierarchical directory structure of various data types are uniformly aligned, enabling the converted overall data to be directly recognized, read, parsed, called, and differentially processed by the modeling and identification update module. After format conversion, a standardized and uniform data format is output, meeting the full-process access requirements for subsequent channel 3D model accuracy verification, topology reconstruction, and hazard labeling updates.

[0095] The modeling and recognition update module is used to acquire historical standard inspection datasets and update the channel 3D model based on channel business security requirement data, historical standard inspection datasets, and standard inspection datasets.

[0096] Specifically, the modeling and recognition update module is configured with a model update control bus, a differential comparison engine, a 3D topology reconstruction unit, and a multi-dimensional verification unit; among which:

[0097] The model update control bus establishes hardware-level linkage links with the format conversion unit of the heterogeneous data preprocessing module, the data service layer, and the operation and maintenance closed-loop module, respectively, for timing synchronization and priority control of data interaction.

[0098] Furthermore, the model update control bus simultaneously interfaces with the format conversion unit within the heterogeneous data preprocessing module, the backend data service layer, and the frontend operation and maintenance closed-loop module, establishing a dedicated hardware-level linkage link at the physical layer and opening up the transmission channel for underlying data and control commands between modules. It unifies and coordinates the runtime sequence of all cross-module data and command interactions, defining the sequential rhythm of standard inspection dataset input, historical standard inspection dataset retrieval, model reconstruction parameter distribution, verification signal feedback, log data writing, and operation and maintenance log reading. It divides different business operations and data interactions into priority levels, prioritizing high-precision hardware operations such as point cloud differential comparison and 3D topology reconstruction, while adapting routine data reading and writing, log retrieval, and other business operations to the appropriate timing window. It completes the distribution of control commands and the transparent transmission of business data according to the inherent communication protocol of the hardware link, standardizing the operating rhythm and data transmission and reception sequence of each unit, allowing all related modules to work collaboratively according to a unified timing rhythm.

[0099] The differential comparison engine has a built-in point cloud data cache and differential feature extraction circuit. It is used to cache the standard inspection dataset output by the format conversion unit and the historical standard inspection dataset called by the data service layer. It supports parallel differential comparison of the standard inspection dataset and the historical standard inspection dataset through hardware logic circuit.

[0100] Furthermore, the differential comparison engine integrates an independently deployed point cloud data cache and a dedicated difference feature extraction circuit. The point cloud data cache is divided into isolated storage partitions, receiving standard inspection datasets that have been regularized and output by the format conversion unit, and historical standard inspection datasets loaded from the data service layer. The two sets of point cloud data from different time periods are temporarily latched and partitioned. A parallel computing path is built using built-in dedicated hardware logic circuits, simultaneously loading the complete set of point cloud resources for the current and historical time periods from the cache. The registration and alignment process for the two sets of point cloud data is completed independently within the hardware circuit. According to the hardware's fixed computing logic, voxel decomposition and unit-by-unit content comparison are performed, sequentially identifying features such as differences in spatial point distribution, changes in entity outline morphology, and changes in vegetation spatial occupancy. Scattered difference units are clustered and regularized according to spatial connectivity characteristics, generating a complete channel spatial point cloud difference feature set. All comparison operations are executed in parallel using the hardware circuits throughout the process, without relying on upper-layer software process scheduling, achieving high-speed extraction of temporal point cloud difference features.

[0101] The 3D topology reconstruction unit integrates local reconstruction control circuits and parameter storage registers. It is used to complete vertex correction and model reconstruction in different regions based on the accuracy threshold constraints in the channel service security requirement data. Simultaneously, it stores the changes in point cloud geometric parameters during the reconstruction process into the registers.

[0102] Furthermore, the 3D topology reconstruction unit integrates the built-in local reconstruction control circuit and dedicated parameter storage register. First, it extracts the accuracy threshold of the channel facility point cloud from the channel service security requirement data, loading the accuracy constraints into the local reconstruction control circuit as the execution benchmark for the reconstruction operation. It receives the boundary information of the difference region set output by the differential comparison engine, locks the spatial range requiring reconstruction correction, and calls the real point cloud positions within the standard inspection dataset as the basis for morphological replication. Following the hardware-level fixed topology reconstruction logic, it fine-tunes the position and corrects the coordinates of the model mesh vertices one by one within the defined difference region. Based on the actual point cloud contour trend, it re-organizes the connection relationships between mesh nodes, reshaping the local spatial topology architecture to adapt to model morphological changes caused by minor deformations of infrastructure, tree branch extensions, and crown boundary expansion. It statistically analyzes the point position offset, contour changes, and spatial morphological changes before and after reconstruction for each region, summarizing them into a complete set of point cloud geometric parameter changes, which are then uniformly written into the parameter storage register for latching and retention.

[0103] The multi-dimensional verification unit is equipped with a precision verification circuit and a spatial compliance verification circuit. It is used to read the precision threshold of the point cloud of the channel facility and the spatial distance constraint of the tree obstacle, and to perform parallel dual verification on the reconstructed local model. If the verification passes, a model write-back permission signal is generated; if the verification fails, a graded anomaly signal is generated, and the anomaly type and anomaly area are marked simultaneously.

[0104] Furthermore, the multi-dimensional verification unit is equipped with independently operating accuracy verification circuits and spatial compliance verification circuits. These two types of circuits operate independently and in parallel, respectively undertaking the tasks of verifying model fitting accuracy and verifying the compliance of tree barrier spatial spacing. The unit reads the pre-stored accuracy threshold of the point cloud of the access facility and sends it to the accuracy verification circuit. Using this as a benchmark, it performs fit checks on each piece of the topologically reconstructed local model, verifying the model's degree of restoration and fit of the infrastructure shape and tree 3D contours, as well as the completeness of details. Simultaneously, it reads the pre-stored tree barrier spatial distance constraints and sends them to the spatial compliance verification circuit. It searches the spatial separation status of the updated tree 3D shape and surrounding access infrastructure, checking the proximity of vegetation extension structures to the facility framework. Both verifications proceed in parallel. Once all verification content meets the constraint standards, a model write-back allow signal is output. If the constraint standards are not met, a graded anomaly signal is generated based on the degree of deviation, simultaneously marking the anomaly category and its spatial location. Key information elements generated throughout the entire model update process are collected sequentially, including the spatial coordinate range of the identified difference areas, the changes in point cloud geometric parameters stored in the parameter storage register, the update timestamps corresponding to the model update behavior, the final verification judgment results output by the multi-dimensional verification unit, and the data source identifiers that distinguish between new and old inspection data sources. All types of information are structured and organized according to a preset log solidification format, and each entry is entered into the dedicated log partition of the data service layer through a hardware direct connection path using a low-level solidification writing method, generating standardized update logs with tamper-proof characteristics. A fixed call interface is provided, allowing the operation and maintenance closed-loop module to retrieve complete log entries at any time. Based on the regional coordinates, parameter change amplitude, time stamps, and verification conclusions recorded in the logs, the operation and maintenance closed-loop module is assisted in performing conditional screening of inspection areas within the second region set, completing target area delineation and inspection strategy upgrade determination.

[0105] The operation and maintenance closed-loop module is used to determine the target area based on the updated channel 3D model and inspection upgrade rules, update the tree obstacle inspection strategy of the target area from the second channel tree obstacle inspection strategy to the first channel tree obstacle inspection strategy, and perform a second inspection on the target area.

[0106] Specifically, the operation and maintenance closed-loop module includes: a model log access unit, an upgrade rule parsing unit, a target area identification unit, and a policy switching and secondary inspection scheduling unit, wherein:

[0107] The model log access unit establishes a hardware interaction path between the operation and maintenance closed-loop module, the modeling and identification update module, and the data service layer. It proactively accesses the full data of the updated channel 3D model and retrieves complete update log content. The update log internally collects key information such as the spatial range of difference areas, changes in point cloud geometric parameters, update time markers, verification conclusions, and data source attribution markers. It synchronously associates and binds the updated tree 3D data parameters and hazard labeling data refresh content corresponding to each inspection area. It only selectively retrieves model data and log records related to all inspection areas previously included in the second area set, excluding area information from the first area set, thus reducing the scope of invalid data traversal. It organizes and archives the accessed model structure parameters, tree morphological change parameters, spatial spacing change information, and log traceability entries according to the inspection area dimension, ensuring consistency between the accessed data and the model update version timeline, and complete field associations.

[0108] The upgrade rule parsing unit has a built-in, fixed-storage inspection upgrade rule knowledge base. This knowledge base organizes complete strategy upgrade judgment logic in a structured entry format, setting trigger conditions only for inspection areas within the second region set. It breaks down the judgment dimensions within each inspection upgrade rule, using tree height increment changes, crown expansion trends, narrowing spatial distance between trees and access facilities, increased risk levels within hazard labeling data, and growth trend prediction results encroaching on the framework as core judgment items. It systematically organizes the matching logic and hierarchical relationships of various judgment items, establishing a judgment mechanism where any single dimension satisfying a rule item triggers a region upgrade, while simultaneously locking the first region set from rule matching screening. The parsed inspection upgrade rule logic is cached one by one in the internal rule register, ensuring that each rule item can be called and compared at any time.

[0109] The target area identification unit uses data from each inspection area in the second region set cached by the model log access unit. It retrieves the updated position offset, height increment, crown width change rate, and spatial distance change from access facilities for each tree within the corresponding area, simultaneously matching and retrieving the updated growth trend prediction results and risk level information. It then calls each rule entry cached by the upgrade rule parsing unit, using a single inspection area as the smallest discrimination unit, and verifies the parameter changes of all trees within that area against each rule. Inspection areas exhibiting accelerated vertical growth, outward crown extension, continuously decreasing distance from access facilities, or growth trends pointing towards the facility structure and upward changes in risk level are identified. Inspection areas meeting any rule trigger condition are spatially marked and their scope locked. All marked areas are then systematically grouped according to spatial boundaries, forming a target area set with clear attributes and complete boundaries. The entire screening process is completed only within the second region set, avoiding meaningless repetitive judgment processes.

[0110] The strategy switching and secondary inspection scheduling unit receives the target area set spatial boundary information output by the target area identification unit, retrieves the original bound configuration of the second-channel tree obstacle inspection strategy for the target area, and decouples the conventional data acquisition parameters from the corresponding general-purpose inspection UAV model. It retrieves the standard configuration template of the first-channel tree obstacle inspection strategy from the preset strategy template library, batch-adapts the encrypted data acquisition parameter system for the target area, and matches it with dedicated inspection UAV models suitable for tall vegetation, undulating terrain, and close-range fine scanning conditions. It re-plans the dedicated inspection flight route and flight path coverage according to the target area spatial boundary, matching the flight operation mode and onboard payload start / stop sequence corresponding to the encrypted data acquisition. It generates an updated strategy command stream, which is sent to the collaborative acquisition module via the hardware bus. The secondary inspection operation process is initiated within the target area according to the new encrypted acquisition standard, fully acquiring the original data of the channel facility 3D point cloud, vegetation 3D point cloud, and supporting positioning attitude imagery, completing the strategy switching and closed-loop verification acquisition process from ordinary standard inspection to refined encrypted inspection.

[0111] The data service layer is used to store the channel's 3D model, historical standard inspection datasets, channel business security requirements data, partitioning rules, target tree determination conditions, inspection area overlap parameters, inspection upgrade rules, and model update logs.

[0112] Specifically, the data service layer includes: a 3D model storage and management unit, a historical inspection data archiving unit, a business security requirements library unit, a unified rule operation and maintenance unit, and a log library unit, among which:

[0113] The 3D model storage management unit is specifically designed to handle the entire process of centralized archiving, version management, interface calls, and incremental write-back of channel 3D models. It hierarchically organizes infrastructure data, tree 3D data, and hazard annotation data attached to the tree 3D data using a unified structured directory architecture. A hierarchical and domain-based storage organization method is adopted, dividing the space into independent storage partitions according to the inspection area, and archiving the model topology, point geometry information, and attribute annotation fields for each corresponding area. A standardized data retrieval interface is provided to respond to model retrieval requests initiated by the inspection strategy planning module, modeling and identification update module, and operation and maintenance closed-loop module, retrieving the complete model content of the corresponding level as needed according to the area range. It receives new version model data after verification by the modeling and identification update module, partially overwrites the model content of different areas using an incremental update mechanism, retains historical version links without overall overwriting, synchronously maintains the model version time-series hierarchy, and supports multi-module time-sharing reuse, on-demand retrieval, and iterative write-back.

[0114] The historical inspection data archiving unit centrally collects historical standard inspection datasets generated from each operation. It establishes a multi-dimensional classification and archiving catalog system based on acquisition time sequence, operation area, and equipment model, systematically storing complete inspection data for each batch after registration, denoising, fusion, and format standardization. Independent storage spaces are allocated according to data type, retaining 3D point clouds of passage facilities, 3D point clouds of vegetation, aerial imagery, and various auxiliary positioning attitude and acquisition parameter identification information, ensuring that the field structure of each set of historical standard inspection datasets is completely consistent with the currently generated standard inspection dataset. Responding to the parallel retrieval needs of the differential comparison engine in the modeling and recognition update modules, it loads corresponding historical datasets by time sequence batch and regional range, supporting direct reading and caching into the computing unit by hardware circuitry. It synchronously accepts log-related data generated from each model update, establishing a correlation index between historical inspection data and model update logs, facilitating subsequent time-series differential comparison, version tracing, and anomaly backtracking retrieval, maintaining the orderly retention and controllable access of inspection data throughout its entire lifecycle.

[0115] The Business Security Requirements Library unit specifically stores channel business security requirements data, dividing it into independent entries and partitions according to constraint categories. It systematically organizes three core constraint categories: point cloud accuracy thresholds for channel facilities, spatial distance constraints for tree obstacles, and accuracy conditions triggering model updates. Each control constraint is defined in a structured logical entry format, specifying its applicable scenarios and distinguishing the applicable boundaries for infrastructure modeling accuracy control, spatial spacing control between trees and channel facilities, and model partial reconstruction trigger condition control. A fixed read interface is provided, allowing the modeling and recognition update modules to retrieve corresponding constraint entries as needed during topology reconstruction, accuracy verification, and spatial compliance verification phases. Constraint entries are kept fixed and cannot be arbitrarily rewritten; only overall version iteration updates are allowed, and temporary local modifications are not permitted. This provides a unified and unchanging benchmark for point cloud reconstruction accuracy control, spatial spacing compliance judgment, and differential region screening triggers, providing standardized and directly matchable constraint rule sources for hardware computing units throughout the entire process.

[0116] The unified operations and maintenance unit centrally collects various logical rules and control parameters, including zoning rules, target tree determination conditions, inspection area overlap parameters, and inspection upgrade rules. Independent rule storage partitions are created based on business module affiliation, corresponding to the business needs of the inspection strategy planning module, heterogeneous data preprocessing module, and operations and maintenance closed-loop module. Zoning rules, target tree determination conditions, and inspection area overlap parameters are organized into structured logic that can be matched line by line, allowing the inspection strategy planning module to complete channel zoning, target tree marking, and delineation of adjacent area overlap ranges. Inspection upgrade rules are fixed into entries based on determination dimensions, setting trigger logic such as tree morphological changes, crown expansion, spatial spacing narrowing, and risk level changes only for the second area set, allowing the operations and maintenance closed-loop module to compare and screen target areas line by line. A unified standard interface is provided for rule retrieval, entry parsing, and logic matching, ensuring all business modules share the same baseline rule system and avoiding inconsistencies in rule versions across modules. Simultaneously, a correlation index is built between rules and operations and maintenance logs, supporting a fully rule-driven, traceable, and reusable business operation mode.

[0117] The log library unit is the sole fixed storage unit for model update logs, solely responsible for log storage and not performing log analysis. It employs a hardware-level fixed writing mechanism, establishing a direct physical link with the modeling and identification update modules. It receives verified update log data and writes it using a low-level fixed method, ensuring the logs are tamper-proof, non-deletable, and unmodifiable. Log fields are fixed and include the coordinate range of the difference region, changes in point cloud geometric parameters, update timestamp, verification result, data source identifier, version number, and operator identifier. The fields are complete, formatted uniformly, and sequentially ordered. Built-in time-series, spatial, and version indexes are used to construct multi-dimensional indexes based on update time, difference region, and model version. This supports fast retrieval, precise location, and batch export by time range, region range, version number, and verification result. A standardized log reading interface is provided externally to respond to log retrieval requests from the operation and maintenance closed-loop module and the modeling and identification update module. It supports sequential reading, batch reading, cached reading, and direct hardware-level reading, providing complete log support for target region determination, model backtracking, and anomaly tracing.

[0118] The technical solution of this invention has the following advantages:

[0119] The real-time channel obstacle modeling system based on point cloud data provided in this embodiment first establishes a fixed interaction link between the inspection strategy planning module and the underlying data service layer, streamlining the internal business scheduling logic of the system, improving the overall smoothness of task planning, and reducing the obstruction of instruction interaction between modules. Secondly, the collaborative acquisition module, based on a terminal scheduling architecture and UAV cluster onboard hardware, coordinates the operation sequence and task allocation of multiple devices, mitigating the impact of inconsistent operation paces between devices and ensuring continuous and orderly on-site acquisition. Next, the heterogeneous data preprocessing module, based on its built-in processing unit, handles multi-source inspection data, relying on its own hardware processing capabilities to complete the entire process of registration, denoising, and fusion, alleviating the computational pressure on the main system business and improving the throughput of large-volume point cloud image data processing. Furthermore, the modeling and recognition update module, based on its own computing hardware, handles time-series comparison and 3D reconstruction, independently completing model difference analysis and local reconstruction calculations without consuming resources from other system modules, improving the operational efficiency of channel 3D model iteration and refresh, and making the model state update process more stable and reliable. Then, a business linkage path is formed through the operation and maintenance closed-loop module and the front-end acquisition and back-end modeling modules. Based on the system's inherent transmission link, area determination, strategy adjustment, and secondary inspection task distribution are completed, improving the tightness of the entire operation and maintenance business link. Finally, the data service layer adopts a layered storage architecture to carry all kinds of business data of the entire system, providing unified data reading and writing support for each functional module, standardizing the data access order of the entire system, improving the adaptability and capacity when multiple modules retrieve data simultaneously, and avoiding operational delays caused by centralized data access. By implementing this solution, the significant shortcomings of related channel tree obstacle inspection and modeling technologies in terms of the rationality of inspection area planning, adaptability of acquisition operation mode, accuracy of multi-source acquisition data processing, timeliness of channel 3D model updates, and the continuity and dynamic adjustment capability of inspection and maintenance processes have been resolved. These shortcomings make it difficult to adapt to the high-precision, real-time, and refined tree obstacle inspection and modeling operation and maintenance needs of large-scale linear infrastructure channels.

[0120] In this embodiment, the real-time channel tree obstacle modeling system based on point cloud data is presented in the form of functional units. Here, a unit refers to an ASIC (Application Specific Integrated Circuit), a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

Claims

1. A method for real-time modeling of channel tree obstacles based on point cloud data, characterized in that, The method includes: Based on the 3D model of the channel and the partitioning rules, the inspection channel is partitioned to obtain a first region set and a second region set. A first channel tree obstacle inspection strategy and a second channel tree obstacle inspection strategy are configured for the first region set and the second region set, respectively. The first channel tree obstacle inspection strategy is an encrypted collection strategy for the first region set, and the second channel tree obstacle inspection strategy is a standard collection strategy for the second region set. Execute the first channel tree obstacle inspection strategy and the second channel tree obstacle inspection strategy to obtain the first inspection data and the second inspection data; The first and second inspection data are subjected to cross-machine registration, point cloud adaptive denoising, multi-machine point cloud incremental fusion and data format standardization to generate a standard inspection dataset. Obtain historical standard inspection datasets, and update the channel 3D model based on the channel service security requirement data, the historical standard inspection datasets, and the standard inspection datasets; Based on the updated 3D model of the channel and the inspection upgrade rules, the target area is determined, the tree obstacle inspection strategy of the target area is updated from the second channel tree obstacle inspection strategy to the first channel tree obstacle inspection strategy, and a second inspection is performed on the target area.

2. The method according to claim 1, characterized in that, The 3D model of the passageway includes infrastructure data, 3D tree data, and hazard annotation data attached to the 3D tree data. The hazard annotation data includes the location, height, crown width, distance from passageway facilities, growth trend prediction results, and risk level of each tree. The passageway is divided into zones based on the 3D model and zoning rules, resulting in a first set of zones and a second set of zones, including: Analyze the hazard labeling data in the 3D model of the passageway and mark trees with a risk level higher than the preset level as target trees; Based on the constraints of the coverage area of ​​the drone inspection operation, the tree location, height, crown width, distance from the passage facilities, and growth trend prediction results of each target tree, the inspection passage is divided into several inspection areas, with overlapping areas between adjacent inspection areas. Based on the zoning rules, inspection areas containing the target tree are assigned to the first region set, and inspection areas not containing the target tree are assigned to the second region set.

3. The method according to claim 2, characterized in that, The configuration of the first channel tree obstacle inspection strategy and the second channel tree obstacle inspection strategy for the first region set and the second region set respectively includes: Based on the tree location, height, crown width, distance from passage facilities, and growth trend prediction results of each target tree, encrypted data collection operation parameters and corresponding inspection drones are configured for each inspection area in the first area set to generate the first passage tree obstacle inspection strategy. The encrypted data collection operation parameters include flight altitude, flight path overlap, point cloud sampling frequency, and number of collection viewpoints. Configure unified routine data collection parameters and corresponding inspection drone models for each inspection area in the second area set to generate a second-channel tree obstacle inspection strategy. The routine data collection parameters include flight altitude, flight path overlap, point cloud sampling frequency, and number of collection viewpoints. The point cloud sampling frequency and number of collection viewpoints in the routine data collection parameters are lower than those in the encrypted data collection parameters.

4. The method according to claim 3, characterized in that, Both the first and second inspection data include 3D point clouds of passage facilities, 3D point clouds of vegetation, aerial imagery data, GNSS positioning data, data acquisition time sequence timestamps and corresponding acquisition operation parameters, IMU attitude data, raw LiDAR scan data, and UAV model identifiers. The first and second inspection data undergo cross-machine registration, adaptive point cloud denoising, multi-machine point cloud incremental fusion, and data format standardization to generate a standard inspection dataset, including: Using a registration algorithm based on the GNSS positioning data and IMU attitude data of each inspection drone, the first inspection data and the second inspection data are uniformly registered to the channel global coordinate system; Obtain the denoising threshold corresponding to each inspection drone, and use the density clustering algorithm to denoise the registered first and second inspection data based on the denoising threshold corresponding to each inspection drone. The denoising threshold corresponding to each inspection drone is set based on the point cloud acquisition accuracy. The first and second inspection data after noise reduction are incrementally fused based on the geographical location information of each inspection area to obtain a fused inspection dataset. The fused inspection dataset is converted into a standard format adapted to the channel 3D model update to generate a standard inspection dataset.

5. The method according to claim 4, characterized in that, The step of obtaining the historical standard inspection dataset and updating the channel 3D model based on the channel service security requirement data, the historical standard inspection dataset, and the standard inspection dataset includes: Based on the security requirements data of the channel business, the accuracy threshold of the point cloud of channel facilities, the spatial distance constraint of tree obstacles, and the accuracy conditions for triggering model updates are extracted. Point cloud registration and alignment and voxelization difference calculation are performed on the standard inspection dataset and the historical standard inspection dataset to generate a channel space point cloud difference feature set. Based on the channel spatial point cloud difference feature set and the model update triggering accuracy condition, a set of difference regions is obtained by filtering. Based on the standard inspection dataset, under the constraint of the point cloud accuracy threshold of the channel facilities, local vertex correction and topology reconstruction are performed on each difference region, and the changes in point cloud geometric parameters before and after reconstruction are recorded. Based on the changes in the point cloud geometric parameters, update the infrastructure data and tree 3D data, and calculate the position offset, height increment, crown change rate, and spatial distance change value of each tree from the access facilities. Based on the location offset, height increment, crown width change rate, spatial distance change value between each tree and the passage facility, and the preset tree obstacle risk assessment rules, update the growth trend prediction results and risk level in the hazard labeling data; The updated model is validated for accuracy based on the point cloud accuracy threshold of the access facility, and spatial compliance is validated based on the spatial distance constraint of the tree barrier.

6. A real-time modeling system for channel tree obstacles based on point cloud data, characterized in that, The system includes: an inspection strategy planning module, a collaborative data acquisition module, a heterogeneous data preprocessing module, a modeling and identification update module, an operation and maintenance closed-loop module, and a data service layer; wherein: The inspection strategy planning module is used to call the channel 3D model stored in the data service layer, perform inspection channel partitioning based on the channel 3D model and partitioning rules, obtain a first region set and a second region set, and configure a first channel tree obstacle inspection strategy and a second channel tree obstacle inspection strategy for the first region set and the second region set respectively. The first channel tree obstacle inspection strategy is an encrypted collection strategy for the first region set, and the second channel tree obstacle inspection strategy is a standard collection strategy for the second region set. The collaborative acquisition module includes a cluster of inspection drones and a collaborative scheduling terminal, used to receive and execute the first channel tree obstacle inspection strategy and the second channel tree obstacle inspection strategy, obtain the first inspection data and the second inspection data, and transmit the first inspection data and the second inspection data to the heterogeneous data preprocessing module in real time. The heterogeneous data preprocessing module is used to perform cross-machine registration, point cloud adaptive denoising, multi-machine point cloud incremental fusion and data format standardization on the first inspection data and the second inspection data to generate a standard inspection dataset, and transmit the standard inspection dataset to the modeling and recognition update module. The modeling and recognition update module is used to acquire historical standard inspection datasets and update the channel 3D model based on channel business security requirement data, the historical standard inspection datasets, and the standard inspection datasets. The operation and maintenance closed-loop module is used to determine the target area based on the updated channel 3D model and inspection upgrade rules, update the tree obstacle inspection strategy of the target area from the second channel tree obstacle inspection strategy to the first channel tree obstacle inspection strategy, and perform a second inspection on the target area. The data service layer is used to store the channel's 3D model, historical standard inspection dataset, channel business security requirement data, partitioning rules, target tree determination conditions, inspection area overlap parameters, inspection upgrade rules, and model update logs.

7. The system according to claim 6, characterized in that, The inspection strategy planning module is equipped with a channel 3D model data interface and a strategy distribution bus; wherein: The channel 3D model data interface is used to establish a two-way data interaction link with the data service layer, supporting the calling and writing back of the channel 3D model; The strategy distribution bus is used to synchronously distribute the first channel tree obstacle inspection strategy and the second channel tree obstacle inspection strategy to the collaborative acquisition module in the form of an instruction stream.

8. The system according to claim 7, characterized in that, The collaborative acquisition module is equipped with a multi-machine collaborative scheduling bus, an airborne acquisition unit, and a real-time data feedback link; wherein: The multi-drone collaborative scheduling bus is used to receive the instruction stream issued by the inspection strategy planning module and to perform multi-drone timing coordination, area division coordination and payload working status collaborative management. The airborne acquisition unit integrates a lidar payload, a GNSS positioning module, and an IMU attitude module, and is used to simultaneously acquire and generate 3D point clouds of passage facilities, 3D point clouds of vegetation, aerial image data, GNSS positioning data, IMU attitude data, and raw lidar scanning data. The real-time data feedback link is used to encapsulate the collected first and second inspection data into a business data stream and push it to the heterogeneous data preprocessing module in real time.

9. The system according to claim 8, characterized in that, The heterogeneous data preprocessing module is configured with a cross-machine registration interface, an adaptive denoising engine, an incremental fusion storage unit, and a format conversion unit; wherein: The cross-machine registration interface is used to access the GNSS positioning data and IMU attitude data of each inspection drone, and to uniformly register the first inspection data and the second inspection data to the channel global coordinate system. The adaptive denoising engine has a built-in denoising threshold mapping table corresponding to different drone models. It uses a density clustering algorithm to adapt the threshold based on the point cloud acquisition accuracy and performs differentiated point cloud denoising processing. The incremental fusion storage unit is used to partition and cache the denoised inspection data, and to perform hierarchical incremental fusion and data version tracing based on the geographical location information constraints of the inspection area. The format conversion unit is used to convert the fused inspection dataset into a standard format that matches the access specifications of the modeling and recognition update module.

10. The system according to claim 9, characterized in that, The modeling and recognition update module is configured with a model update control bus, a differential comparison engine, a 3D topology reconstruction unit, and a multi-dimensional verification unit; wherein: The model update control bus establishes hardware-level linkage links with the format conversion unit of the heterogeneous data preprocessing module, the data service layer, and the operation and maintenance closed-loop module, respectively, for timing synchronization and priority control of data interaction. The differential comparison engine has a built-in point cloud data cache and differential feature extraction circuit, which is used to cache the standard inspection dataset output by the format conversion unit and the historical standard inspection dataset called by the data service layer. It supports parallel differential comparison of the standard inspection dataset and the historical standard inspection dataset through hardware logic circuit. The three-dimensional topology reconstruction unit integrates a local reconstruction control circuit and a parameter storage register. It is used to complete the vertex correction and model reconstruction of the difference region through hardware-level topology reconstruction logic based on the accuracy threshold constraints in the channel service security requirement data, and simultaneously store the changes in point cloud geometric parameters during the reconstruction process into the register. The multi-dimensional verification unit is equipped with a precision verification circuit and a spatial compliance verification circuit, which are used to read the precision threshold of the channel facility point cloud and the spatial distance constraint of the tree obstacle, respectively, and perform parallel dual verification on the reconstructed local model. If the verification passes, a model write-back permission signal is generated; if the verification fails, a graded anomaly signal is generated, and the anomaly type and anomaly area are marked simultaneously.

Images