Aerial survey data processing system for pipeline conflict early warning

Abstract

The application relates to the fields of aerial surveying and mapping and point cloud data processing, in particular to a pipeline conflict early warning aerial surveying data processing system; the system comprises a data analysis module, a voxel parameter generation module, a point cloud down-sampling module and a pipeline conflict detection module; the system synchronizes original point cloud and unmanned aerial vehicle flight state data, dynamically generates voxel parameters and carries out down-sampling; the core of the system is to output simplified point cloud containing a grid scale label; when the grid size is smaller than a preset threshold value, pipeline extraction and clearance distance calculation are carried out in a fine grid area, and pipeline conflict early warning is output; the conventional grid area is exempted from the identification process; the application realizes dynamic adjustment of the down-sampling scale according to the flight state to prevent loss of key features, and based on the grid label, the pipeline is extracted on demand, so that the overall computing load of the system is effectively reduced.

Description

Technical Field

[0001] This invention relates to the field of aerial surveying and point cloud data processing technology, specifically to an aerial surveying data processing system for pipeline conflict early warning. Background Technology

[0002] In existing pipeline inspection point cloud processing based on airborne lidar, downsampling typically employs a globally uniform voxel scale or relies solely on geometric features and local density. This approach has the following drawbacks: The existing method does not include flight status data such as speed, acceleration, and yaw changes of the UAV in the process of determining the downsampling scale, resulting in a serious mismatch between the point cloud processing scale and the complexity of the actual situation. In stable flight areas, global uniform downsampling will generate point redundancy and cause the system's computational load to exceed the preset limit. In complex flight conditions such as deceleration, turning, and detouring, the failure to dynamically adjust the processing scale can easily lead to the smoothing and erasure of key clearance features, which seriously affects the accuracy of subsequent conflict warning judgment. Existing downsampling processes often only output a simplified point cloud set, lacking contextual identifiers related to voxel scale and processing sequence. This forces subsequent detection modules to perform indiscriminate distance calculations across the entire region, making it impossible to precisely focus computational resources on high-risk areas. At the same time, due to the lack of smooth constraints on continuous voxel parameters between adjacent processing cycles, point cloud breaks and mesh boundary flickering are easily generated at the high-to-low resolution switching points, which disrupts the continuity of pipeline target extraction and the stability of net distance calculation. Therefore, there is an urgent need for a point cloud processing and labeling method that can dynamically adapt to changes in flight status in order to solve the problems of key feature loss and wasted computing resources, and improve the real-time performance and accuracy of pipeline conflict early warning. Summary of the Invention

[0003] The purpose of this invention is to provide an aerial survey data processing system for pipeline conflict early warning, addressing the following technical problems: Existing pipeline inspection point cloud processing technology based on airborne lidar has significant shortcomings in dynamically adjusting the downsampling scale and accurately focusing computational resources under complex flight conditions. There is an urgent need for an aerial survey data processing system for pipeline conflict early warning that can dynamically adapt to changes in flight status for point cloud processing and labeling, while also considering feature preservation and resource control. The purpose of this invention can be achieved through the following technical solutions: The system includes a data parsing module, a voxel parameter generation module, a point cloud downsampling module, and a pipeline conflict detection module. The data parsing module synchronously receives the raw point cloud data stream generated by the lidar and the flight status data output by the UAV flight control system. Using the lidar's scanning shutter time as the reference timestamp, it aligns the flight status data to the reference timestamp and outputs a data combination containing motion state components and point cloud sets. The voxel parameter generation module receives motion state components, generates voxel parameters for the current frame based on the motion state components, and sends the voxel parameters for the current frame along with the corresponding timestamp to the point cloud downsampling module; the point cloud downsampling module receives the point cloud set and the voxel parameters for the current frame, performs downsampling processing on the point cloud set based on the voxel parameters for the current frame, and outputs the downsampled simplified point cloud as well as the grid scale label and time label bound to the simplified point cloud; The pipeline conflict detection module receives a simplified point cloud, a grid scale label, and a time label. When the voxel size represented by the grid scale label is less than a preset scale threshold, the corresponding area is identified as a fine grid retention area. Pipeline targets are identified in the simplified point cloud of the fine grid retention area, point cloud clusters identified as pipelines are extracted, the net distance between two or more point cloud clusters identified as pipelines is calculated, and a warning is made based on a preset net distance threshold, outputting the pipeline conflict warning result. When the voxel size represented by the grid scale label is greater than or equal to the preset scale threshold, the corresponding area is identified as a regular grid area and the pipeline target identification process is waived.

[0004] Preferably, the data parsing module includes a data receiving interface unit, a time reference extraction unit, and an interpolation and alignment unit. The data receiving interface unit synchronously receives the raw point cloud data stream containing three-dimensional coordinates and reflection intensity generated by the lidar, as well as the flight status data output by the UAV flight control system. The flight status data includes at least instantaneous velocity vector, instantaneous acceleration vector, and instantaneous yaw rate. The time reference extraction unit extracts the scanning shutter time of the lidar as a reference timestamp. The interpolation and alignment unit interpolates and aligns the flight status data based on the reference timestamp to obtain the motion state components aligned to the reference timestamp, and combines the motion state components with the point cloud set corresponding to the reference timestamp to output the data combination.

[0005] Preferably, the voxel parameter generation module includes a state parameter calculation unit, a voxel volume calculation unit, an aspect ratio calculation unit, and a voxel size calculation unit. The state parameter calculation unit receives motion state components, extracts the instantaneous acceleration vector and instantaneous yaw rate, and calculates flight state parameters. The voxel volume calculation unit receives flight state parameters and calculates the basic voxel volume by combining the preset maximum voxel volume and the preset minimum voxel volume. The aspect ratio calculation unit receives the instantaneous velocity vector in the motion state components and calculates the ratio between the flight direction dimension and the cross-sectional feature dimension based on the scalar magnitude of the instantaneous velocity vector as the aspect ratio. The voxel size calculation unit combines the basic voxel volume and the aspect ratio to calculate and output the size parameters along the three coordinate directions as the voxel parameters of the current frame.

[0006] Preferably, the state parameter calculation unit multiplies the scalar of the instantaneous acceleration vector with the first proportional coefficient, and adds the product of the instantaneous yaw rate and the second proportional coefficient to obtain the flight state parameters; the voxel volume calculation unit uses the product of the negative of the adjustment coefficient and the flight state parameters as an exponent to calculate the exponent of the natural constant to obtain the attenuation coefficient; the difference between the preset maximum voxel volume and the preset minimum voxel volume is multiplied by the attenuation coefficient and then added to the preset minimum voxel volume to obtain the basic voxel volume.

[0007] Preferably, the point cloud downsampling module includes a parameter smoothing unit, a mesh generation unit, a representative point extraction unit, and a label attaching unit. The parameter smoothing unit receives the current frame voxel parameters from multiple consecutive frames and smooths them to obtain smoothed voxel parameters. The mesh generation unit receives the smoothed voxel parameters and constructs a local octree mesh based on them. The representative point extraction unit receives a set of point clouds at the same timestamp, assigns the set to the local octree mesh, and extracts representative points from each mesh to obtain a simplified point cloud. The label attaching unit attaches a corresponding mesh scale label and time label to each point cloud block in the simplified point cloud and outputs the result.

[0008] Preferably, the pipeline conflict detection module includes a region screening unit, a topology exclusion unit, a clearance calculation unit, and an early warning judgment unit. The region screening unit receives a simplified point cloud and its bound grid scale label. When the voxel size represented by the grid scale label is greater than a preset scale threshold, it is classified as a low-risk region point cloud cluster; otherwise, it is classified as a high-risk region point cloud cluster. The topology exclusion unit receives the low-risk region point cloud cluster and performs a topology exclusion operation on the low-risk region point cloud cluster based on the spatial connection status between the point cloud clusters as a preset topology relationship. Point cloud clusters that are not connected to other point cloud clusters and whose length is less than a preset length threshold are deleted as point cloud clusters that meet the preset topology exclusion conditions. The clearance calculation unit receives point cloud clusters in high-risk areas, identifies pipeline targets in the point cloud clusters in high-risk areas, and uses nearest neighbor search to calculate the clearance, obtaining the minimum Euclidean distance between any two point cloud clusters identified as pipelines; the early warning determination unit receives the minimum Euclidean distance and combines it with a preset clearance threshold to generate a pipeline conflict early warning result.

[0009] Preferably, the early warning determination unit is set with a preset safe clearance threshold and a preset extreme clearance threshold, and the preset extreme clearance threshold is less than the preset safe clearance threshold; when the minimum Euclidean distance is greater than or equal to the preset safe clearance threshold, a safety warning record is generated; when the minimum Euclidean distance is less than the preset safe clearance threshold but greater than the preset extreme clearance threshold, a first-level pipeline conflict early warning signal is generated; when the minimum Euclidean distance is less than or equal to the preset extreme clearance threshold, a second-level pipeline conflict early warning signal is generated, and a review instruction is output to the external work order system.

[0010] Preferably, the system also includes a parameter adjustment module; the parameter adjustment module receives the frequency record of pipeline conflict early warning results returned by the pipeline conflict detection module, and compares it with the historical conflict records that have been manually verified to obtain the early warning frequency and false alarm rate; when the early warning frequency based on the preset time window and preset flight state interval is greater than the preset early warning frequency threshold, the parameter adjustment module selects at least one of the adjustment coefficient, the first proportional coefficient and the preset minimum voxel volume according to the preset priority to increase the adjustment, otherwise the original parameter value is maintained; When the false alarm rate is greater than the preset false alarm rate threshold, the parameter adjustment module reduces the parameter value of the adjustment coefficient; otherwise, the original value of the adjustment coefficient remains unchanged; and the updated preset minimum voxel volume is less than the preset maximum voxel volume.

[0011] Preferably, the system also includes a result output module, which includes a front-end display interface unit, a report generation unit, and an external interface unit. The front-end display interface unit receives pipeline conflict warning results and flight trajectories generated from flight status data, and identifies and displays the simplified point cloud location and flight trajectory that triggered the pipeline conflict warning results in a 3D map view. The report generation unit integrates the time label, grid scale label, and minimum Euclidean distance value when the pipeline conflict warning results occur to generate a daily pipeline inspection report. The external interface unit pushes the daily pipeline inspection report and the pipeline conflict warning results to an external equipment maintenance platform.

[0012] Preferably, the nearest neighbor search is implemented using the KD tree algorithm combined with spatial partitioning.

[0013] Compared with the prior art, the present invention has the following beneficial effects: This invention aligns flight status data to generate voxel parameters, enabling dynamic adjustment of the downsampling scale based on flight status and avoiding the loss of key clearance features under complex operating conditions. Simultaneously, based on the grid scale labels of the downsampling output, the system eliminates the need for conventional grid region identification processes, performing pipeline extraction and clearance calculation only on the fine-grid retained areas, effectively reducing the overall computational load while ensuring the accuracy of early warnings. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of this application and the prior art, the accompanying drawings used in the description of the embodiments and the prior art will be briefly introduced below.

[0015] Figure 1 This is a block diagram of an aerial survey data processing system for pipeline conflict early warning according to the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0017] like Figure 1 As shown, an aerial survey data processing system for pipeline conflict early warning includes: a data parsing module, a voxel parameter generation module, a point cloud downsampling module, and a pipeline conflict detection module; The data parsing module synchronously receives the raw point cloud data stream generated by the lidar and the flight status data output by the UAV flight control system. Using the lidar's scanning shutter time as the reference timestamp, it aligns the flight status data to the reference timestamp and outputs a data combination containing motion state components and point cloud sets. The voxel parameter generation module generates voxel parameters for the current frame based on motion state components, and sends the voxel parameters for the current frame along with the corresponding timestamp to the point cloud downsampling module. The point cloud downsampling module performs downsampling processing on the point cloud set based on the voxel parameters of the current frame, and outputs a simplified point cloud and a grid scale label and time label bound to the simplified point cloud; The pipeline conflict detection module identifies pipeline targets by performing pipeline target identification on the simplified point cloud of the fine mesh retention area when the voxel size represented by the grid scale label is smaller than the preset scale threshold. It also extracts the point cloud clusters identified as pipelines, calculates the net distance between two or more point cloud clusters identified as pipelines, and outputs the pipeline conflict warning result according to the preset net distance threshold. When the voxel size represented by the grid scale label is greater than or equal to the preset scale threshold, the corresponding area is identified as a regular grid area and the pipeline target identification process is waived.

[0018] The data parsing module includes a data receiving interface unit, a time base extraction unit, and an interpolation and alignment unit; The data receiving interface unit is used to synchronously receive the raw point cloud data stream containing three-dimensional coordinates and reflection intensity generated by the lidar, as well as the flight status data output by the UAV flight control system. The flight status data includes at least the instantaneous velocity vector, instantaneous acceleration vector, and instantaneous yaw rate. The time reference extraction unit is used to extract the scanning shutter time of the LiDAR as the reference timestamp; The interpolation and alignment unit is used to interpolate and align the flight state data based on the reference timestamp to obtain the motion state component aligned to the reference timestamp, and then combine the motion state component with the point cloud set corresponding to the reference timestamp to output the data combination.

[0019] The voxel parameter generation module includes a state parameter calculation unit, a voxel volume calculation unit, an aspect ratio calculation unit, and a voxel size calculation unit; The state parameter calculation unit receives motion state components, extracts the instantaneous acceleration vector and instantaneous yaw rate, and calculates flight state parameters; The voxel volume calculation unit receives flight status parameters and calculates the basic voxel volume by combining the preset maximum voxel volume and the preset minimum voxel volume. The aspect ratio calculation unit receives the instantaneous velocity vector in the motion state component and calculates the ratio between the flight direction dimension and the cross-sectional feature dimension based on the scalar magnitude of the instantaneous velocity vector as the aspect ratio. The voxel size calculation unit combines the basic voxel volume and aspect ratio to calculate and output the size parameters along the three coordinate directions as the voxel parameters of the current frame.

[0020] This embodiment provides an aerial survey data processing system for pipeline conflict early warning. The system takes the raw point cloud data stream output by lidar and the flight status data output by the UAV flight control system as input. First, it establishes a unified time reference based on the scanning shutter time. Then, it generates the voxel parameters of the current frame based on the aligned motion state components. The voxel parameters of the current frame are used to downsample the point cloud set. Finally, it performs pipeline target identification and clearance calculation only on the simplified point cloud represented by the grid scale label as the fine grid retention area, and outputs the pipeline conflict early warning result. This processing chain allows for the direct use of flight state changes to guide downsampling scale adjustments without prior calculation of point cloud geometric complexity, thereby balancing the preservation of key clearance features with overall computational resource load control. Furthermore, based on the above implementation method, the data parsing module, voxel parameter generation module, point cloud downsampling module, and pipeline conflict detection module can be executed sequentially according to the processing cycle within the same computing platform. Instead of using a suspended data jump method between modules, the flow is completed through continuous references to data objects in memory. Specifically, the raw point cloud data stream generated by the lidar can be written into the point cloud buffer in the driver layer, and the flight status data output by the flight control system can be written into the status buffer in the flight control communication thread. After the scanning shutter time is extracted, the data parsing module reads the data segments within the same processing cycle from the above buffer, generates a data combination object after time alignment, and releases the original index after the data combination object is read by the voxel parameter generation module, retaining only the motion state components and point cloud set required for the current cycle to reduce the processing burden caused by repeated copying. Furthermore, based on the above implementation method, the original point cloud data stream and flight state data are acquired. According to the scanning shutter time of the lidar, the original point cloud data stream and flight state data are time-aligned to generate a data combination. Here, the data combination refers to the correspondence between a point cloud set and a motion state component under the same reference timestamp, such as a point cloud set within the same scanning cycle. With the instantaneous velocity vector at that moment Instantaneous acceleration vector and instantaneous yaw rate The pairing results; In specific implementation, the data receiving interface unit synchronously receives the raw point cloud data stream containing three-dimensional coordinates and reflection intensity, as well as flight status data containing at least instantaneous velocity vector, instantaneous acceleration vector, and instantaneous yaw rate; the time reference extraction unit extracts the scanning shutter time from the point cloud data header information as the reference timestamp; the interpolation and alignment unit performs linear interpolation on the flight status data in adjacent processing cycles based on the reference timestamp, so that the velocity, acceleration, and yaw rate all fall on the same time reference; By embedding UAV flight status data into the point cloud processing starting point through time alignment, the correspondence between flight status changes and local situation complexity can be obtained, serving as the basis for subsequent downsampling control; multi-source data alignment is achieved based on a unified time reference to avoid voxel parameter mismatch caused by time sequence misalignment; Furthermore, based on the above implementation method, motion state components are obtained, and voxel parameters are calculated for the motion state components according to the instantaneous acceleration vector, instantaneous yaw rate, and instantaneous velocity vector to generate the current frame voxel parameters; wherein, the current frame voxel parameters refer to the size parameters along the three coordinate directions. , , For example, when the dimension parameter along the flight direction is greater than the dimension parameter in the cross-sectional direction, the flight direction stretching state corresponds to the preset stretching condition; the flight state parameter refers to the comprehensive quantity characterized by the change in acceleration and the change in yaw. For example, the deceleration and circling phase corresponds to the flight state parameter higher than the preset threshold, and the stable straight flight phase corresponds to the flight state parameter lower than the preset threshold. In practical implementation, the state parameter calculation unit first extracts the scalar magnitude of the instantaneous acceleration vector and the absolute value of the instantaneous yaw rate to quantify the degree of UAV maneuver changes within the current processing cycle and provide a basis for scaling the basic voxel volume; correspondingly, the flight state parameters can be calculated using the following formula:

[0021] Among them, a first proportionality coefficient is introduced to adjust the contribution of acceleration changes to the voxel contraction trend. To characterize the ability of longitudinal deceleration or acceleration to indicate the complexity of the situation, a second proportionality coefficient is introduced to adjust the strength of the contribution of yaw rate to the voxel contraction tendency. This characterizes the ability of lateral attitude adjustments, such as detouring and turning, to indicate the complexity of a situation; preferably, and The normalized weights can be set based on the statistical results of historical line-flight data. For example, they can be set according to the reciprocal of the fluctuation degree of the corresponding state quantity to improve the comparability between components of different dimensions. After obtaining the flight state parameters, the voxel volume calculation unit calculates the basic voxel volume by combining the rule that complex regions correspond to smaller voxel volumes and stable regions correspond to larger voxel volumes; accordingly, it can be calculated using the following formula:

[0022] in, As a natural constant, calculate the preset maximum voxel volume based on the upper limit of coarse-grained sampling in the corresponding low-complexity region. The preset minimum voxel volume corresponding to the lower limit of fine-grained sampling in high-complexity regions. Expand and utilize adjustment coefficients The effect of changes in flight status parameters on the voxel volume decay rate is controlled to adjust the system's sensitivity to complex flight maneuvers; Should be greater than To ensure that the voxel scale has effective shrinkage space; its value can be set according to the point cloud processing capability and early warning accuracy requirements. For example, a first volume upper limit that matches the point cloud amount of the regular inspection area and a second volume lower limit that matches the target net distance resolution requirements can be selected. After obtaining the basic voxel volume, the aspect ratio calculation unit calculates the aspect ratio based on the scalar magnitude of the instantaneous velocity vector and the field-of-view overlap characteristics along the flight direction; correspondingly, the aspect ratio can be calculated using the following formula:

[0023] Among them, a velocity stretching coefficient is introduced to control the effect of velocity changes on the degree of voxel stretching in the flight direction. This characterizes the extent to which the heading redundancy point can be compressed as flight speed increases; preferably, It can be set to a positive value, and its magnitude can be matched and set according to the sensor scanning frequency and the common cruise speed. For example, to enlarge the relative cross-sectional dimension of the flight direction dimension to a preset ratio under medium cruise conditions, it can be selected to make... Values ​​greater than 1; After obtaining the basic voxel volume and aspect ratio, the voxel size calculation unit calculates the dimensional parameters along the three coordinate directions; since the cross-sectional direction is responsible for pipeline clearance identification, the cross-sectional dimensions can be kept the same, i.e. And in combination with volume relationships calculate: ; Calculated Used to characterize voxel dimensions in the flight direction to control the merging scale of adjacent points in the flight direction. and Used to characterize voxel dimensions in the cross-sectional direction to control the ability to retain details in the critical direction of pipeline spacing; By coordinating changes in acceleration, yaw rate, and velocity, anisotropic voxel parameters matching the flight state are generated to identify processing areas where there may be clearance risks; the voxel parameters of the current frame are adaptively determined based on motion state components. Furthermore, based on the above implementation method, the point cloud set and the voxel parameters of the current frame are obtained. According to the voxel parameters of the current frame, the point cloud set is downsampled to generate a simplified point cloud and a grid scale label and a time label bound to the simplified point cloud. The grid scale label refers to the identification information that characterizes the voxel size level corresponding to the current point cloud block. For example, when it is smaller than a preset scale threshold, it is marked as a fine grid preservation area, and when it is greater than or equal to the preset scale threshold, it is marked as a regular grid area. The time label refers to the time index used to identify the processing cycle to which the point cloud block belongs. The determination of the preset scale threshold can be quantified based on the physical feature size of the target pipeline and its expected visibility. For example, it can be set to 1.5 to 3 times the nominal diameter of the pipeline to ensure that the point cloud resolution in the fine mesh preservation area is sufficient to restore the pipeline topology features. Specifically, after reading the point cloud set and the voxel parameters of the current frame, the point cloud downsampling module can construct the corresponding local voxel grid based on the voxel parameters of the current frame, and record the point cloud index of each grid cell in memory using a multidimensional array. For multiple points falling into the same grid cell, a representative point can be retained as a simplified point cloud output, and the scale level and reference timestamp of the grid cell can be written into the same data structure to form a labeled simplified point cloud block, which can be directly read by the subsequent pipeline collision detection module. Since the voxel parameters of the current frame already have time alignment results, the point cloud downsampling module can complete the grid division and label binding in the same processing cycle after reading, without having to look up the flight control data again. The grid scale label and time label are simultaneously fixed during the downsampling stage, so that the subsequent detection module can filter and process objects according to the label; the simplified point cloud output and label binding are realized based on the voxel parameters of the current frame, avoiding repeated scanning of the full point cloud in the subsequent detection stage. Furthermore, based on the above implementation method, a simplified point cloud, a grid scale label, and a time label are obtained. Based on the grid scale label, the simplified point cloud is classified into regions and the net distance is determined to generate pipeline conflict warning results. Among them, the fine grid reserved region refers to the region where the voxel size represented by the grid scale label is smaller than the preset scale threshold, such as the voxel region below the preset scale threshold corresponding to the deceleration, turning, or hovering phase. A regular grid area refers to a region where the voxel size represented by the grid scale label is greater than or equal to a preset scale threshold, such as a voxel region that is equal to or higher than the preset scale threshold during the uniform straight flight phase. The preset net distance threshold can be quantified and determined based on the general safety isolation distance required by the inspection industry standard and combined with the wind deflection redundancy of the specific inspection environment, and serves as the initial judgment benchmark for determining whether there is a potential risk. Specifically, after reading the tagged simplified point cloud, the pipeline conflict detection module first sieves the point cloud blocks according to the grid scale label; then it performs pipeline target identification on the simplified point cloud in the fine grid retention area, extracts the point cloud clusters identified as pipelines, and calculates the net distance between two or more pipeline point cloud clusters. During pipeline target identification, the covariance matrix is ​​calculated for each simplified point and its nearest neighbors, and eigenvalues ​​are extracted. The eigenvalues ​​of the first principal component are then calculated. With the eigenvalues ​​of the second principal components absolute ratio As a linear extension feature, when the linear extension feature is greater than a preset judgment threshold, it is marked as a pipeline candidate point. Based on the spatial Euclidean distance, adjacent pipeline candidate points with a distance less than the preset clustering distance threshold are clustered and merged into a point cloud cluster that is identified as a pipeline. When the clearance is less than the preset clearance threshold, a warning status bit is written and a pipeline conflict warning result is output; for regular grid areas, the pipeline target identification process is directly exempted to reduce the amount of computation in irrelevant areas; the warning result can be output to the result cache in the form of a structured result object in the current processing cycle for subsequent display or reporting modules to call. By first screening with grid scale labels, fine identification is concentrated in the simplified point cloud represented by the grid scale label as the fine grid preservation area; the processing scale is established by using flight state components, the basic voxel volume is maintained at or above the preset basic voxel threshold in the UAV uniform speed inspection area, and shrank to below the preset basic voxel threshold in the deceleration and detour area. Combined with flight direction stretching to suppress heading redundancy points, regional detection and early warning output are realized. In this application embodiment, the system can be deployed in the airborne computing unit or ground processing terminal corresponding to the UAV aerial survey mission. After the newly acquired raw point cloud data stream and flight status data are continuously written into their respective buffers, the data parsing module reads and generates data combinations according to the processing cycle. The voxel parameter generation module generates the voxel parameters of the current frame based on the motion state components in the data combination. The point cloud downsampling module generates a simplified point cloud with grid scale labels and time labels. When the pipeline conflict detection module detects that the clearance is less than the preset clearance threshold, it encapsulates the warning result into a structured record containing time labels, area identifiers, and clearance judgment results, and writes it into the result buffer to trigger subsequent display interface identifiers or external result call actions. Through this implementation method, new inspection data can be continuously processed from data alignment and downsampling to warning output after entering the system. In this embodiment, the state parameter calculation unit and the voxel volume calculation unit are configured to: process the scalar of the instantaneous acceleration vector and the instantaneous yaw rate through the first proportional coefficient and the second proportional coefficient to obtain the flight state parameters; obtain the attenuation coefficient according to the flight state parameters and the adjustment coefficient; and determine the basic voxel volume according to the attenuation coefficient, the preset maximum voxel volume and the preset minimum voxel volume.

[0024] The point cloud downsampling module includes a parameter smoothing unit, a mesh generation unit, a representative point extraction unit, and a label attachment unit; The parameter smoothing unit receives the current frame voxel parameters of multiple consecutive frames and performs smoothing processing on the current frame voxel parameters of multiple consecutive frames to obtain smoothed voxel parameters. The mesh generation unit receives the smoothed voxel parameters and constructs a local octree mesh based on the smoothed voxel parameters; The representative point extraction unit receives a set of point clouds at the same timestamp, distributes the point cloud set into a local octree grid, extracts representative points from the point cloud in each grid, and obtains a simplified point cloud. The label attaching unit attaches a corresponding grid scale label and time label to each point cloud block in the simplified point cloud and outputs the result.

[0025] Flight state parameters are mapped to basic voxel volumes using an exponential decay method, causing the voxel volumes to decrease as the random motion intensity increases. Simultaneously, the voxel parameters of the current frame are smoothed across multiple consecutive frames, and a local octree mesh is constructed based on the smoothed voxel parameters to output a simplified point cloud and its label results. Furthermore, based on the above implementation method, the scalar of the instantaneous acceleration vector and the instantaneous yaw rate are obtained. According to the first proportional coefficient, the second proportional coefficient, and the adjustment coefficient, the motion state components are exponentially mapped to determine the basic voxel volume. Among them, the attenuation coefficient refers to the exponential term that decreases as the flight state parameters increase. For example, during smooth line following, the attenuation coefficient approaches the preset attenuation upper limit, and during sharp turns or deceleration, the attenuation coefficient approaches the preset attenuation lower limit. The basic voxel volume refers to the volume reference used for constraining voxel size calculation in the current processing cycle. Specifically, the flight state parameters are obtained by first multiplying the scalar of the instantaneous acceleration vector by the first proportional coefficient, and then adding the product of the instantaneous yaw rate and the second proportional coefficient. Next, the attenuation coefficient is obtained by using the product of the negative of the adjustment coefficient and the flight state parameters as an exponent to calculate the exponentiation of the natural constant. Finally, the base voxel volume is obtained by multiplying the difference between the preset maximum voxel volume and the preset minimum voxel volume by the attenuation coefficient and then adding the preset minimum voxel volume. The relationship can be expressed as follows: ; ;

[0026] The attenuation coefficient introduced in the above calculation Used to convert the degree of change in flight status into a voxel volume retention ratio; when the maneuver change is below a preset maneuver threshold. If the attenuation exceeds a preset attenuation threshold, the coarser-grained sampling space is expanded accordingly; when the maneuver change exceeds a preset maneuver threshold, the corresponding change is also expanded. If the attenuation level is below a preset attenuation threshold, the sampling point density in the local detail region will be increased accordingly; the aforementioned adjustment coefficient The value of is used to control the exponential decay slope. An example value can be a certain value in the empirical positive range so that the voxel shrinkage rate matches the range of maneuver changes in the inspection task. The basic voxel volume is determined by continuous exponential mapping to achieve smooth transition and continuous adjustable shrinkage of the voxel scale, and avoid distortion of the region boundary caused by abrupt changes in the voxel scale. Furthermore, based on the above implementation method, the voxel parameters of the current frame of multiple consecutive frames are obtained. According to the sliding window and smoothing weight, the voxel parameters of the current frame of multiple consecutive frames are smoothed to generate smoothed voxel parameters. Here, multiple consecutive frames refer to a sequence of voxel parameters generated sequentially within adjacent processing cycles, such as within several consecutive scan cycles. , , Etc; smoothed voxel parameters refer to the size parameters used for actual mesh construction after time-series weighting, such as the size result where the current period parameter has the main weight and the adjacent historical period parameters have the secondary weight; In practical implementation, the parameter smoothing unit can maintain a length of [length missing] in memory. The parameter window, The preset window size can be an integer greater than 1 to cover multiple consecutive adjacent processing cycles; whenever a new current frame voxel parameter is written to the window, the parameter that entered the window earliest can be overwritten after the current calculation is completed; Normalized Gaussian weights Calculate a weighted average of the parameters within the window, for example: ,in, This represents the number of delayed frames relative to the current period, and satisfies the weighted summation. The conditions are to ensure the dimensional consistency and numerical stability of the voxel size after smoothing. Correspondingly, and It can be calculated in a similar manner; this weighted calculation uses a delay relative to the current period within a sliding window. Flight direction dimension parameters per processing cycle and combined with weights Adjusting the influence of voxel parameters on the current mesh construction at different processing cycles typically allows parameters closer to the current processing cycle to contribute more, preserving real-time performance while suppressing the impact of isolated jumps on mesh size; this also addresses situations where the parameter window is not full due to system startup or communication packet loss. The frame boundary timing condition, the parameter smoothing unit establishes a dynamic normalization mechanism: Only the weights corresponding to the currently cached valid historical frames. Perform truncation, summation, and normalization; when there is only one frame of data in the window, directly set... The voxel parameters of the current frame are used as the smoothing output to avoid weight loss or null pointer calculation caused by temporal data discontinuity, thereby ensuring the integrity of the temporal logic of the smoothing calculation and the continuous operation of the system. Temporal smoothing is performed by sliding window and smoothing weight, incorporating the changing trend of continuous flight maneuvers into the grid construction basis, suppressing disturbances in a single processing cycle, and avoiding the formation of breaks in adjacent regions due to parameter changes. Furthermore, based on the above implementation method, smoothed voxel parameters are obtained. Based on the smoothed voxel parameters, the point cloud set is divided into grids and representative points are extracted to generate a simplified point cloud. Among them, the local octree grid refers to the hierarchical grid structure established in the local space of the current point cloud according to the smoothed voxel parameters. For example, the grid structure that maintains a division below the preset cross-sectional grid scale threshold in the cross-sectional direction and generates grid cells with an aspect ratio greater than the preset ratio in the flight direction; the representative point refers to the point used to represent the distribution of the point cloud in the same grid. For example, the centroid point can be used as the output point corresponding to the grid. In the specific implementation, after the mesh generation unit reads the smoothed voxel parameters, it can first establish a local spatial partition boundary according to the enclosing range of the point cloud set in the current processing cycle, and then construct a local octree mesh accordingly; the representative point extraction unit allocates the point cloud set under the same timestamp to each mesh unit according to the coordinate index, and extracts representative points in each mesh unit to form a simplified point cloud; The local octree mesh is not unfolded at a fixed depth, but is locally partitioned under the constraint of smoothed voxel parameters, thus maintaining a compatible spatial organization between fine-grained and regular regions within the same processing cycle. When initializing the root node, the mesh generation unit establishes an anisotropic bounding box boundary according to the size ratio of the smoothed voxel parameters. When the local octree is progressively segmented in 3D space, the aspect ratio of the leaf nodes inherits the anisotropic state. The local octree mesh is constructed and representative points are extracted based on the smoothed voxel parameters. Furthermore, based on the above implementation method, a simplified point cloud is obtained. According to the grid division result and the reference time information, a grid scale label and a time label are attached to each point cloud block in the simplified point cloud to generate a labeled simplified point cloud output result. Here, a point cloud block refers to a set of simplified points output from the same grid cell or the same local grid region, such as a set of several representative points in the same fine-grained region. The grid scale label is used to characterize the scale level of the grid cell to which the point cloud block belongs, and the time label is used to characterize the processing cycle corresponding to the point cloud block. Specifically, after the representative points are extracted, the label-attached unit can write the size parameter summary of the corresponding grid unit into the label field and simultaneously write the reference timestamp; the tagged simplified point cloud results can be temporarily stored in memory as a structured array for subsequent detection modules to read by label, and the intermediate index is released after reading is completed; the simplified point cloud is accompanied by a scale context for direct use in the detection process; the simplified point cloud is synchronized with scale and time information based on label attachment; The voxel parameters of the current frame are weighted by a parametric smoothing unit to make the scale of the local octree mesh change continuously; the basic voxel volume is calculated by an exponential decay relationship to maintain a small mesh scale in high-complexity regions, thus maintaining the spatial continuity of the simplified point cloud and the integrity of pipeline cluster extraction. In the application embodiment, the point cloud downsampling module can be deployed in an independent computing thread within the aerial survey data processing process; the parameter smoothing unit maintains the current frame voxel parameter window for multiple consecutive frames; the mesh generation unit constructs a local octree mesh after reading the smoothed voxel parameters; the representative point extraction unit maps the point cloud set corresponding to the current time label to the mesh and outputs a simplified point cloud; the label attaching unit writes the mesh scale label and time label into the data field of the simplified point cloud; the labeled simplified point cloud result is written into the result buffer for the detection thread to read, thereby triggering subsequent region filtering and net distance calculation actions; In this embodiment, the pipeline conflict detection module includes a region filtering unit, a topology exclusion unit, a clearance calculation unit, and an early warning determination unit; The region filtering unit receives the simplified point cloud and the bound grid scale label. When the voxel size represented by the grid scale label is greater than the preset scale threshold, it is classified as a low-risk region point cloud cluster; otherwise, it is classified as a high-risk region point cloud cluster. The topology exclusion unit receives low-risk area point cloud clusters and performs topology exclusion operations on the low-risk area point cloud clusters based on the spatial connection status between point cloud clusters as a preset topology relationship. Point cloud clusters that are not connected to other point cloud clusters and whose length is less than a preset length threshold are deleted as point cloud clusters that meet the preset topology exclusion conditions. The clearance calculation unit receives point cloud clusters in high-risk areas, identifies pipeline targets in the point cloud clusters in high-risk areas, and uses nearest neighbor search to calculate the clearance, obtaining the minimum Euclidean distance between any two point cloud clusters identified as pipelines. The early warning determination unit receives the minimum Euclidean distance and combines it with a preset net distance threshold to generate a pipeline conflict early warning result.

[0027] The early warning determination unit is set with a preset safe clearance threshold and a preset extreme clearance threshold, and the preset extreme clearance threshold is less than the preset safe clearance threshold. The early warning judgment unit is configured to generate a safety warning record when the minimum Euclidean distance is greater than or equal to the preset safety clearance threshold. When the minimum Euclidean distance is less than the preset safe clearance threshold and greater than the preset limit clearance threshold, a first-level pipeline conflict early warning signal is generated. When the minimum Euclidean distance is less than or equal to the preset limit clearance threshold, a second-level pipeline conflict early warning signal is generated, and a review instruction is output to the external work order system.

[0028] The system also includes a parameter adjustment module; The parameter adjustment module is used to receive the frequency record of pipeline conflict early warning results returned by the pipeline conflict detection module, and compare it with the historical conflict records that have been manually verified to obtain the early warning frequency and false alarm rate. When the frequency of warnings based on a preset time window and a preset flight state interval exceeds a preset warning frequency threshold, the parameter adjustment module selects at least one of the adjustment coefficient, the first proportional coefficient, and the preset minimum voxel volume according to the preset priority and increases it; otherwise, it keeps the original parameter value. When the false alarm rate exceeds a preset false alarm rate threshold, the parameter adjustment module decreases the parameter value of the adjustment coefficient; otherwise, it keeps the original value of the adjustment coefficient unchanged. And it satisfies the constraint that the updated preset minimum voxel volume is less than the preset maximum voxel volume.

[0029] The system also includes a results output module, which comprises a front-end display interface unit, a report generation unit, and an external interface unit. The front-end display interface unit is used to receive pipeline conflict warning results and flight trajectories generated from flight status data, and to mark and display the simplified point cloud location and flight trajectory that triggered the pipeline conflict warning results in the 3D map view; The report generation unit is used to integrate the time label, grid scale label and minimum Euclidean distance value when the pipeline conflict early warning result occurs to generate a daily pipeline inspection report. The external interface unit is used to push daily pipeline inspection reports and pipeline conflict early warning results to the external equipment maintenance platform.

[0030] Nearest neighbor search is implemented using the KD-tree algorithm combined with spatial partitioning.

[0031] This implementation method further explains the pipeline conflict detection, early warning classification, parameter adjustment, and result output process. The main processing flow of this implementation method is as follows: First, the simplified point cloud is divided into low-risk area point cloud clusters and high-risk area point cloud clusters according to the grid scale label. Then, topological exclusion is performed on the low-risk area point cloud clusters, and pipeline identification and nearest neighbor distance calculation are performed on the high-risk area point cloud clusters. The classification early warning results are generated based on the minimum Euclidean distance, and the early warning results are used for parameter adjustment, result display, report generation, and external push. Furthermore, based on the above implementation method, a simplified point cloud and its bound grid scale labels are obtained. According to a preset scale threshold, the simplified point cloud is used for region screening to determine low-risk and high-risk point cloud clusters. The preset scale threshold refers to the scale judgment condition used to distinguish between regular grids and fine-grained grids. For example, voxel sizes larger than the threshold correspond to regular grids, and voxel sizes smaller than or equal to the threshold correspond to fine-grained grids. Low-risk point cloud clusters refer to point cloud clusters formed at voxel sizes larger than the preset scale threshold, and high-risk point cloud clusters refer to point cloud clusters formed at voxel sizes smaller than or equal to the preset scale threshold. Specifically, the region filtering unit reads the grid scale label from the labeled simplified point cloud results and compares the scale label with a preset scale threshold. When the voxel size represented by the grid scale label is greater than the preset scale threshold, the corresponding point cloud block is classified into a low-risk region point cloud cluster; otherwise, it is classified into a high-risk region point cloud cluster. This classification result can be cached in the detection thread with two sets of point cloud cluster indices for subsequent topology exclusion unit and net distance calculation unit to read. Region filtering is achieved based on grid scale labels, and the processed objects are subdivided into low-risk area point cloud clusters and high-risk area point cloud clusters. Furthermore, based on the above implementation method, low-risk area point cloud clusters are obtained, and topological exclusion is performed on the low-risk area point cloud clusters according to the spatial connection status between the point cloud clusters. Point cloud clusters that meet the preset topological exclusion conditions are identified and deleted. Among them, the spatial connection status refers to whether there is a continuous connection relationship between adjacent point cloud clusters. For example, two points cloud clusters that are connected to each other in space can be regarded as connected, and isolated and suspended short point cloud clusters can be regarded as unconnected. The preset topological exclusion conditions refer to the combination of conditions that are not connected to other point cloud clusters and have a length less than a preset length threshold. Specifically, after the topology exclusion unit reads the point cloud clusters in the low-risk area, it can first determine whether they are connected to other point cloud clusters based on the relative distribution relationship between the point cloud clusters. The determination is made using a quantified geometric association rule: extract the geometric endpoints of any two point cloud clusters, calculate the nearest Euclidean distance between the endpoints, and calculate the tangent vectors of each cluster at the corresponding endpoints. The specific calculation rules for the tangential vector are as follows: within the point cloud cluster, search for a set of local nearest neighbor points within a preset radius centered on the corresponding geometric endpoint. Construct a covariance matrix using the three-dimensional coordinates of the local nearest neighbor point set, and calculate the eigenvalues ​​and eigenvectors of the covariance matrix. Extract the first principal component vector corresponding to the largest eigenvalue as the actual tangential vector at the endpoint. Among them, the preset connection spacing threshold can be quantitatively set according to the physical continuity of the target pipeline and the maximum allowable breakpoint distance, for example, set to 0.5 meters to 1.5 meters; the preset collinear angle can be set according to the maximum natural curvature of the target pipeline, for example, set to 15 degrees to 30 degrees, thereby quantitatively representing the connection trend in space. Next, calculate the length of the unconnected point cloud clusters, that is, extract the vector Pi of all points in the unconnected point cloud clusters into the first principal component vector. The dot product mapping result on, and using the formula The projection range span is calculated as the judgment length; if the length is less than the preset length threshold, it is deleted as a point cloud cluster that meets the preset topology exclusion conditions. Here, a preset length threshold is used to remove point cloud clusters that are short, isolated, and do not have continuous pipeline structure characteristics. Its value can be set in combination with the minimum effective pipeline segment scale in the inspection situation; low-risk area point cloud clusters are screened out based on spatial connection status and length threshold. Furthermore, based on the above implementation method, a point cloud cluster of high-risk area is obtained, and the net distance of pipeline targets in the point cloud cluster of high-risk area is calculated according to the nearest neighbor search to generate the minimum Euclidean distance; wherein, the minimum Euclidean distance refers to the shortest spatial distance between any two point cloud clusters that are identified as pipelines, such as the distance between the nearest points of two adjacent pipelines in a local intersection area; the nearest neighbor search refers to the distance calculation method used to quickly find the nearest point pair between two sets of points; Specifically, after the net distance calculation unit identifies pipeline targets in the point cloud clusters of high-risk areas, it can establish spatial indexes for each point cloud cluster identified as a pipeline. To improve query efficiency, the nearest neighbor search is implemented using the KD tree algorithm combined with spatial partitioning: first, the extreme coordinates of the global three-dimensional bounding box of all point cloud clusters within the calculation area are extracted. And according to the preset partition side length threshold Using the formula Generate a uniform three-dimensional spatial cell mesh, wherein, This indicates the 3D index number of the grid along the X-coordinate axis. To perform the rounding up operation, the point cloud falling within the same grid 3D index is divided into corresponding local spatial partitions, and then a KD tree index structure is built within each local spatial partition to reduce the global search range. When it is necessary to calculate the net distance between any two point cloud clusters identified as pipelines, the nearest neighbor of the point cloud cluster of the other pipeline is queried point by point from the point cloud cluster of one pipeline to obtain a candidate distance set, and the minimum value among them is taken as the minimum Euclidean distance. Spatial partitioning narrows the local comparison range, and KD-trees are used to build a query index for the point set within the local area; based on the KD-tree algorithm combined with spatial partitioning, net distance search is performed in high-risk areas to calculate the minimum Euclidean distance, avoiding global traversal search; Furthermore, based on the above implementation method, the minimum Euclidean distance is obtained, and a warning judgment is made on the minimum Euclidean distance according to the preset safe clearance threshold and the preset extreme clearance threshold to generate a pipeline conflict warning result; wherein, the preset safe clearance threshold refers to the judgment condition that characterizes the lower limit of the safe distance, the preset extreme clearance threshold refers to the judgment condition that characterizes the severe proximity state, and the preset extreme clearance threshold is less than the preset safe clearance threshold; the safety prompt record, the first-level pipeline conflict warning signal and the second-level pipeline conflict warning signal correspond to different clearance risk levels respectively; In actual calibration, the preset safe clearance threshold can be determined by combining the voltage level of the pipeline being tested or the minimum safe distance specified in the industry power transmission and distribution specifications, plus the redundancy of equipment ranging error. For example, it can be set to 4 to 6 meters. The preset extreme clearance threshold can be set according to the critical distance that causes physical collision or direct air discharge breakdown. For example, it can be set to 1.5 to 2.5 meters, which serves as a clear quantitative boundary to distinguish between general warnings and triggering emergency response review instructions. Specifically, after reading the minimum Euclidean distance, the early warning determination unit generates a safety warning record when the minimum Euclidean distance is greater than or equal to a preset safety clearance threshold; generates a first-level pipeline conflict early warning signal when the minimum Euclidean distance is less than the preset safety clearance threshold but greater than a preset limit clearance threshold; and generates a second-level pipeline conflict early warning signal when the minimum Euclidean distance is less than or equal to a preset limit clearance threshold, and outputs a review instruction to the external work order system. The review instruction can be encapsulated as structured task data, including at least a time tag, a corresponding point cloud region identifier, and a risk level, to trigger the review process in the external work order system. A tiered early warning output is achieved based on dual threshold determination, and corresponding actions are taken for net distance results with different risk levels. Furthermore, based on the above implementation method, the frequency records of pipeline conflict early warning results and historical conflict records confirmed by manual review are obtained. According to the preset time window and preset flight state interval, the adjustment coefficient, the first proportional coefficient, and the preset minimum voxel volume are adjusted to determine the updated parameter values. Among them, the preset time window refers to the statistical interval used to count the occurrence of early warnings within a certain processing cycle, such as the cumulative interval of early warnings for several consecutive inspection cycles; the preset flight state interval refers to the conditional interval for merging and counting similar flight state parameter ranges or similar speed and acceleration states. Specifically, the parameter adjustment module receives the frequency record of pipeline conflict early warning results returned by the pipeline conflict detection module, and compares it with the historical conflict records that have been manually verified to obtain the early warning frequency and false alarm rate. The quantitative calculation rule for the false alarm rate is as follows: count the total number of records that have been manually verified and confirmed not to have generated actual net distance conflict within the corresponding preset time window, and divide it by the total number of early warning results output by the system within the same time window. When the frequency of warnings based on a preset time window and a preset flight state interval exceeds a preset warning frequency threshold, at least one of the adjustment coefficient, the first proportional coefficient, and the preset minimum voxel volume is increased according to the preset priority; otherwise, the original parameter value is maintained. When the false alarm rate exceeds a preset false alarm rate threshold, the parameter adjustment module decreases the parameter value of the adjustment coefficient; otherwise, the original value of the adjustment coefficient remains unchanged. The updated preset minimum voxel volume must be less than the preset maximum voxel volume. When executing specific control commands for parameter adjustment, a clear quantitative iteration mechanism can be used to increase or decrease the adjustment. For example, each time an adjustment is triggered, the adjustment step size can be 5% of the current value of the corresponding parameter or a set fixed step value. For preset priorities, the adjustment coefficient can be increased by a fixed step size according to the system's sensitivity to parameters from high to low. If the warning frequency still does not decrease below the preset warning frequency threshold after a preset statistical verification period, such as after updating the historical statistical window containing manual review results 5 times, the next priority is triggered, and the first proportional coefficient or the preset minimum voxel volume is increased by a step size of 10%. This achieves closed-loop control and avoids system downsampling instability caused by a single drastic adjustment. In the above process, when the adjustment coefficient increases, it indicates that the effect of flight state parameters on voxel volume contraction is enhanced; when the first proportional coefficient increases, it indicates that the contribution of acceleration change to complexity judgment is enhanced; when the preset minimum voxel volume increases, it indicates that the minimum sampling accuracy boundary is shifted upward, which is used to control the overly fine sampling range; based on the warning frequency and false alarm rate, the parameters are adjusted and the parameter values ​​are corrected by comparing the warning results with the manual review records. Furthermore, based on the above implementation method, the pipeline conflict early warning result, flight trajectory, time tag, grid scale tag, and minimum Euclidean distance value are obtained. According to the display, reporting, and external push requirements, the early warning result is processed for output, an inspection report is generated, and pushed to the external equipment maintenance platform. Among them, the flight trajectory refers to the trajectory result generated by the flight status data, such as the trajectory line of the inspection path in the 3D map view; the simplified point cloud position that triggers the early warning result refers to the marked position of the corresponding early warning area in the 3D map. Specifically, the front-end display unit receives pipeline conflict early warning results and flight trajectories generated from flight status data, and marks and displays the simplified point cloud location and flight trajectory that triggered the early warning results in the 3D map view; the report generation unit integrates the time label, grid scale label, and minimum Euclidean distance value when the pipeline conflict early warning results occur to generate a pipeline daily inspection report; the external interface unit encapsulates the pipeline daily inspection report and pipeline conflict early warning results into data packets adapted to the external equipment maintenance platform and pushes them; the output of the external interface unit does not stop at the result recording level, but directly triggers the external maintenance platform to receive the corresponding early warning information so that it can enter the subsequent operation and maintenance processing flow; The result output module realizes the display of early warning information, report generation and external push, and incorporates the early warning results into the inspection business chain; the region is filtered according to the grid scale label, the topology exclusion is performed on low-risk areas, and the nearest neighbor search is performed on high-risk areas using the KD tree algorithm combined with spatial partitioning; the early warning judgment unit divides the risk level according to the preset safe net distance threshold and the preset extreme net distance threshold and generates corresponding output actions. In this application example, the system can be deployed in the data processing platform for inspection tasks. After the tagged simplified point cloud enters the detection thread, the area filtering unit generates low-risk area point cloud clusters and high-risk area point cloud clusters. The topology exclusion unit deletes low-risk area point cloud clusters that meet the preset topology exclusion conditions. The net distance calculation unit uses the KD tree algorithm combined with spatial partitioning to obtain the minimum Euclidean distance. The early warning judgment unit outputs a safety prompt record, a first-level pipeline conflict early warning signal, or a second-level pipeline conflict early warning signal, and outputs a review instruction to the external work order system when the second-level pipeline conflict early warning signal is generated. Meanwhile, the parameter adjustment module adjusts relevant parameters based on the cumulative early warning results and manual review records. The result output module displays the early warning information in the 3D map view, generates a daily pipeline inspection report, and pushes it to the external equipment maintenance platform. Thus, the newly entered inspection data can continuously form a complete processing process from detection, judgment, adjustment to output.

[0032] In a specific application embodiment of the present invention, a drone performs inspection of a pipeline with a nominal diameter of 0.2 meters, with a preset maximum voxel volume V. max =3.375 cubic meters, preset minimum voxel volume V min =0.008 cubic meters, first proportionality coefficient α=0.6, second proportionality coefficient β=0.4, adjustment coefficient γ=2.0, speed stretching coefficient =0.05, the preset scale threshold is 2.5 times the nominal diameter of the pipeline, i.e., 0.5 meters.

[0033] During the steady cruise phase, the scalar of the instantaneous acceleration vector =0.2 m / s², instantaneous yaw rate =0.05 radians per second, scalar of instantaneous velocity vector =8 meters per second; Flight status parameters attenuation coefficient ; Basic voxel volume cubic meters; aspect ratio Dimensions in the cross-sectional direction Meters, dimensions in flight direction If the voxel size is greater than the preset scale threshold of 0.5 meters, the corresponding area is marked as a regular grid area, eliminating the need for pipeline target identification.

[0034] After entering the deceleration and circling phase, the flight status data for three consecutive processing cycles are as follows: Cycle t-2 meters per second squared radians per second meters per second, calculated using the above formula. , , cubic meter, , rice, meters; period t-1 meters per second squared radians per second meters per second, calculated , , cubic meter, rice, meters; period t meters per second squared radians per second meters per second, calculated , , cubic meter, , rice, rice.

[0035] The parametric smoothing unit uses a window size of N=3, with normalized Gaussian weights W0=0.5, W1=0.3, and W2=0.2, satisfying the following conditions: Smoothed voxel parameters Rice, similarly The voxel size after smoothing is still less than the preset scale threshold of 0.5 meters. The corresponding area is marked as the fine mesh retention area, and the mesh scale changes continuously between adjacent processing cycles without point cloud breakage or mesh boundary flicker.

[0036] The pipeline collision detection module performs pipeline target identification on the simplified point cloud of the fine-mesh preserved region: the first principal component eigenvalue of the covariance matrix of the nearest neighbor set of a simplified point. eigenvalues ​​of the second principal component Linear extension feature If the distance between adjacent pipeline candidate points is greater than the preset judgment threshold of 3.0, the point is marked as a pipeline candidate point; if the distance between adjacent pipeline candidate points is 0.18 meters and less than the clustering distance threshold of 0.3 meters, the points are clustered and merged into a point cloud cluster that is identified as a pipeline.

[0037] For low-risk point cloud clusters identified by the regular grid area, topological exclusion is performed: if the nearest Euclidean distance between the geometric endpoints of two point cloud clusters is 0.7 meters, which is less than the preset connection spacing threshold of 1.0 meter, and the angle between the tangent vectors of the first principal components at the two endpoints is 12 degrees, which is less than the preset collinearity angle of 20 degrees, they are determined to be connected and retained; another point cloud cluster not connected to other point cloud clusters has a projection range span of Point cloud clusters that are less than the preset length threshold of 2.0 meters are deleted as meeting the preset topology exclusion criteria.

[0038] The clearance calculation unit extracts the global 3D bounding box of the region to be calculated, with an X-direction range of 10.0 meters to 40.0 meters, and a preset threshold for the side length of the partition. Rice, then Similarly, , A total of 48 local spatial partitions were generated, and KD-tree indexes were built for each. Through nearest neighbor search, the coordinates of the nearest point pairs between two point cloud clusters identified as pipelines were found to be 22.35 meters, 11.20 meters, and 4.60 meters, and 22.35 meters, 12.40 meters, and 4.60 meters, respectively, with the minimum Euclidean distance being... rice.

[0039] The preset safe clearance threshold is set to 5.0 meters, and the preset extreme clearance threshold is set to 2.0 meters; because If the distance is less than or equal to the preset limit clearance threshold, the early warning judgment unit generates a second-level pipeline conflict early warning signal and outputs a review instruction containing time stamp, area identifier and risk level to the external work order system; if the minimum Euclidean distance in another area of ​​the same flight is calculated to be 3.6 meters, which is between the preset limit clearance threshold and the preset safe clearance threshold, a first-level pipeline conflict early warning signal is generated accordingly.

[0040] The parameter adjustment module uses a preset time window of 30 consecutive inspection cycles. Within this window, the warning frequency is 18 times. If the number of warnings exceeds the preset warning frequency threshold by 10 times, the adjustment coefficient will be adjusted in steps of 5% according to the preset priority. The error rate was increased from 2.0 to 2.1; 3 records within this window were manually verified and confirmed to have no actual net distance conflict, resulting in a false alarm rate of [missing information]. If the false alarm rate is less than the preset false alarm rate threshold of 20%, the adjustment coefficient will no longer be reduced.

[0041] The original point cloud for this flight contained approximately 520,000 points. After dynamic downsampling, the point cloud was reduced to approximately 29,000 points. About 78% of the point cloud blocks were marked as regular grid areas, eliminating the need for pipeline target identification. The time taken for net distance calculation in a single processing cycle was reduced from approximately 2.6 seconds for global indiscriminate calculation to approximately 0.4 seconds. Furthermore, all six actual net distance conflicts confirmed by manual review were detected. This concrete data demonstrates that the application's application, by dynamically adjusting the downsampling scale based on flight status, can avoid the loss of key net distance features under complex operating conditions and effectively reduce the overall computational load of the system while ensuring the accuracy of early warning.

[0042] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. An aerial survey data processing system for pipeline conflict early warning, characterized in that, include: The module includes a data parsing module, a voxel parameter generation module, a point cloud downsampling module, and a pipeline conflict detection module. The data parsing module synchronously receives the raw point cloud data stream generated by the lidar and the flight status data output by the UAV flight control system. Using the lidar's scanning shutter time as the reference timestamp, the flight status data is aligned to the reference timestamp, and a data combination containing motion state components and point cloud sets is output. The voxel parameter generation module generates voxel parameters for the current frame based on the motion state components, and sends the voxel parameters for the current frame along with the corresponding timestamp to the point cloud downsampling module. The point cloud downsampling module performs downsampling processing on the point cloud set according to the voxel parameters of the current frame, and outputs a simplified point cloud and a grid scale label and time label bound to the simplified point cloud; The pipeline conflict detection module determines the corresponding area as a fine mesh retention area when the voxel size represented by the grid scale label is smaller than a preset scale threshold. It then performs pipeline target identification on the simplified point cloud of the fine mesh retention area, extracts the point cloud clusters identified as pipelines, calculates the net distance between two or more point cloud clusters identified as pipelines, and outputs the pipeline conflict warning result according to the preset net distance threshold. When the voxel size represented by the grid scale label is greater than or equal to the preset scale threshold, the corresponding area is determined to be a regular grid area and the pipeline target identification process is waived.

2. The aerial survey data processing system for pipeline conflict early warning according to claim 1, characterized in that, The data parsing module includes a data receiving interface unit, a time base extraction unit, and an interpolation and alignment unit. The data receiving interface unit is used to synchronously receive the raw point cloud data stream containing three-dimensional coordinates and reflection intensity generated by the lidar, as well as the flight status data output by the UAV flight control system. The flight status data includes at least the instantaneous velocity vector, instantaneous acceleration vector, and instantaneous yaw rate. The time reference extraction unit is used to extract the scanning shutter time of the lidar as a reference timestamp; The interpolation and alignment unit is used to interpolate and align the flight state data based on the reference timestamp to obtain the motion state component aligned to the reference timestamp, and to combine the motion state component with the point cloud set corresponding to the reference timestamp to output the data combination.

3. The aerial survey data processing system for pipeline conflict early warning according to claim 2, characterized in that, The voxel parameter generation module includes a state parameter calculation unit, a voxel volume calculation unit, an aspect ratio calculation unit, and a voxel size calculation unit. The state parameter calculation unit receives the motion state components, extracts the instantaneous acceleration vector and instantaneous yaw rate, and calculates the flight state parameters. The voxel volume calculation unit receives the flight status parameters and calculates the basic voxel volume by combining the preset maximum voxel volume and the preset minimum voxel volume. The aspect ratio calculation unit receives the instantaneous velocity vector in the motion state component and calculates the ratio between the flight direction dimension and the cross-sectional feature dimension based on the scalar magnitude of the instantaneous velocity vector as the aspect ratio. The voxel size calculation unit combines the basic voxel volume and the aspect ratio to calculate and output the size parameters along the three coordinate directions as the voxel parameters of the current frame.

4. The aerial survey data processing system for pipeline conflict early warning according to claim 3, characterized in that, The state parameter calculation unit and the voxel volume calculation unit are configured to: process the scalar of the instantaneous acceleration vector and the instantaneous yaw rate through a first proportional coefficient and a second proportional coefficient to obtain the flight state parameters; obtain the attenuation coefficient based on the flight state parameters and the adjustment coefficient; and determine the basic voxel volume based on the attenuation coefficient, the preset maximum voxel volume, and the preset minimum voxel volume.

5. The aerial survey data processing system for pipeline conflict early warning according to claim 1, characterized in that, The point cloud downsampling module includes a parameter smoothing unit, a grid generation unit, a representative point extraction unit, and a label attachment unit; The parameter smoothing unit receives the current frame voxel parameters of multiple consecutive frames and performs smoothing processing on the current frame voxel parameters of multiple consecutive frames to obtain smoothed voxel parameters. The mesh generation unit receives the smoothed voxel parameters and constructs a local octree mesh based on the smoothed voxel parameters. The representative point extraction unit receives a set of point clouds under the same timestamp, allocates the set of point clouds to the local octree grid, extracts representative points of the point cloud in each grid, and obtains the simplified point cloud. The label attaching unit attaches a corresponding grid scale label and time label to each point cloud block in the simplified point cloud and outputs the result.

6. The aerial survey data processing system for pipeline conflict early warning according to claim 1, characterized in that, The pipeline conflict detection module includes a region filtering unit, a topology exclusion unit, a net distance calculation unit, and an early warning determination unit; The region filtering unit receives the simplified point cloud and the bound grid scale label. When the voxel size represented by the grid scale label is greater than a preset scale threshold, it is classified as a low-risk region point cloud cluster; otherwise, it is classified as a high-risk region point cloud cluster. The topology exclusion unit receives the low-risk area point cloud clusters and performs a topology exclusion operation on the low-risk area point cloud clusters based on the spatial connection status between the point cloud clusters as a preset topology relationship. Point cloud clusters that are not connected to other point cloud clusters and whose length is less than a preset length threshold are deleted as point cloud clusters that meet the preset topology exclusion conditions. The net distance calculation unit receives the point cloud cluster of the high-risk area, identifies the pipeline targets in the point cloud cluster of the high-risk area, and uses nearest neighbor search to calculate the net distance to obtain the minimum Euclidean distance between any two point cloud clusters that are identified as pipelines. The early warning determination unit receives the minimum Euclidean distance and generates the pipeline conflict early warning result by combining it with the preset net distance threshold.

7. The aerial survey data processing system for pipeline conflict early warning according to claim 6, characterized in that, The warning determination unit is set with a preset safe clearance threshold and a preset extreme clearance threshold, and the preset extreme clearance threshold is less than the preset safe clearance threshold. The early warning determination unit is configured to generate a safety alert record when the minimum Euclidean distance is greater than or equal to the preset safety clearance threshold. When the minimum Euclidean distance is less than the preset safe clearance threshold and greater than the preset limit clearance threshold, a first-level pipeline conflict early warning signal is generated. When the minimum Euclidean distance is less than or equal to the preset limit clearance threshold, a second-level pipeline conflict warning signal is generated, and a review instruction is output to the external work order system.

8. The aerial survey data processing system for pipeline conflict early warning according to claim 4, characterized in that, The system also includes a parameter adjustment module; The parameter adjustment module is used to receive the frequency record of pipeline conflict early warning results returned by the pipeline conflict detection module, and compare it with the historical conflict records that have been manually verified to obtain the early warning frequency and false alarm rate. When the frequency of warnings based on a preset time window and a preset flight state interval exceeds a preset warning frequency threshold, the parameter adjustment module selects at least one of the adjustment coefficient, the first proportional coefficient, and the preset minimum voxel volume according to a preset priority and increases it; otherwise, it maintains the original parameter value. When the false alarm rate exceeds a preset false alarm rate threshold, the parameter adjustment module decreases the parameter value of the adjustment coefficient; otherwise, it maintains the original value of the adjustment coefficient. And it satisfies the constraint that the updated preset minimum voxel volume is less than the preset maximum voxel volume.

9. The aerial survey data processing system for pipeline conflict early warning according to claim 6, characterized in that, The system also includes a result output module, which includes a front-end display interface unit, a report generation unit, and an external interface unit. The front-end display interface unit is used to receive the pipeline conflict warning result and the flight trajectory generated by the flight status data, and to mark and display the simplified point cloud location that triggered the pipeline conflict warning result and the flight trajectory in the three-dimensional map view; The report generation unit is used to integrate the time tag, grid scale tag and minimum Euclidean distance value when the pipeline conflict early warning result occurs to generate a pipeline daily inspection report. The external interface unit is used to push the daily pipeline inspection report and the pipeline conflict early warning result to the external equipment maintenance platform.

10. The aerial survey data processing system for pipeline conflict early warning according to claim 6, characterized in that, The nearest neighbor search is implemented using the KD-tree algorithm combined with spatial partitioning.

Images