Unmanned aerial vehicle gas inspection system and method based on dynamic route planning

Abstract

This invention discloses a UAV gas inspection system and method based on dynamic route planning, relating to the field of gas inspection technology. The environmental modeling module acquires basic spatial data and building structure data within the inspection area, constructs a three-dimensional spatial environment model, identifies flyable spaces, and generates a flight space constraint model. The dynamic route planning module acquires a list of inspection targets, filters inspection targets based on the list, and generates a set of candidate inspection waypoints. Based on the candidate waypoint set and the flight space constraint model, a UAV inspection path is generated through coverage optimization and path cost optimization. The gas detection module collects and analyzes gas concentration data in real time, identifies abnormal concentrations, and triggers an alarm mechanism. The privacy protection module performs real-time analysis of the collected inspection images, identifies sensitive areas in the images, and automatically occludes or blurs these sensitive areas, achieving both gas safety detection and user privacy protection.

Description

Technical Field

[0001] This invention relates to the field of gas inspection technology, specifically to a UAV gas inspection system and method based on dynamic flight path planning. Background Technology

[0002] With the rapid development of high-rise residential buildings in cities and towns, the scale of residential gas usage is constantly expanding, and the demand for gas safety inspections in high-rise residential buildings is increasing. For vacant units or units that cannot be found during visits, traditional manual in-home inspection methods suffer from difficulties in entry, low inspection efficiency, and limited coverage, making it difficult to detect potential gas leaks in a timely manner. To improve inspection efficiency, in recent years, drones equipped with gas detection equipment have begun to be used for low-altitude non-contact inspections. By inspecting the building facade and window areas, suspected gas leak locations can be quickly identified.

[0003] However, existing drone gas inspection technology still has certain limitations in practical applications. On the one hand, most inspection drones generally use fixed preset routes to perform inspection tasks, which cannot adapt to the three-dimensional structure of high-rise buildings, obstacles between buildings, and the dynamic distribution of vacant households. This easily leads to blind spots and redundant inspection paths, thus affecting inspection efficiency and coverage. On the other hand, gas inspection data is disconnected from the operation and maintenance management system. Dynamic route planning lacks an effective coordination mechanism with gas laser detection, privacy protection, and gas operation and maintenance management. This results in insufficient detection accuracy, lack of protection for user privacy, and inability to process inspection data in a closed loop, failing to meet the requirements of efficiency, safety, and compliance for gas leak inspection in high-rise residential buildings. Summary of the Invention

[0004] The purpose of this invention is to provide a UAV gas inspection system and method based on dynamic route planning. By constructing a dynamic route planning mechanism for UAV gas inspection and combining gas detection data processing and user privacy protection strategies, the UAV gas inspection task can adaptively plan routes according to the structure of high-rise buildings and the distribution of vacant households, thus achieving the dual requirements of gas safety detection and user privacy protection.

[0005] The technical solution to achieve the objective of this invention is as follows:

[0006] On the one hand, the present invention provides a UAV gas inspection system based on dynamic flight path planning, including an environmental modeling module, a dynamic flight path planning module, a gas detection module, a privacy protection module and a task management module, and the modules work together.

[0007] The environmental modeling module is used to acquire basic spatial data and building structure data within the inspection area, construct a three-dimensional spatial environment model, identify flyable spaces based on the three-dimensional spatial environment model, and generate a flight space constraint model.

[0008] The dynamic route planning module obtains the list of inspection targets, filters the inspection targets based on the list and generates a set of candidate inspection waypoints. Based on the set of candidate inspection waypoints and the flight space constraint model, it generates the UAV inspection path through coverage optimization and path cost optimization.

[0009] The gas detection module collects and analyzes gas concentration data in real time during the drone's flight along the inspection route, identifies abnormal concentrations, and triggers an alarm mechanism.

[0010] The privacy protection module performs real-time analysis on the collected inspection images, identifies sensitive areas in the inspection images, and automatically occludes or blurs the sensitive areas.

[0011] The task management module is used for creating, issuing, monitoring the status of, archiving, and handling of hidden dangers of inspection tasks, and synchronously receives and stores inspection data uploaded by drones.

[0012] Furthermore, the environment modeling module includes an information acquisition unit, a modeling unit, and a spatial analysis unit;

[0013] The information acquisition unit is used to collect basic spatial data and building structure data of high-rise residential buildings within the inspection area, and to preprocess and coordinate system 1 to obtain basic environmental dataset;

[0014] Specifically, by accessing the city's geographic information system (GIS) platform or retrieving building survey drawings, information such as building geographic coordinates, building plan outline data, and floor distribution is obtained. Furthermore, a drone equipped with a lidar performs a pre-scan flight over the inspection area, collecting three-dimensional point cloud data of the building surface through the onboard lidar to supplement detailed information on the building facade outline, concave and convex structure information, and obstacle information. Preprocessing operations such as noise reduction, missing data completion, duplicate data filtering, and smoothing are performed on the multi-source data to obtain a basic environmental dataset.

[0015] The modeling unit is used to construct a three-dimensional spatial environment model of the inspection area based on the basic environmental dataset;

[0016] The building plan outline and building height parameters are extracted from the basic environmental dataset. Boundary points are extracted from the building plan outline, and the boundaries of the building plan outline are smoothed using a polygon outline fitting algorithm. The plan outline is stretched and modeled vertically according to the standard floor height to obtain a 3D geometric model. Point cloud data of the building facade is extracted using a point cloud segmentation algorithm to obtain the building facade point cloud data. A surface fitting algorithm is used to geometrically reconstruct the building facade point cloud to obtain a 3D surface model. The 3D geometric model and the 3D surface model of the same building are spatially registered to obtain the 3D building model of a single building.

[0017] Furthermore, a point cloud segmentation algorithm is used to extract 3D point clouds of environmental elements, and a land feature classification algorithm is used to identify environmental elements in the inspection area. A geometric fitting algorithm is used to perform 3D modeling of different types of environmental elements to obtain environmental element models. The 3D models of multiple buildings and environmental element models in the inspection area are spatially combined. A unified spatial relationship is established based on the spatial coordinates of the 3D building models and environmental element models of each building. The 3D models are then integrated using a spatial stitching algorithm to form a 3D spatial environment model.

[0018] The spatial analysis unit is used to identify and analyze the flight space of UAVs based on the three-dimensional spatial environment model, and to construct a flight space constraint model that can be used for route planning.

[0019] Specifically, the spatial boundaries of each building's 3D model and environmental element model are extracted, and a safety buffer zone is constructed outside each building and environmental obstacle based on the UAV's safety distance requirements. A morphological dilation algorithm is used to generate the corresponding safe obstacle avoidance area, resulting in a basic flight space constraint model. Under a unified coordinate system, a voxelization algorithm is used to discretize the 3D space of the inspection area into regular 3D mesh units, with each mesh unit corresponding to a voxel unit. Based on the spatial relationship between each voxel unit and the basic flight space constraint model, the voxel units are labeled with attributes to identify flyable voxels. A 3D 26-domain neighborhood search algorithm is used to identify connected regions of flyable voxels, extracting a continuously distributed set of flyable voxels. Flight path structures that allow UAVs to pass through these regions are extracted, resulting in a 3D flight space constraint model that includes flyable areas and flight path structures.

[0020] Furthermore, the dynamic route planning module includes a planning unit and a dynamic path generation unit;

[0021] The planning unit is used to screen inspection targets, locate the spatial position of inspection targets, and generate a set of candidate inspection waypoints before the UAV takes off.

[0022] Specifically, the inspection user data and historical safety inspection records are retrieved from the gas operation and maintenance platform, and vacant households, households that are not visited, and households that have not been inspected for a long time are extracted as the inspection targets for this inspection, thus obtaining a list of targets to be inspected;

[0023] Based on the building number, unit number, floor number, and apartment number of the inspection target, and in conjunction with a pre-established 3D spatial environment model, the 3D spatial position of each inspection target on the building facade is determined. A preset inspection distance d is set in front of the building facade, and the inspection target is spatially offset along the normal direction of the building facade to generate corresponding inspection waypoints. The above steps are repeated to generate a set of candidate inspection waypoints. k represents the number of candidate waypoints for inspection.

[0024] The dynamic path generation unit is used to generate the optimal UAV inspection path by jointly solving the candidate inspection waypoint set and flight space constraint model through coverage optimization and path cost optimization.

[0025] Specifically, the candidate inspection waypoints are mapped one by one to the flight space constraint model. The voxel attributes of each candidate inspection waypoint are determined. If a candidate inspection waypoint is located in a non-flying voxel or restricted flight voxel area, the nearest flyable voxel is found within a preset range around it using a 3D neighborhood search algorithm. The center point of that voxel unit is used as the corrected candidate inspection waypoint, ensuring that all candidate waypoints are in areas that the UAV can safely reach.

[0026] In the following path planning, the flight space constraint model is used as the spatial boundary of the path search;

[0027] The UAV gas detection module can detect targets within a certain spatial range during flight. Its detection beam spreads in a cone shape, forming an elliptical coverage area on the building facade. Based on this principle, the spatial distance between candidate inspection waypoints and each inspection target is first calculated, and then the detection is performed within a preset effective detection distance. Within the system, based on the horizontal and vertical detection field of view, the horizontal projection width and vertical projection height of the detection area on the building facade are calculated. If the target being inspected is located within this projection area, it is considered a candidate inspection waypoint. It can effectively cover the inspection targets, establish the coverage relationship between candidate inspection waypoints and inspection targets, and select the smallest subset of waypoints that can cover all inspection targets from the candidate waypoint set. Let the candidate inspection waypoints be... The set of inspection targets covered is The set of inspection waypoints is obtained by solving the waypoint coverage optimization problem. , making Where S is the set of inspection targets, the goal is to minimize the size of the set of inspection waypoints. This allows for the coverage of all inspection targets with as few inspection waypoints as possible, thereby reducing the number of times the drone stops and the flight time during the inspection process.

[0028] Preferably, a greedy coverage optimization algorithm is used to solve the problem. The basic idea is to select the waypoint that can cover the most uncovered inspection targets in each iteration and add it to the waypoint set, and update the set of uncovered inspection targets until all inspection targets are covered. This method can obtain near-optimal waypoint coverage results with low computational complexity.

[0029] Determining the optimal set of inspection waypoints Subsequently, UAV inspection path planning is performed using inspection waypoints as path nodes, and the three-dimensional A* path search algorithm is used to solve the problem. Specifically, the UAV takeoff point is used as the starting node and the inspection waypoints are used as the target nodes. A path search grid is constructed in the three-dimensional constrained space. A comprehensive path cost function is constructed in the path planning, taking into account the flight distance and altitude variation factors. The formula for calculating the comprehensive path cost function is as follows:

[0030] ,

[0031] in, It is a comprehensive path cost. It's the cost of flight distance. It is the cost of high change. and These are weighting coefficients, which prioritize paths with shorter flight distances and smaller altitude changes in path planning. The formulas for calculating flight distance cost and altitude change cost are as follows:

[0032] ,

[0033] ,

[0034] in, Let represent the altitude of the i-th inspection waypoint. By optimizing the comprehensive path cost function, the optimal flight path for UAV inspection is obtained, which enables the UAV to reduce the number of inspection waypoints and reduce the overall flight distance and altitude variation while ensuring coverage of all inspection targets, thereby improving the execution efficiency of gas inspection tasks.

[0035] Furthermore, the gas detection module includes a detection unit and an analysis and alarm unit; the gas detection module is mounted on the UAV flight platform and works in conjunction with the dynamic route planning module to perform real-time gas detection during the UAV's inspection flight according to the planned inspection path;

[0036] The detection unit is used to collect gas concentration data in real time. It adopts a projection laser methane detection sensor with a preset maximum effective detection distance of 50m. The current position of the UAV is used as the detection origin. A spatial detection field of view is formed based on the horizontal and vertical detection angles of the sensor. An effective detection projection area is formed on the building facade area within this field of view. After the UAV takes off, the detection unit starts synchronously and continuously collects gas concentration data at a preset sampling frequency.

[0037] The analysis and alarm unit preprocesses and identifies anomalies in the collected gas concentration data. Specifically, it filters and reduces noise in the collected gas concentration data, compares the data with a preset safety threshold to determine if a gas leak exists. When the gas concentration exceeds the preset safety threshold, a local alarm mechanism is triggered, sending a hover verification command to the UAV flight control system. This causes the UAV to hover briefly at its current position. During hovering, the detection results are verified through continuous high-frequency sampling across multiple frames to determine if a persistent gas leak exists. When multiple consecutive samples confirm a gas leak in the current area, the system associates and encapsulates the current detection location coordinates, UAV-captured images, gas concentration data, and corresponding timestamp information to generate an alarm message, which is then uploaded to the task management module in real time via a wireless communication link.

[0038] Furthermore, the privacy protection module includes an identification unit and a processing protection unit; the privacy protection module is deployed on the UAV flight platform, and through a lightweight target detection model, it performs real-time identification and processing of user privacy information during the inspection image data acquisition stage, and completes privacy processing before data transmission and storage, so as to achieve full-process privacy information protection, with a focus on automatically occluding or blurring faces and sensitive information in the inspection images.

[0039] The identification unit collects inspection image data through an airborne vision camera device and identifies privacy-sensitive information in the inspection images frame by frame in real time.

[0040] Specifically, the recognition unit uses a lightweight target detection model to detect each frame of the collected inspection images, identify the face region and the clothing region inside the balcony, and output the corresponding target detection box coordinates. To ensure real-time processing capability during the inspection process, the target detection model runs on the edge computing device of the UAV. The detection results are filtered by a preset confidence threshold, and the detection results with a confidence level higher than the preset confidence threshold are retained and marked as sensitive areas.

[0041] The privacy processing unit is used to automatically occlude or blur detected sensitive areas. Based on the location coordinates of the sensitive areas output by the recognition unit, the privacy processing unit performs privacy protection processing on the corresponding areas in the original inspection image. Specifically, when a face or clothing area inside a balcony is detected, Gaussian blur processing is performed on the area. By smoothing the pixels of the area, the image details are reduced, making the corresponding area unrecognizable. Alternatively, pixel block occlusion can be used to replace the corresponding area with fixed color blocks or low-resolution pixel blocks to obtain privacy-processed inspection image data. The privacy-processed inspection image data is used for subsequent data transmission and storage to reduce the risk of sensitive data exposure in the communication link and ensure that the inspection data uploaded to the task management module does not contain identifiable personal privacy information.

[0042] Preferably, the object detection model adopts a lightweight single-stage object detection network structure, such as a lightweight object detection model based on the YOLO series. The object detection model is obtained by joint training on an open-source face dataset and a balcony clothing image dataset.

[0043] Furthermore, the task management module includes a management unit and a processing unit;

[0044] The management unit is used to manage the entire process of UAV inspection tasks, including task creation, issuance, execution, and archiving. Maintenance personnel can create UAV inspection tasks, enter the inspection area range, retrieve the pre-built 3D spatial environment model and flight space constraint model, and associate them with the inspection target list through the gas maintenance platform interface. After the task configuration is completed, the gas maintenance platform will send the inspection task instructions to the UAV flight control system and related functional modules in real time through interface communication to start the inspection task.

[0045] During the inspection mission, the system continuously receives inspection data uploaded by the drone. When the inspection mission is completed, the management unit automatically archives all types of data generated during the inspection process to form a standardized inspection file. At the same time, it performs statistical analysis on the inspection mission and generates inspection statistical reports.

[0046] The processing unit is used to uniformly receive data generated during the inspection process and execute abnormal linkage handling.

[0047] Specifically, when the processing unit receives alarm information uploaded by the gas detection module, it immediately pops up an alarm prompt on the operation and maintenance platform interface, simultaneously displaying information such as the gas leak location, gas concentration value, detection time, and drone-captured images. It also automatically associates the user information and historical safety inspection records of the corresponding inspection target, generates a gas hazard handling work order, and pushes it to the operation and maintenance personnel's terminal, specifying the handling time limit and handling requirements. Subsequently, the operation and maintenance personnel can go to the site for verification or repair based on the work order information, and provide feedback on the handling results through the operation and maintenance personnel's terminal. The platform automatically updates the gas hazard status based on the feedback information.

[0048] On the other hand, the present invention provides a method for UAV gas inspection based on dynamic flight path planning, comprising the following steps:

[0049] Acquire basic spatial data and building structure data within the inspection area, construct a three-dimensional spatial environment model, identify flyable spaces based on the three-dimensional spatial environment model, and generate a flight space constraint model;

[0050] Create an inspection task on the gas operation and maintenance platform, obtain a list of items to be inspected, filter inspection targets based on the list and generate a set of candidate inspection waypoints, and generate a drone inspection path based on the set of candidate inspection waypoints and the flight space constraint model through coverage optimization and path cost optimization.

[0051] The drone is controlled to perform inspection flights according to the inspection route, and the gas concentration data is collected and analyzed in real time during the flight to identify abnormal concentrations and trigger alarm mechanisms.

[0052] Collect and analyze inspection image data in real time, identify sensitive areas in the inspection images, and automatically mask or blur sensitive areas.

[0053] The gas operation and maintenance platform receives and stores inspection data transmitted by drones and performs closed-loop processing of alarm information.

[0054] Compared with the prior art, the significant advantages of this invention are:

[0055] 1. By constructing a three-dimensional spatial environment model of the inspection area, and performing flight space constraint analysis on this basis, the spatial constraints of the UAV inspection route are realized. Dynamic planning is carried out for inspection targets that are difficult to inspect manually, such as vacant households and households that are not encountered during visits. The inspection coverage and path cost are comprehensively optimized, and the optimal inspection path is automatically generated, effectively filling the blind spots of manual inspection, improving the completion rate and efficiency of inspection tasks, and reducing the time and energy consumption caused by repeated flights and ineffective inspections.

[0056] 2. Integrating gas detection, privacy protection, and task management mechanisms, it enables real-time detection of gas concentration and anomaly alarms during inspections. By identifying and processing sensitive areas in inspection images in real time, it protects user privacy and security. At the same time, combined with the task management module, it provides unified management of the inspection task execution process, inspection data, and hazard handling procedures, realizing closed-loop management from inspection execution, anomaly detection to hazard handling, thereby improving the safety, standardization, and management efficiency of gas inspection work. Attached Figure Description

[0057] Figure 1 A flowchart of a drone-based gas inspection system based on dynamic flight path planning;

[0058] Figure 2 This is a flowchart of a UAV gas inspection method based on dynamic flight path planning. Detailed Implementation

[0059] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0060] This invention discloses a UAV gas inspection system and method based on dynamic route planning, which is fully adaptable to non-contact inspection scenarios of natural gas leaks in vacant units of high-rise residential buildings and users who are not encountered during visits.

[0061] like Figure 1 As shown, this invention discloses a UAV gas inspection system based on dynamic flight path planning, including an environment modeling module, a dynamic flight path planning module, a gas detection module, a privacy protection module, and a task management module, with each module working collaboratively.

[0062] The environmental modeling module is used to acquire basic spatial data and building structure data within the inspection area, construct a three-dimensional spatial environment model, identify flyable spaces based on the three-dimensional spatial environment model, and generate a flight space constraint model.

[0063] The dynamic route planning module obtains the list of inspection targets, filters the inspection targets based on the list and generates a set of candidate inspection waypoints. Based on the set of candidate inspection waypoints and the flight space constraint model, it generates the UAV inspection path through coverage optimization and path cost optimization.

[0064] The gas detection module collects and analyzes gas concentration data in real time during the drone's flight along the inspection route, identifies abnormal concentrations, and triggers an alarm mechanism.

[0065] The privacy protection module performs real-time analysis on the collected inspection images, identifies sensitive areas in the inspection images, and automatically occludes or blurs the sensitive areas.

[0066] The task management module is used for creating, issuing, monitoring the status of, archiving, and handling of hidden dangers of inspection tasks, and synchronously receives and stores inspection data uploaded by drones.

[0067] Furthermore, the environment modeling module includes an information acquisition unit, a modeling unit, and a spatial analysis unit;

[0068] The information acquisition unit is used to collect basic spatial data and building structure data of high-rise residential buildings within the inspection area, and to preprocess and coordinate system 1 to obtain basic environmental dataset;

[0069] In this embodiment, spatial data within the inspection area is collected using a dual-mode approach: retrieval of surveying data combined with on-site real-scene data acquisition. By accessing a city geographic information system (GIS) platform or retrieving architectural survey drawings, information such as building geographic coordinates, building number information, building height data, building plan outline data, and floor distribution is obtained. Furthermore, a drone equipped with a lidar performs a pre-scan flight over the inspection area, acquiring 3D point cloud data of the building surface using the lidar to supplement detailed information on the building facade outline, information on concave and convex structures such as balconies or air conditioners, and information on obstacles such as trees, streetlights, and structures. Preprocessing operations are performed on the multi-source data, including: removing noise points from the original data, completing missing data, filtering and merging duplicate or redundant data, and smoothing the boundaries of the building outline data. Through the above data cleaning and fusion processing, a basic environmental dataset describing the spatial structural characteristics of the buildings in the inspection area is obtained.

[0070] The modeling unit is used to construct a three-dimensional spatial environment model of the inspection area based on the basic environmental dataset;

[0071] Building plan outlines and building height parameters are extracted from the basic environmental dataset. Boundary points are extracted from the building plan outlines, and the boundaries of the building plan outlines are smoothed using a polygon contour fitting algorithm, based on the standard floor height of the building. The planar building outline is stretched and modeled in the vertical direction to obtain a three-dimensional geometric model. For example, the Douglas-Puk thinning algorithm is used to thin and smooth the boundary points of the building planar outline, with the distance threshold set to 0.2m.

[0072] Point cloud data of the building facade is extracted using a point cloud segmentation algorithm to obtain the building facade point cloud data. Based on this, a surface fitting algorithm is used to geometrically reconstruct the building facade point cloud. As an optional implementation method, a three-dimensional surface model of the building facade is generated by using least squares surface fitting or a surface reconstruction method based on triangular mesh, or by using meshing to construct a three-dimensional mesh structure of the building facade to obtain a three-dimensional surface model. For example, the point cloud data of the building facade is extracted using a Euclidean clustering segmentation algorithm with a clustering distance of 0.5m to obtain the building facade point cloud data. Then, the least squares method is used to perform quadratic surface fitting on the building facade point cloud to obtain a three-dimensional surface model of the building facade.

[0073] Since the spatial coordinate systems used by different data sources may differ, it is necessary to perform a unified coordinate system transformation on the multi-source data. As an optional implementation method, the geographic coordinates of the GIS platform are used as the reference coordinate system. The 3D point cloud data obtained by scanning is spatially registered with the reference coordinate system to obtain multi-source data with a unified coordinate system. The 3D geometric model and 3D surface model of the unified building are then fused to obtain the 3D model of a single building. For example, building corners, roof corners, etc. are selected as spatial registration control points. Each building has no less than 4 evenly distributed control points, and the ICP iterative nearest point algorithm is used for registration.

[0074] Furthermore, the environmental elements within the inspection area are identified and modeled. The three-dimensional point cloud of the environmental elements is extracted by the point cloud segmentation algorithm, and then the environmental elements in the inspection area are identified by the ground feature classification algorithm, including obstacles such as trees, street lights, utility poles, billboards and other structures. The geometric fitting algorithm is used to perform three-dimensional modeling of different types of environmental elements to obtain environmental element models. For tree structures, a multi-layer cylindrical combination model can be used for fitting; for pole-shaped structures such as street lights and utility poles, a columnar geometric model can be used for fitting; and for planar structures such as billboards and fences, a planar fitting algorithm can be used to generate the corresponding three-dimensional structural model.

[0075] After obtaining the 3D model of a single building, the 3D models of multiple buildings and environmental elements in the inspection area are spatially combined. A unified spatial relationship is established based on the spatial coordinates of the 3D models of each building and the environmental element models. The 3D models are then integrated through a spatial stitching algorithm to form a 3D spatial environment model that includes building structure information and environmental obstacle information.

[0076] It should be noted that in the above processing steps, when smoothing the building plan outline boundary, extracting the point cloud data of the building facade, and geometrically reconstructing the point cloud of the building facade, the present invention does not limit the polygon contour fitting algorithm, point cloud segmentation algorithm, and surface fitting algorithm used. Conventional polygon contour fitting algorithms, point cloud segmentation algorithms, and surface fitting algorithms in the field can be used.

[0077] The spatial analysis unit is used to identify and analyze the flight space of UAVs based on the three-dimensional spatial environment model, and to construct a flight space constraint model that can be used for route planning;

[0078] Based on the three-dimensional spatial environment model, spatial constraint calculations are performed on the building structures and environmental obstacles in the inspection area. Specifically, the spatial boundaries of each building's three-dimensional model and environmental element model are extracted, and a safety buffer zone is constructed outside each building and environmental obstacle based on the UAV's safety distance requirements. For example, a horizontal safety buffer distance of 0.5-2.0m is set, and a morphological dilation algorithm is used to generate the corresponding safe obstacle avoidance area. Through the above processing, the main building structure and environmental obstacles are transformed into a spatial obstacle area with safety distance constraints, thus obtaining the basic flight space constraint model.

[0079] Under a unified coordinate system, the three-dimensional space of the inspection area is discretized into regular three-dimensional mesh units using a voxelization algorithm. Each mesh unit corresponds to a voxel node. Based on the spatial relationship between each voxel unit and the basic flight space constraint model, the voxel units are marked with attributes. Specifically, when a voxel unit overlaps with the building 3D model / environmental element model, the voxel unit is marked as a non-flying voxel. When the voxel unit is within the safety buffer zone, it is marked as a restricted-flying voxel. When the voxel unit is within the safe flight zone, it is marked as a flyable voxel.

[0080] Further regional connectivity analysis is performed on the three-dimensional voxel space model to identify flyable channels within the inspection area. Specifically, a three-dimensional 26-domain neighborhood search algorithm is used to identify connected regions of flyable voxels, extract continuously distributed sets of flyable voxels, and extract flight channel structures that can be passed through these regions. Through the above steps of spatial constraint calculation, spatial voxel modeling, and flight channel identification, a three-dimensional flight space constraint model containing flyable regions and flight channel structures can be obtained.

[0081] Furthermore, the dynamic route planning module includes a planning unit and a dynamic path generation unit;

[0082] The planning unit is used to screen inspection targets, locate the spatial position of inspection targets, and generate a set of candidate inspection waypoints before the UAV takes off.

[0083] The inspection team retrieved user data and historical safety inspection records from the gas operation and maintenance platform, extracted vacant households, households that were not visited, and households that had not been inspected for a long time as the inspection targets for this task, and filtered out users who had completed manual safety inspections and had no abnormalities, thus obtaining the list of inspection targets for this task.

[0084] Based on the building number, unit number, floor number, and apartment number of the inspection target, and combined with the pre-established three-dimensional spatial environment model, the three-dimensional spatial coordinates of each inspection target on the building facade are obtained. Let the spatial coordinates of the j-th inspection target be denoted as . ,in, , and These represent the position coordinates of the inspection target in the 3D spatial environment model. As an optional implementation method, the corresponding 3D building model is matched according to the building number, and the corresponding building facade area is determined according to the unit number and household number. An inspection distance d is preset in front of the building facade, and the inspection target is spatially offset along the normal direction of the building facade to generate the corresponding inspection waypoint. The calculation formula is as follows:

[0085] ,

[0086] in, Let be the spatial coordinates of the candidate inspection waypoints, and n be the unit normal vector of the building facade, which is calculated from the building facade data in the 3D spatial environment model using a surface differentiation algorithm. Repeat the above steps to generate a set of candidate inspection waypoints. k represents the number of candidate patrol points. This set of candidate patrol points is used to describe the detection positions that the UAV may use when performing gas detection. For example, the value of the preset patrol distance is determined according to the UAV obstacle avoidance safety requirements and the effective detection focal length of the UAV laser methane sensor. The preset patrol distance d is preferably 1m.

[0087] The dynamic path generation unit is used to generate the optimal UAV inspection path by jointly solving the candidate inspection waypoint set and flight space constraint model through coverage optimization and path cost optimization.

[0088] The candidate waypoints are mapped one by one into the flight space constraint model. The voxel attributes of each candidate waypoint are determined. If a candidate waypoint is located in a non-flying voxel or restricted flight voxel area, the nearest flyable voxel unit within 1m around the candidate waypoint is found using a 3D neighborhood search algorithm in this embodiment. The center point of the voxel unit is used as the corrected candidate waypoint to ensure that all candidate waypoints are in areas that the UAV can safely reach.

[0089] After all candidate waypoints have been corrected, the spatial distance and coverage effectiveness with the inspection target are recalculated based on the corrected waypoint coordinates to ensure that the corrected waypoints still meet the detection coverage requirements before proceeding to subsequent coverage optimization and path planning.

[0090] In the following path planning, the flight space constraint model is used as the spatial boundary of the path search;

[0091] Furthermore, the gas detection module on the UAV can detect inspection targets within a certain spatial range around any waypoint on its flight path. The gas detection module emits a detection beam towards the building facade; in this embodiment, its maximum effective detection distance is... The detection beam has a horizontal detection field of view. and vertical detection field of view The beam of light forms a conical diffusion structure in space and an elliptical detection projection area on the building facade. Based on this principle, when the UAV is located at the candidate inspection waypoint... When calculating the spatial distance between the candidate waypoint and the target, the formula is as follows:

[0092] ,

[0093] in, Let be the distance between the i-th candidate waypoint and the j-th inspection target, when satisfying At that time, the projection range of the detection beam on the building facade is calculated based on the detection field of view, and its horizontal projection width is... and vertical projection height They are respectively:

[0094] ,

[0095] ,

[0096] Based on the above projection range, a corresponding detection coverage area is formed on the building facade. The position of the inspection target within this detection coverage area is further determined. When the offset of the inspection target on the projection plane satisfies:

[0097] ,

[0098] in, and Let $\mathbf$ be the offset of the inspection target relative to the center of the projection of the detection beam on the building facade. When the above formula is satisfied, it is considered a candidate inspection waypoint. Capable of inspecting targets To perform effective coverage detection, establish the coverage relationship between candidate inspection waypoints and inspection targets, and select the smallest subset of waypoints from the candidate waypoint set that can cover all inspection targets. Let the candidate inspection waypoints be... The set of inspection targets covered is The set of inspection waypoints is obtained by solving the waypoint coverage optimization problem. , making Where S is the set of inspection targets, the goal is to minimize the size of the set of inspection waypoints. This allows for the coverage of all inspection targets with as few inspection waypoints as possible, thereby reducing the number of times the drone stops and the flight time during the inspection process.

[0099] As an optional implementation, a greedy coverage optimization algorithm is used to solve the problem. The optimal cruise point set is initialized as an empty set, and the set of uncovered inspection targets is initialized as the target inspection set S. Iterative optimization is then performed. The basic idea is to prioritize adding the waypoint that can cover the most uncovered inspection targets to the optimal inspection waypoint set in each iteration, and update the set of uncovered inspection targets until the set of uncovered inspection targets is empty, indicating that all inspection targets have been covered. The iteration then terminates, and the optimal inspection waypoint set is output. This method can achieve near-optimal waypoint coverage results with lower computational complexity.

[0100] Determining the optimal set of inspection waypoints Subsequently, inspection waypoints are used as path nodes for UAV inspection path planning. Since the UAV flight path needs to simultaneously meet the three-dimensional constraint space limit and flight distance optimization requirements, a comprehensive path cost function is constructed in the path planning process and solved using the three-dimensional A* path search algorithm. As an optional implementation, the 3D A* path search algorithm includes: constructing a path search grid within the 3D flight space constraint model with the UAV takeoff point as the starting node and the inspection waypoint as the target node. The grid step size is set to 0.5m according to the UAV fuselage size. Initializing the open list and the closed list, adding the starting node to the open list and setting the comprehensive path cost of the starting node to 0, determining the spatial boundary of the path search as the wingable voxel region in the 3D flight space constraint model, selecting the node with the minimum comprehensive path cost from the open list as the current expansion node, and moving it to the closed list; traversing the wingable neighboring grid nodes with the current node as the center, calculating the comprehensive path cost of each neighboring node; if the neighboring node is not added to the open list, calculating the heuristic function value using the 3D Manhattan distance, updating the node cost, and adding it to the open list; if the neighboring node is already in the open list, comparing the current calculated cost with the original cost, and retaining the path with the smaller cost; when the target inspection waypoint is added to the closed list, completing the current segment path search; sequentially completing the path concatenation of all inspection waypoints in the above manner until the last inspection waypoint is included in the closed list, terminating the iteration.

[0101] Specifically, considering factors such as flight distance and altitude variations, and to avoid redundant flights and invalid detections, the formula for calculating the comprehensive path cost function is as follows:

[0102] ,

[0103] in, It is a comprehensive path cost. It's the cost of flight distance. It is the cost of high change. and These are weighting coefficients, which are jointly calibrated through simulation debugging or field tests based on the characteristics of UAV gas inspection operations. For example, the weighting coefficient for flight distance cost. Weighting coefficients for the cost of high variability In path planning, paths with shorter flight distances and smaller changes in flight altitude are prioritized. The formulas for calculating flight distance cost and altitude change cost are as follows:

[0104] ,

[0105] ,

[0106] in, Represents the altitude of the i-th inspection waypoint and the flight distance cost. The sum of the three-dimensional Euclidean distances between adjacent path nodes, plus the cost of height variation. It is the cumulative value of the absolute value of the height difference between adjacent path nodes; by optimizing the solution of the comprehensive path cost function, the optimal flight path of the UAV inspection is obtained, so that the UAV can reduce the number of inspection waypoints and reduce the overall flight distance and flight altitude change while ensuring coverage of all inspection targets, thereby improving the execution efficiency of gas inspection tasks.

[0107] Furthermore, the gas detection module includes a detection unit and an analysis and alarm unit; the gas detection module is mounted on the UAV flight platform and works in conjunction with the dynamic route planning module to perform real-time gas detection during the UAV's inspection flight according to the planned inspection path;

[0108] The detection unit is used to collect real-time gas concentration data. In this embodiment, the detection unit adopts a projection-type laser methane detection sensor with a preset maximum effective detection distance of 50m. Using the current position of the drone as the detection origin, a spatial detection field of view is formed based on the sensor's horizontal and vertical detection angles. Within this field of view, an effective detection projection area is formed on the building facade. After the drone takes off, the detection unit starts synchronously and continuously collects gas concentration data at a preset sampling frequency. For example, the horizontal detection angle of the projection-type laser methane detection sensor is... The vertical detection angle is Detection sensitivity Response time It can quickly capture subtle changes in gas concentration, meeting the detection needs of real-time inspection by drones;

[0109] The analysis and alarm unit preprocesses and identifies anomalies in the collected gas concentration data. In this embodiment, a sliding window averaging filter method is used to reduce noise in the gas concentration data and remove abnormal interference points. Subsequently, the gas concentration data is compared with a preset safety threshold. When the gas concentration is within the safe range, the current concentration data and corresponding time information are recorded and synchronously transmitted to the task management module for storage. When the gas concentration exceeds the preset safety threshold, a local alarm mechanism is triggered, sending a hover verification command to the UAV flight control system. This causes the UAV to hover briefly at its current position. During the hover, the detection results are verified through continuous multi-frame high-frequency sampling to determine whether there is a persistent gas leak. For example, when the UAV flight control system receives the hover verification command, it hovers at its current position for 3 seconds and samples at a sampling frequency of 10Hz. When the gas concentration of five consecutive samples exceeds the preset safety threshold, a gas leak is confirmed. When a gas leak is confirmed in the current area, the system associates and encapsulates the current detection location coordinates, UAV-captured images, gas concentration data, and corresponding timestamp information to generate alarm information, which is then uploaded to the task management module in real time via a wireless communication link.

[0110] Furthermore, the privacy protection module includes an identification unit and a processing protection unit; the privacy protection module is deployed on the UAV flight platform, and through a lightweight target detection model, it performs real-time identification and processing of user privacy information during the inspection image data acquisition stage, and completes privacy processing before data transmission and storage, so as to achieve full-process privacy information protection, with a focus on automatically occluding or blurring faces and sensitive information in the inspection images.

[0111] The identification unit is deployed on the edge computing device of the UAV flight platform. It collects inspection image data through the airborne vision camera device and identifies privacy-sensitive information in the inspection image frame by frame in real time.

[0112] In this embodiment, the recognition unit focuses on identifying information areas in the inspection images that may involve user privacy, such as face areas and areas where clothes are drying on balconies. The recognition unit uses a lightweight target detection model to perform frame-by-frame detection on the collected inspection image frames, identify face areas and areas where clothes are drying on balconies in the images, and output the corresponding target detection box coordinates. To ensure real-time processing capability during the inspection process, the target detection model runs on the edge computing device of the UAV. The detection results are filtered by a preset confidence threshold, and the detection results with a confidence level higher than the preset confidence threshold are retained and marked as sensitive areas. The image frames that are missed are processed by continuing the sensitive area coordinates of the previous image frame.

[0113] The privacy processing unit is used to automatically occlude or blur detected sensitive areas. In this embodiment, the privacy processing unit performs privacy protection processing on the corresponding area in the original inspection image according to the location coordinates of the sensitive area output by the recognition unit. Specifically, when a face or clothing area inside a balcony is detected, Gaussian blur processing is performed on the area. By smoothing the pixels of the area, the image details are reduced, making the corresponding area unrecognizable. In another embodiment, a pixel block occlusion method can also be used to replace the corresponding area with a fixed color block or a low-resolution pixel block to further enhance the privacy protection effect and obtain the privacy-processed inspection image data. The privacy-processed inspection image data is used for subsequent data transmission and storage to reduce the risk of sensitive data exposure in the communication link and ensure that the inspection data uploaded to the task management module does not contain identifiable personal privacy information.

[0114] Preferably, the object detection model adopts a lightweight single-stage object detection network structure, such as a lightweight object detection model based on the YOLO series. The object detection model is obtained through joint training on an open-source face dataset and a balcony clothing image dataset. The specific training process is as follows: The collected balcony scene images are manually labeled, marking the target bounding boxes for the balcony clothing areas. The labeled balcony clothing image dataset is combined with the open-source face dataset to construct a complete model training dataset. Data augmentation techniques such as random scaling, random cropping, and brightness perturbation are used to enhance the training dataset and improve the model's generalization ability. The enhanced training dataset is divided into a training set, a validation set, and a test set in a 7:2:1 ratio. The training process uses the Adam optimizer, with its parameters set as follows: initial learning rate... Set as The weight decay coefficient is set to The training batch size is 64, and the training is conducted for 150 epochs. An early stopping strategy is adopted to prevent the model from overfitting. When the prediction error of the validation set no longer decreases for 10 consecutive training epochs, the early stopping mechanism is triggered, and the model parameters with the smallest current validation error are saved to obtain the trained target detection model.

[0115] Furthermore, the task management module includes a management unit and a processing unit;

[0116] The management unit is used for the entire process management of drone inspection tasks, from creation, issuance, execution, and archiving.

[0117] In this embodiment, maintenance personnel create drone inspection tasks, input the inspection area range, retrieve the pre-built three-dimensional spatial environment model and flight space constraint model, and associate the inspection target list through the gas maintenance platform operation interface. After the task configuration is completed, the inspection task instructions are sent to the drone flight control system and related modules in real time through the gas maintenance platform interface to start the execution of the inspection task.

[0118] During the inspection process, the system continuously receives inspection data uploaded by the drone. Once the inspection is completed, the management unit automatically archives all types of data generated during the inspection, including gas concentration data, alarm information, drone flight paths, and inspection waypoint information, and associates them with the corresponding task information to form a standardized inspection file. At the same time, the gas operation and maintenance platform automatically calculates key indicators such as task completion rate, hazard detection rate, and inspection coverage, and generates corresponding inspection statistical reports to facilitate task evaluation and management analysis by operation and maintenance personnel.

[0119] The processing unit is used to uniformly receive data generated during the inspection process and execute coordinated actions.

[0120] The processing unit receives alarm information uploaded by the gas detection module in real time and performs anomaly processing on the received data. Optionally, the processing unit can also communicate with the gas operation and maintenance platform through a data interface to realize information exchange between drone inspection data and manual safety inspection data.

[0121] When an alarm message is received from the gas detection module, an alarm notification immediately pops up on the operation and maintenance platform interface, simultaneously displaying information such as the gas leak location, gas concentration value, detection time, and drone-captured images. It also automatically associates the user information and historical safety inspection records of the corresponding inspection target, generates a gas hazard handling work order, and pushes it to the operation and maintenance personnel's terminal, specifying the handling time limit and requirements. Subsequently, the operation and maintenance personnel can go to the site to verify or repair the issue based on the work order information and provide feedback on the handling results through their terminals. The platform automatically updates the gas hazard status based on the feedback information, thereby achieving closed-loop management of the gas hazard handling process.

[0122] like Figure 2 As shown, this invention discloses a method for UAV gas inspection based on dynamic flight path planning, including the following steps:

[0123] Acquire basic spatial data and building structure data within the inspection area, construct a three-dimensional spatial environment model, identify flyable spaces based on the three-dimensional spatial environment model, and generate a flight space constraint model;

[0124] Create an inspection task on the gas operation and maintenance platform, obtain a list of items to be inspected, filter inspection targets based on the list and generate a set of candidate inspection waypoints, and generate a drone inspection path based on the set of candidate inspection waypoints and the flight space constraint model through coverage optimization and path cost optimization.

[0125] The drone is controlled to perform inspection flights according to the inspection route, and the gas concentration data is collected and analyzed in real time during the flight to identify abnormal concentrations and trigger alarm mechanisms.

[0126] Collect and analyze inspection image data in real time, identify sensitive areas in the inspection images, and automatically mask or blur sensitive areas.

[0127] The gas operation and maintenance platform receives and stores inspection data transmitted by drones and performs closed-loop processing of alarm information.

[0128] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A UAV gas inspection system based on dynamic flight path planning, characterized in that, include: The environmental modeling module is used to acquire basic spatial data and building structure data within the inspection area, construct a three-dimensional spatial environment model, identify flyable spaces based on the three-dimensional spatial environment model, and generate a flight space constraint model. The dynamic route planning module obtains the list of inspection targets, filters the inspection targets based on the list and generates a set of candidate inspection waypoints. Based on the set of candidate inspection waypoints and the flight space constraint model, it generates the UAV inspection path through coverage optimization and path cost optimization. The gas detection module collects and analyzes gas concentration data in real time during the drone's flight along the inspection route, identifies abnormal concentrations, and triggers an alarm mechanism. The privacy protection module performs real-time analysis on the collected inspection images, identifies sensitive areas in the images, and automatically occludes or blurs these sensitive areas.

2. The UAV gas inspection system based on dynamic route planning as described in claim 1, characterized in that, The environment modeling module includes an information acquisition unit and a modeling unit; The information collection unit obtains building structure data of the inspection area through urban geographic information system or building surveying data, and performs pre-scanning with UAV equipped with lidar to obtain three-dimensional point cloud data of building facades. After pre-processing and unifying the coordinate system, a basic environmental dataset is obtained. The modeling unit obtains a 3D geometric model and a 3D surface model based on the basic environmental dataset, performs spatial registration, and obtains a 3D architectural model of a single building. The 3D point cloud data of environmental elements is processed to obtain an environmental element model. Multiple 3D building models and environmental element models are spatially combined to obtain a 3D spatial environment model.

3. The UAV gas inspection system based on dynamic flight path planning as described in claim 2, characterized in that... The environment modeling module also includes a spatial analysis unit; Based on the three-dimensional spatial environment model, the spatial analysis unit constructs a safety buffer zone outside each building and environmental obstacle according to the safety distance requirements of UAVs. The three-dimensional space of the inspection area is discretized into regular three-dimensional mesh units using a voxelization algorithm. Each mesh unit corresponds to a voxel unit. According to the spatial relationship between each voxel unit and the basic flight space constraint model, the voxel units are labeled with attributes to identify flyable voxel units. Furthermore, the flight channel structure that can be passed by UAVs is extracted through connectivity analysis to obtain the flight space constraint model.

4. The UAV gas inspection system based on dynamic route planning as described in claim 1, characterized in that, The dynamic route planning module includes a planning unit, which filters inspection targets, locates the spatial positions of the inspection targets, and generates a set of candidate inspection waypoints, specifically: From the user data of the inspection, vacant households, households that were not visited during the visit, and households that have not been inspected for a long time were selected as the targets of this inspection, and a list of targets to be inspected was obtained. Based on the inspection target information and a pre-established three-dimensional spatial environment model, the three-dimensional spatial position of the inspection target on the building facade is obtained. An inspection distance is preset in front of the building facade, and the inspection target is spatially offset along the normal direction of the building facade to generate corresponding inspection waypoints. The above steps are repeated to generate a set of candidate inspection waypoints.

5. The UAV gas inspection system based on dynamic route planning as described in claim 4, characterized in that, The dynamic route planning module also includes a dynamic path generation unit. Under the conditions of the candidate inspection waypoint set and the flight space constraint model, the dynamic path generation unit generates the optimal UAV inspection path through a collaborative solution of coverage optimization and path cost optimization, including: The candidate waypoints are mapped one by one into the flight space constraint model. The positions of the waypoints located in the non-flying voxels or restricted flight voxel regions are corrected to obtain the corrected candidate waypoint geometry. The smallest subset of waypoints that can cover all inspection targets is selected from the corrected candidate waypoint set. The inspection waypoint set is obtained by solving the waypoint coverage optimization problem.

6. The UAV gas inspection system based on dynamic route planning as described in claim 5, characterized in that, After obtaining the set of inspection waypoints, the dynamic route planning unit uses the inspection waypoints as path nodes to perform UAV inspection path planning within the flight space constraint model, including: Using the UAV takeoff point as the starting node and the inspection waypoint as the target node, a path search grid is constructed in the three-dimensional constrained space. In the path planning, a comprehensive path cost function including flight distance cost and altitude change cost is constructed. By minimizing the comprehensive path cost function, the UAV inspection path is obtained by using the three-dimensional A* path search algorithm.

7. The UAV gas inspection system based on dynamic route planning as described in claim 1, characterized in that, The gas detection module includes a detection unit and an analysis and alarm unit, which performs real-time gas detection during the drone's inspection flight according to the planned inspection path. The detection unit performs spatial range detection on the building facade during the drone's flight, and collects real-time gas concentration data of the inspection area. The analysis and alarm unit preprocesses and identifies anomalies in the collected gas concentration data, compares the detection results with preset safety thresholds, and triggers an alarm mechanism when the gas concentration exceeds the preset safety threshold. It controls the drone to perform hovering verification, and after confirming the existence of gas leakage, it associates and encapsulates the current detection location coordinates, inspection images, gas concentration and time information to form alarm information, and uploads it to the task management module.

8. The UAV gas inspection system based on dynamic route planning as described in claim 1, characterized in that, The privacy protection module includes an identification unit and a processing unit, which performs real-time identification and processing of privacy-sensitive information in the acquired images during the inspection image data acquisition phase, including: The recognition unit uses a preset target detection model to perform frame-by-frame detection on the real-time collected inspection images, identify privacy information in the inspection images, and output the corresponding target detection box coordinates. Detection results with confidence scores higher than the preset confidence threshold are retained and marked as sensitive areas. The privacy processing unit automatically occludes or blurs the sensitive areas to obtain privacy-processed inspection image data, which is then used for subsequent data transmission and storage.

9. The UAV gas inspection system based on dynamic route planning as described in claim 1, characterized in that, The system also includes a task management module, which comprises a management unit and a processing unit; The management unit is used for creating, issuing, executing, monitoring and archiving inspection tasks. After the task configuration is completed, it issues the inspection task instructions to the UAV flight control system to start the inspection task. After the inspection task is completed, it automatically archives the inspection data and generates inspection statistics. The processing unit receives data generated during the inspection process and performs abnormal linkage processing. When it receives alarm information uploaded by the gas detection module, it issues an alarm prompt on the operation and maintenance platform interface and generates a gas hazard handling work order. It pushes relevant information to the operation and maintenance personnel's terminal for subsequent handling and updates the gas hazard status based on the handling feedback.

10. A method for UAV gas inspection based on dynamic flight path planning, used to implement the UAV gas inspection system based on dynamic flight path planning as described in any one of claims 1-9, characterized in that, Includes the following steps: Acquire basic spatial data and building structure data within the inspection area, construct a three-dimensional spatial environment model, identify flyable spaces based on the three-dimensional spatial environment model, and generate a flight space constraint model; Create an inspection task on the gas operation and maintenance platform, obtain a list of items to be inspected, filter inspection targets based on the list and generate a set of candidate inspection waypoints, and generate a drone inspection path based on the set of candidate inspection waypoints and the flight space constraint model through coverage optimization and path cost optimization. The drone is controlled to perform inspection flights according to the inspection route, and the gas concentration data is collected and analyzed in real time during the flight to identify abnormal concentrations and trigger alarm mechanisms. Collect and analyze inspection image data in real time, identify sensitive areas in the inspection images, and automatically mask or blur sensitive areas. The gas operation and maintenance platform receives and stores inspection data transmitted by drones and performs closed-loop processing of alarm information.

Images