An unmanned aerial vehicle route automatic planning method and system fusing multi-source perception

Abstract

The application provides a kind of unmanned aerial vehicle route automatic planning method and system of fusion multi-source perception, it is related to data processing technical field, the method includes: selecting the base point pose vector corresponding to left and right cross arm endpoint and preset symmetry constraint condition, deduce initial hover reference coordinate;According to the initial hover reference coordinate, the target hover coordinate is obtained by offset correction according to track deviation parameter and multi-dimensional pose parameter;According to the target hover coordinate, the expected flight height is measured and calculated in combination with the preset safety distance and the camera angle of view parameter;According to the expected flight height and the target hover coordinate, the inspection route control instruction containing heading angle and camera pitch angle is output.The application can realize the autonomous positioning and dynamic adjustment of unmanned aerial vehicle inspection route, optimize the reliability of hover reference, and at the same time, consider flight safety and the convenience of automatic operation.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to an automatic route planning method and system for unmanned aerial vehicles (UAVs) that integrates multi-source sensing. Background Technology

[0002] In power transmission line inspection scenarios, it is often necessary to use aerial equipment to take close-up photos of key components of transmission towers. Existing positioning and hovering control methods mostly rely on satellite positioning coordinates and discrete visual feature points to directly calculate the relative position of the aerial equipment and the target component. However, tower structures often have multiple components with spatially intersecting and local deformation features. Under complex perspectives, existing methods sometimes struggle to accurately determine the true spatial inclusion relationship between the target component and the aerial equipment's field of view, and often fail to accurately extract the geometric deformation trends of local micro-areas of components. For example, when conducting close-up inspections of the crossarm end insulators of a double-circuit tower, the crossarm metal structure... Due to the influence of occlusion and changes in lighting angle, traditional visual processing methods may confuse the spatially overlapping insulator attachment points with the crossarm back contour. Since most methods rely solely on global feature matching, they may fail to identify the minute spatial deflections of the insulator string caused by wind load disturbances. Under such conditions, existing hovering reference calculation logic often simply connects discrete feature points directly without analyzing local structural deformations segment by segment, which may cause the actual hovering position of the flight equipment to deviate from the theoretical safe area of ​​the target component. This spatial analysis deviation may cause inaccurate shooting angles, potentially leading to flight path deviations when approaching complex tower structures, thereby increasing the risk of flight collisions. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a method and system for automatic planning of UAV routes that integrates multi-source perception, which can realize the autonomous positioning and dynamic adjustment of UAV inspection routes, optimize the reliability of hovering reference, and at the same time take into account flight safety and the convenience of automated operation.

[0004] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: Firstly, a method for automatic flight path planning for unmanned aerial vehicles (UAVs) that integrates multi-source perception, the method comprising: Step 1: Acquire binocular image sequences, inertial navigation measurement data, and satellite positioning data during UAV flight; Step 2: Based on the binocular image sequence, identify the left and right crossarm endpoints and left and right insulator hanging points of the tower, and extract the pose vectors of four base points; construct the spatial configuration base domain based on the base point pose vectors and obtain the initial point set, and use each point in the point set as a candidate point; construct the detection vector with the candidate point as the starting point, calculate the net number of direction flips across the boundary line segment of the spatial configuration base domain, determine the spatial inclusion relationship based on the net number of direction flips, screen the effective reference points and divide them to obtain micro-area geometric elements; Step 3: Based on the vertex coordinate distribution of multiple micro-region geometric elements, analyze the spatial deformation gradient between adjacent elements point by point to obtain the trajectory correction parameters. Step 4: Based on the temporal changes of inertial navigation measurement data, satellite positioning data, and base point pose vector, accumulate the total spatiotemporal drift of pose deviation frame by frame to obtain multidimensional pose parameters; select the base point pose vectors corresponding to the endpoints of the left and right crossarms and the preset symmetric constraints to derive the initial hovering reference coordinates. Step 5: Correct the initial hovering reference coordinates based on the trajectory correction parameters and multi-dimensional pose parameters to obtain the target hovering coordinates; Step 6: Calculate the desired flight altitude based on the target hovering coordinates, combined with the preset safety distance and gimbal field of view parameters; Step 7: Based on the desired flight altitude and target hovering coordinates, output inspection route control commands including heading angle and gimbal pitch angle.

[0005] Secondly, an automatic flight path planning system for unmanned aerial vehicles (UAVs) that integrates multi-source perception includes: The acquisition module is used to acquire binocular image sequences, inertial navigation measurement data, and satellite positioning data during UAV flight. The processing module is used to identify the left and right crossarm endpoints and left and right insulator hanging points of the tower based on the binocular image sequence, and extract the pose vectors of four base points; construct the spatial configuration base domain based on the base point pose vectors and obtain an initial point set, and use each point in the point set as a candidate point; construct a detection vector starting from the candidate point, calculate the net number of direction flips across the boundary line segment of the spatial configuration base domain, determine the spatial inclusion relationship based on the net number of direction flips, screen the effective reference points and divide them into micro-area geometric elements; The analysis module is used to analyze the spatial deformation gradient between adjacent elements point by point based on the vertex coordinate distribution of multiple micro-region geometric elements, and obtain the trajectory correction parameters. The derivation module is used to accumulate the total spatiotemporal drift of the pose deviation frame by frame based on the temporal changes of inertial navigation measurement data, satellite positioning data and base point pose vectors to obtain multidimensional pose parameters; and to derive the initial hovering reference coordinates by selecting the base point pose vectors corresponding to the endpoints of the left and right crossarms and the preset symmetry constraints. The correction module is used to correct the initial hovering reference coordinates based on the trajectory correction parameters and multi-dimensional pose parameters to obtain the target hovering coordinates. The calculation module is used to calculate the desired flight altitude based on the target hovering coordinates, combined with the preset safety distance and gimbal field of view parameters; The output module is used to output inspection route control commands, including heading angle and gimbal pitch angle, based on the desired flight altitude and target hovering coordinates.

[0006] Thirdly, a computing device, comprising: One or more processors; A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to implement the method.

[0007] The above-described solution of the present invention has at least the following beneficial effects: To address the industry pain points of crossarm obstruction and uneven lighting in complex working conditions such as double-circuit towers, which easily lead to visual confusion between insulator hanging points and crossarm outlines, and misjudgment of spatially overlapping feature points, this invention constructs a spatial configuration base domain for key tower components. By quantitatively determining the spatial inclusion relationship of candidate points through the net number of boundary direction flips of the detection vector, it optimizes the coarse spatial judgment method of directly connecting feature points, making it easier to distinguish between overlapping components on the front and rear sides of the tower structure and internal and external spatial regions. This reduces the probability of false detections and misjudgments in visual inspection under complex perspectives, and helps improve the accuracy of spatial modeling of key inspection parts of the tower. By performing micro-region geometric element subdivision of the tower spatial configuration base domain and analyzing the spatial deformation gradient of adjacent elements point by point, it can quantify the local subtle deformation and attitude deflection of tower components. The deformation parameters are incorporated into the flight path correction process, enabling adaptive fine-tuning of the flight path to follow the deformation of the tower structure. This helps to improve the problems of insufficient adaptability of fixed flight paths and deviation of inspection points in high wind environments. Attached Figure Description

[0008] Figure 1 This is a schematic flowchart of an automatic route planning method for unmanned aerial vehicles (UAVs) that integrates multi-source perception, provided by an embodiment of the present invention.

[0009] Figure 2 This is a schematic diagram of an automatic flight path planning system for unmanned aerial vehicles (UAVs) that integrates multi-source perception, provided by an embodiment of the present invention. Detailed Implementation

[0010] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0011] like Figure 1 As shown, an embodiment of the present invention proposes an automatic route planning method for unmanned aerial vehicles (UAVs) that integrates multi-source perception. The method includes the following steps: Step 1: Acquire binocular image sequences, inertial navigation measurement data, and satellite positioning data during UAV flight; Step 2: Based on the binocular image sequence, identify the left and right crossarm endpoints and left and right insulator hanging points of the tower, and extract the pose vectors of four base points; construct the spatial configuration base domain based on the base point pose vectors and obtain the initial point set, and use each point in the point set as a candidate point; construct the detection vector with the candidate point as the starting point, calculate the net number of direction flips across the boundary line segment of the spatial configuration base domain, determine the spatial inclusion relationship based on the net number of direction flips, screen the effective reference points and divide them to obtain micro-area geometric elements; Step 3: Based on the vertex coordinate distribution of multiple micro-region geometric elements, analyze the spatial deformation gradient between adjacent elements point by point to obtain the trajectory correction parameters. Step 4: Based on the temporal changes of inertial navigation measurement data, satellite positioning data, and base point pose vector, accumulate the total spatiotemporal drift of pose deviation frame by frame to obtain multidimensional pose parameters; select the base point pose vectors corresponding to the endpoints of the left and right crossarms and the preset symmetric constraints to derive the initial hovering reference coordinates. Step 5: Correct the initial hovering reference coordinates based on the trajectory correction parameters and multi-dimensional pose parameters to obtain the target hovering coordinates; Step 6: Calculate the desired flight altitude based on the target hovering coordinates, combined with the preset safety distance and gimbal field of view parameters; Step 7: Based on the desired flight altitude and target hovering coordinates, output inspection route control commands including heading angle and gimbal pitch angle.

[0012] In this embodiment of the invention, addressing the industry pain points of crossarm obstruction and uneven lighting causing visual confusion between insulator hanging points and crossarm outlines, and misjudgment of spatially overlapping feature points under complex working conditions such as double-circuit towers, this invention constructs a spatial configuration base domain for key tower components. By quantitatively determining the spatial inclusion relationship of candidate points through the net number of flips in the boundary direction of the detection vector, it optimizes the coarse spatial judgment method of directly connecting feature points, making it easier to distinguish between overlapping components on the front and rear sides of the tower structure and internal and external spatial regions. This reduces the probability of false detections and misjudgments in visual inspection under complex perspectives, and helps improve the accuracy of spatial modeling of key inspection parts of the tower. By performing micro-region geometric element subdivision of the tower spatial configuration base domain and analyzing the spatial deformation gradient of adjacent elements point by point, it can quantify the local subtle deformation and attitude deflection of tower components. The deformation parameters are incorporated into the flight path correction process, enabling adaptive fine-tuning of the flight path to follow the deformation of the tower structure. This helps to improve the problem of insufficient adaptability of fixed flight paths and the deviation of inspection points under high wind conditions.

[0013] In a preferred embodiment of the present invention, step 1, acquiring the binocular image sequence, inertial navigation measurement data, and satellite positioning data during UAV flight, specifically includes: after initiating the UAV's autonomous inspection process, the UAV ascends to the standard inspection altitude according to a preset takeoff path; the binocular vision camera simultaneously starts continuous shooting, with the left and right lenses maintaining a fixed distance to synchronously acquire images of the tower and lines, outputting two images of the same scene simultaneously with each shot, and continuously shooting to form a complete binocular image sequence, with each set of images automatically marked with the generation time; the inertial navigation acquisition unit built into the fuselage synchronously acquires flight data throughout the entire process, continuously recording fuselage acceleration, rotation angle, and sway offset information, with each set of inertial data matched with the time stamp of the corresponding image to ensure complete correspondence between the image and inertial data time sequence; the satellite positioning receiving terminal on the top of the fuselage continuously receives satellite signals, calculates the UAV's longitude, latitude, and altitude information in real time, and each set of positioning data synchronously matches the image time stamp, completing the unified binding of the three types of data time sequence; all acquired images, inertial measurement data, and satellite positioning data are uniformly stored in the temporary storage space of the onboard computing unit, completing all data acquisition operations in this step.

[0014] This embodiment integrates three types of perception data—airborne vision, satellite positioning, and inertial measurement—to uniformly analyze the three-dimensional structure of the tower. It no longer relies on a single discrete feature point to locate components, and can distinguish the spatial hierarchy of multiple intersecting components, eliminating positioning errors caused by component outline confusion in occluded scenarios.

[0015] In a preferred embodiment of the present invention, step 2, based on the binocular image sequence, identifies the left and right crossarm endpoints and left and right insulator hanging points of the tower, and extracts four base point pose vectors; constructs a spatial configuration base domain based on the base point pose vectors and obtains an initial point set, using each point in the point set as a candidate point; constructs a detection vector starting from the candidate point, calculates the net number of direction flips across the boundary line segment of the spatial configuration base domain, determines the spatial inclusion relationship based on the net number of direction flips, filters effective reference points, and subdivides to obtain micro-region geometric elements, which may include: Step 201: Detect tower components in each frame of the binocular image sequence, identifying the left crossarm end point, right crossarm end point, left insulator hanging point, and right insulator hanging point of the tower, obtaining four key points. Specifically, these include: When the UAV performs inspection operations, the binocular camera on the fuselage continuously captures images of the line tower, and all captured images are stored sequentially in the onboard storage space according to the time sequence of capture; when executing the current step, retrieve a single set of binocular images frame by frame according to the time sequence of storage records. Each set of binocular images contains two independent imaging images, left and right, captured simultaneously under the same inspection scenario; sequentially read all pixel information of the left and right images within a single set of images, distinguishing between two types of targets in the image: metal crossarm structures and insulator suspension structures. The horizontally extending metal component in the middle of the transmission tower is the crossarm. The crossarm extends horizontally to the left and right sides of the image. The farthest metal end of the crossarm extending outward has no additional connecting components, and this end position is the crossarm endpoint. The farthest end of the crossarm extending to the left of the image is defined as the left crossarm endpoint, and the farthest end of the crossarm extending to the right of the image is defined as the right crossarm endpoint. Insulator strings made of insulating material are suspended below the crossarm. The metal fittings at the top of the insulator strings are fixedly connected to the metal components on the lower surface of the crossarm. The fixed points where the two metal components are in contact are the insulator hanging points. The fixed point on the lower left side of the crossarm used to suspend the insulator string is defined as the left insulator hanging point, and the fixed point on the lower right side of the crossarm used to suspend the insulator string is defined as the right insulator hanging point. After the single set of binocular image recognition is completed, the pixel position information of the four key points in the left and right images is recorded respectively, completing the extraction of the four key points of a single frame image.

[0016] Step 202: Based on the pixel disparity of the four key points in the left and right views of the binocular image, calculate the three-dimensional spatial coordinates of each key point relative to the UAV, and extract the pose vectors of the four base points. Specifically, this includes: retrieving the four key points of the tower identified in the single-frame image of step 201, selecting one key point, and reading the horizontal and vertical pixel coordinate parameters of the key point in the left image, as well as the horizontal and vertical pixel coordinate parameters of the same key point in the right image; comparing the offset distance formed by the two sets of horizontal pixel coordinate parameters in the horizontal direction, the offset distance value is the pixel disparity parameter corresponding to the current key point; retrieving the lens baseline spacing parameter and imaging focal length reference parameter stored in the factory calibration of the binocular camera, the lens baseline spacing parameter representing the physical distance between the left and right lenses of the binocular camera. The installation interval distance and the imaging focal length reference parameter represent the optical focal length value of the lens. Combined with the image offset degree reflected by the pixel parallax parameter, the horizontal orientation parameter, vertical height parameter, and straight-line distance parameter of the key point from the drone body are derived step by step. The three sets of spatial orientation and distance parameters are combined to form the complete three-dimensional spatial coordinates of the current key point. The three-dimensional spatial coordinates include the horizontal axis spatial position parameter, the vertical axis spatial position parameter, and the vertical axis spatial position parameter. The three-dimensional spatial coordinates of the remaining three key points are derived and calculated in sequence according to the same operation process. After the coordinate calculation of each key point is completed, the horizontal axis spatial position parameter, vertical axis spatial position parameter, vertical axis spatial position parameter and straight-line distance parameter corresponding to the point are integrated to generate the base point pose vector matching the key point. Four sets of independent base point pose vectors are generated for the four key points.

[0017] Step 203: Based on the positions of the four base point pose vectors in three-dimensional space, connect adjacent base point pose vectors sequentially to form a quadrilateral region, thus constructing the spatial configuration base domain. Specifically, this includes: reading the four sets of base point pose vectors, extracting the horizontal axis spatial position parameters, vertical axis spatial position parameters, and vertical axis spatial position parameters associated with each set of vectors; combining these three sets of parameters to form a complete three-dimensional coordinate system; the four sets of coordinates correspond sequentially to the left crossarm endpoint, the left insulator hanging point, the right crossarm endpoint, and the right insulator hanging point; connecting the spatial points corresponding to the four coordinates in a fixed connection order, with the first connection segment connecting the coordinates of the left crossarm endpoint. The first line connects the coordinates of the left insulator hanging point to the coordinates of the right crossarm endpoint; the second line connects the coordinates of the left insulator hanging point to the coordinates of the right crossarm endpoint; the third line connects the coordinates of the right crossarm endpoint to the coordinates of the right insulator hanging point; and the fourth line connects the coordinates of the right insulator hanging point to the coordinates of the left crossarm endpoint. The three-dimensional spatial coordinates corresponding to the pose vectors of the four base points are read, and a fitting plane is obtained using the least squares method. The four key points are projected onto this fitting plane to obtain four coplanar projection points. Based on the positions of the four coplanar projection points in three-dimensional space, adjacent projection points are connected sequentially to form a planar quadrilateral region, thus constructing the spatial configuration base domain.

[0018] Step 204: Generate uniformly distributed three-dimensional spatial points within the spatial configuration base domain according to a preset spatial sampling step size to obtain an initial point set; treat each spatial point in the initial point set as a candidate point location, specifically including: reading the horizontal axis spatial position parameters, vertical axis spatial position parameters, and vertical axis spatial position parameters of all vertices corresponding to the four boundary line segments of the spatial configuration base domain; calculating the spatial extension interval parameters of the closed quadrilateral region that fully covers the horizontal, vertical, and longitudinal dimensions; the horizontal extension interval includes the minimum and maximum horizontal axis parameters; the vertical extension interval includes the minimum and maximum vertical axis parameters; and the longitudinal extension interval includes the minimum and maximum vertical axis parameters. A pre-defined spatial sampling interval parameter includes three fixed values: horizontal sampling step size, vertical sampling step size, and vertical sampling step size. This parameter uniformly generates a large number of 3D spatial points within the 3D extension range, maintaining a uniform point distribution density throughout the process and preventing local point clustering or the absence of points in certain areas. Each generated 3D point synchronously records its own horizontal, vertical, and longitudinal spatial position parameters. All 3D spatial points generated within the spatial configuration base domain are aggregated and integrated to form a complete initial point set. This process is then iterated through all 3D spatial points within the initial point set, and each independently existing 3D point is individually marked as a candidate point.

[0019] Step 205: Starting from each candidate point, construct probe vectors in the directions of each boundary segment of the spatial configuration base domain. Specifically, this includes: sequentially retrieving the horizontal, vertical, and triangular spatial position parameters corresponding to each candidate point in the initial point set to lock the specific spatial position of the current candidate point within the spatial configuration base domain. Read the spatial extension direction parameters corresponding to each of the four boundary segments of the spatial configuration base domain. These extension direction parameters include spatial direction angle values. Using the three sets of spatial coordinate parameters of the current candidate point as the starting point for vector generation, generate independent probe vectors extending outward in the respective spatial directions of the four boundary segments. Each candidate point corresponds to four non-intersecting and non-overlapping probe vectors, with each probe vector corresponding to one of the boundary segments of the spatial configuration base domain.

[0020] Step 206: For each candidate point, calculate the sign change of the normal component when the probe vector crosses the boundary segment of the spatial configuration base domain. Accumulate the number of sign changes for all boundary segments to obtain the net number of direction flips corresponding to the candidate point. Specifically, this includes: extracting the four probe vectors generated for a candidate point, and taking out each probe vector individually to perform full path tracing; following the complete spatial direction of the probe vector extending outward from the starting point, continuously comparing the spatial positional relationship between the vector's travel path and the four boundary segments of the spatial configuration base domain, and continuously searching for the spatial positions where the two intersect; whenever the probe travel path crosses any boundary segment, immediately retrieve the pre-defined boundary segment... First, retain the normal orientation parameter, compare the positive and negative values ​​of the normal components of the boundary before and after the vector penetrates the boundary. If the positive and negative values ​​are inconsistent, it means that the normal components have switched. Each time a switch occurs, record the number of sign changes. After a single probe vector has completely traversed the entire extension path and completed the comparison and statistics of all intersections with the four boundary line segments, summarize the total number of sign changes generated by this vector throughout the entire process. Then, complete the path tracing and sign count statistics of the remaining three probe vectors in sequence. Add up all the sign change counts obtained from the statistics of the four probe vectors and sum them up. The final sum value is defined as the net number of direction flips corresponding to the current candidate point.

[0021] Step 207: Determine the spatial inclusion relationship of candidate points relative to the spatial configuration base domain based on the net number of orientation flips. Candidate points with an odd number of net orientation flips are determined to be located inside the spatial configuration base domain and marked as valid reference points. Candidate points with an even number of net orientation flips are determined to be located outside the spatial configuration base domain and are eliminated. Specifically, this includes: sequentially retrieving each candidate point in the initial point set and synchronously reading the net number of orientation flips parameter corresponding to that point; performing an integer division operation on the read net number parameter and determining whether there is a remainder after the operation; if there is no remainder after dividing the parameter by two, it means that the current candidate point is located outside the three-dimensional spatial range enclosed by the spatial configuration base domain. The horizontal axis spatial position parameter, vertical axis spatial position parameter, and vertical axis spatial position parameter corresponding to that point are directly deleted, and all data of that point will no longer participate in subsequent spatial subdivision related operations. If a remainder exists after dividing the parameter by two, it means that the current candidate point is completely within the three-dimensional space bounded by the spatial configuration base domain. A distinguishing identifier is added to the point, and all three-dimensional spatial coordinate parameters of the point are completely retained. The point is then designated as a valid reference point. After the determination of a single candidate point is completed, the next candidate point is retrieved and the entire judgment process is repeated until all candidate points in the initial point set are distinguished. After all points have been filtered, the three-dimensional coordinates of all marked valid reference points are integrated and all external point data are removed.

[0022] Step 208: All valid reference points are grouped into a valid point set. The valid point set is then divided by connecting adjacent points to obtain multiple micro-region geometric elements enclosed by adjacent valid reference points. Specifically, this includes: summarizing all valid reference points selected and marked in Step 207; integrating the horizontal, vertical, and triangular spatial position parameters of all points to form a complete valid point set; sequentially extracting the three sets of spatial coordinate parameters from any two sets of valid reference points in the valid point set, substituting them into the distance calculation logic to obtain the three-dimensional straight-line distance between the two points; and comparing the calculated three-dimensional straight-line distance with a preset interval threshold parameter, where the preset interval threshold is a pre-defined root... According to the fixed spatial distance limit set by the refined inspection standard for transmission towers, the maximum spatial distance between two points is determined. If the calculated three-dimensional straight-line distance between two points is less than the limit, the two points are determined to be adjacent. All point pairs in the effective point set are traversed to identify all adjacent point combinations. Spatial segmentation lines are drawn for each group of adjacent points. A large number of segmentation lines divide the complete three-dimensional space covered by the effective point set. After all spatial segmentation is completed, a large number of independent closed small spatial regions are generated inside the space. Each closed region is formed by connecting multiple groups of adjacent effective reference points. This type of closed space is uniformly named micro-geometric surface element.

[0023] This embodiment completes the three-dimensional spatial subdivision by connecting adjacent effective reference points, generating corresponding micro-area geometric elements, and fully restoring the spatial morphology of the local structure around the crossarm end and insulator hanging point. This provides a complete and reliable local spatial structure basis for subsequent dynamic flight path correction calculations. There is no need to collect tower laser point cloud data in advance. The UAV can complete the spatial area division work simultaneously in a single flight over the tower, saving a lot of preliminary work such as field scanning and modeling and manual marking of inspection points.

[0024] In a preferred embodiment of the present invention, step 3, which involves analyzing the spatial deformation gradient between adjacent elements point by point based on the vertex coordinate distribution of multiple micro-area geometric elements to obtain the trajectory correction parameters, may include: Step 301: Based on multiple micro-geometric elements, extract the common connection edges of adjacent micro-geometric elements to obtain a common edge set; based on the common edge set and the vertex coordinates of the micro-geometric elements, calculate the angle deviation between the normals of adjacent micro-geometric elements to obtain the element normal deflection; retrieve all micro-geometric elements, and simultaneously retrieve the vertex horizontal axis spatial position parameters, vertex vertical axis spatial position parameters, and vertex vertical axis spatial position parameters bound to each micro-geometric element. These three sets of parameters jointly determine the fixed position of a single vertex in three-dimensional space; sequentially select two groups of unpaired micro-geometric elements, and extract the three sets of spatial parameters corresponding to all vertices inside the two groups of elements respectively. The coordinate parameters are compared one by one to complete the vertex coordinate comparison. When two sets of face elements have two completely consistent vertex three-axis coordinates, it means that the two face elements are attached to each other in three-dimensional space, and the two sets of micro-region geometric face elements are determined to be adjacent to each other. The spatial line segment formed by connecting the two overlapping vertices is the common connection edge. A single common connection edge binds all the vertex parameters of the two adjacent micro-region geometric face elements. The pairwise comparison of all micro-region geometric face elements is continuously completed, and all the identified common connection edges are collected in a unified manner. Each common connection edge synchronously stores its two endpoint three-axis coordinates and the two sets of face element numbers bound to it. All common connection edges are integrated to form a complete common edge set.

[0025] Select any common connecting edge within the common edge set, and retrieve the three-axis coordinates of all vertices of the two adjacent micro-region geometric elements corresponding to this common edge. First, integrate the coordinates of all vertices of the first group of elements. Based on the positions of all vertices of the closed plane formed by the elements, derive the three-axis extension angle parameters of the overall normal of the plane. This parameter records the degree of inclination of the plane normal towards the horizontal, vertical, and axial axes, serving as the orientation parameter of the first group of elements' normals. Using the same calculation method, integrate the coordinates of all vertices of the other group of adjacent micro-region geometric elements to derive the orientation parameters of the second group of elements' normals. Compare the three-axis inclination angle parameters of the two groups of normals item by item, and calculate the difference in the horizontal, vertical, and axial angles respectively. Combine the three-axis differences to comprehensively calculate the angular offset of the two groups of normals in the three-dimensional space. This comprehensive offset is the normal deflection of the element corresponding to the current common edge.

[0026] Step 302: Based on the continuous distribution of the surface element normal deflection along the common edge set, the spatial variation trend of the surface element normal deflection is deduced segment by segment to obtain the spatial deformation gradient. Specifically, this includes: retrieving the complete common edge set generated in step 301, where each common connecting edge carries the endpoint three-axis spatial coordinate parameters and the corresponding surface element normal deflection parameters; spatially sorting all common connecting edges according to the actual layout order of the tower from top to bottom and from left to right, based on the vertical axis coordinate of the midpoint of the common edge from high to low and the horizontal axis coordinate from small to large; sequentially taking out two adjacent common connecting edges in the sorted order, reading the complete value of the normal deflection bound to the previous common edge, and then reading the complete value of the normal deflection bound to the next common edge, calculating the increase or decrease difference between the two sets of deflection, and simultaneously reading the three-axis spatial interval distance parameter between the midpoints of the two common edges.

[0027] Along the continuous extension of the physical structure of the tower crossarm and insulator hanging points, the sorted common edges are divided into continuous segments. Each segment contains multiple spatially continuous common connecting edges, and the segment boundaries are defined according to the variation of the vertical axis height at the midpoint of the common edge. Within each segment, the increase or decrease difference in the deflection of adjacent common edges and the corresponding spatial interval are recorded sequentially, thus fully recording the rise or fall of the normal deflection within the segment as the spatial position changes. The variation patterns recorded in all segments are integrated and connected to form a complete variation trend of the normal deflection of the entire tower structure as the three-dimensional spatial position changes continuously. This complete variation trend is defined as the spatial deformation gradient.

[0028] Step 303: Based on the spatial deformation gradient and combined with the preset flight path safety tolerance range, calculate the lateral compensation value and longitudinal pitch compensation value of the UAV flight trajectory to obtain the flight path correction parameters. Specifically, this includes: retrieving the full segmental change record of the spatial deformation gradient and simultaneously retrieving the flight path safety tolerance range preset in the transmission line close-range inspection specifications. The flight path safety tolerance range is divided into two types of fixed constraint parameters. The first type is the lateral offset limit parameter, which represents the maximum allowable spatial distance of the UAV fuselage relative to the tower components in the horizontal plane formed by the horizontal and vertical axes. The value is uniformly set according to the width of the UAV fuselage and the safety distance of the live conductor. The second type is the vertical pitch offset limit parameter, which represents the maximum allowable spatial distance of the UAV relative to the tower components in the vertical direction. The value is uniformly set according to the coverage of the gimbal's field of view and the safety gap between the upper and lower insulators. The two types of limit parameters together constitute the complete flight path safety tolerance range.

[0029] The horizontal structural deformation amplitude of each segment of the spatial deformation gradient is recorded and converted according to the horizontal offset limit parameter. When the gradient record shows that the tower locally bulges in the positive direction of the horizontal axis, the spatial scale that needs to be adjusted in the negative direction of the horizontal axis is calculated; when the gradient record shows that the tower locally bulges in the negative direction of the horizontal axis, the spatial scale that needs to be adjusted in the positive direction of the horizontal axis is calculated. The final calculated horizontal adjustment scale is the horizontal compensation value. The vertical structural deformation amplitude of each segment of the spatial deformation gradient is recorded and converted according to the vertical pitch offset limit parameter. When the gradient record shows that the tower locally raises the vertical axis height, the spatial scale that the UAV needs to raise is calculated; when the gradient record shows that the tower locally lowers the vertical axis height, the spatial scale that the UAV needs to lower is calculated. The final calculated vertical adjustment scale is the longitudinal pitch compensation value. The horizontal compensation value and longitudinal pitch compensation value calculated for the same spatial interval are integrated and bound together. The two sets of values ​​are matched with the three-axis coordinates of the spatial interval to form a complete trajectory correction parameter.

[0030] In this embodiment, the lateral compensation value and the longitudinal pitch compensation value can be used in conjunction with the previous pose drift multi-dimensional compensation parameters. The multi-layer spatial constraints simultaneously reduce the drone's hovering position deviation, continuously maintain a safe distance between the drone body and the components of the live pole, reduce the safety hazards of the drone body colliding with metal components during close-range fine inspection, and at the same time ensure that the gimbal shooting point is always close to the key components of the pole, reduce defects such as target offset, local overexposure, and image blur in the inspection image, and reduce the difficulty of identifying defects in the back-end lines.

[0031] In a preferred embodiment of the present invention, step 4, based on the temporal changes of inertial navigation measurement data, satellite positioning data, and base point pose vectors, accumulates the total spatiotemporal drift of pose deviation frame by frame to obtain multidimensional pose parameters; selecting the base point pose vectors corresponding to the endpoints of the left and right crossarms and preset symmetry constraints, and deriving the initial hovering reference coordinates, may include: Step 401: Based on the inertial navigation measurement data and satellite positioning data, align the sampling timestamps and extract the UAV flight attitude sequence to obtain the attitude time sequence. Specifically, this includes: When the UAV conducts inspection flights, the onboard inertial acquisition equipment and satellite signal receiving equipment synchronously and continuously acquire data. Each data acquisition action generates a set of inertial navigation measurement data and a set of satellite positioning data, and each set of data is bound to a timestamp generated at the same time. The inertial navigation measurement data completely records the UAV's lateral movement, longitudinal movement, and vertical movement at the moment of acquisition, while also recording the aircraft's pitch rotation, horizontal rotation, and lateral rotation. Angle information; satellite positioning data is recorded in its entirety at the moment of acquisition, including the UAV's horizontal, vertical, and longitudinal spatial coordinates; all inertial navigation measurement data and satellite positioning data in the onboard storage space are read, all timestamp identifiers are extracted and sorted according to the time sequence, and a set of inertial navigation measurement data and a set of satellite positioning data corresponding to the same timestamp are retrieved in sequence, and the two types of data are integrated into a complete record. A single integrated record is defined as a single frame of flight pose information; all timestamps are continuously traversed to complete data pairing and integration, and all single frame flight pose information is arranged strictly according to the order of acquisition time, and combined to form a pose time sequence.

[0032] Step 402: Based on the temporal changes between the pose timing sequence and the base point pose vectors, compare the pose offset difference between adjacent frames frame by frame to obtain the single-frame pose deviation. Specifically, this includes: retrieving the pose timing sequence and simultaneously retrieving all base point pose vectors. Each set of base point pose vectors is generated and simultaneously bound to the corresponding image capture timestamp. The image capture timestamp and the data acquisition timestamp within the pose timing sequence are kept synchronously corresponding. The flight pose information of two adjacent single frames in the pose timing sequence is extracted sequentially according to the time sequence. The three-axis movement amplitude, three types of fuselage rotation angles, and three-axis... Spatial coordinates are read, and then all data of the same type in the next frame are read. Based on the timestamps of the two sets of frames, the pose vectors of the corresponding two sets of base points are retrieved. The three-axis spatial coordinates of the left crossarm endpoint, the right crossarm endpoint, the left insulator hanging point, and the right insulator hanging point in the previous and next frames are read respectively. The differences in the three-axis movement values ​​of the UAV, the differences in the three types of fuselage rotation angles, and the differences in the three-axis coordinates of the satellite are calculated item by item. At the same time, the differences in the three-axis coordinate changes of the four key points of the tower are calculated item by item. The offset values ​​generated in all dimensions are summarized and merged. The overall offset data after merging is the single-frame pose deviation corresponding to the current adjacent frame.

[0033] Step 403: Based on the continuous distribution of single-frame pose deviations along the flight time axis, the cumulative offset of single-frame pose deviations is calculated globally to obtain the total spatiotemporal drift. Specifically, this includes: retrieving all single-frame pose deviations, which are arranged sequentially according to the flight duration of the UAV from takeoff to the current moment; starting from the first group of single-frame pose deviations corresponding to the initial flight moment, continuously superimposing the three-axis offset values ​​and fuselage rotation angle offset values ​​contained in each subsequent group of single-frame pose deviations along the time progression direction. During the superposition process, the cumulative offset values ​​of the three-axis space and the cumulative offset values ​​of the three types of rotation angles are calculated separately. The superposition process is not interrupted throughout until the single-frame pose deviations corresponding to the last group of flight frames are accumulated; after all temporal single-frame pose deviations are superimposed, the global cumulative offset data is obtained, and this complete global cumulative offset data is defined as the total spatiotemporal drift.

[0034] Step 404: Based on the spatial relationship between the total spatiotemporal drift and the base point pose vector, calculate the compensation distribution of pose error point by point to obtain multi-dimensional pose parameters including spatial position compensation and attitude angle compensation. Specifically, this includes: retrieving the total spatiotemporal drift and simultaneously retrieving the three-axis coordinates of the base point pose vectors corresponding to the four key tower points in the current time sequence; establishing a corresponding association between the various cumulative offset values ​​recorded by the total spatiotemporal drift and the spatial position of each key point; sequentially selecting the three-axis coordinates of the base point pose vector corresponding to any key point in a single frame image; the total spatiotemporal drift completely records the cumulative offset values ​​of the horizontal axis, vertical axis, and vertical direction of the point from the flight start point to the current time. These three types of values ​​represent the overall spatial offset amplitude of the point caused by the combined influence of positioning drift and visual temporal deviation.

[0035] To restore the true spatial coordinates of the key point unaffected by drift, it is necessary to match the corresponding correction scale in a spatial direction completely opposite to the cumulative offset. The cumulative offset values ​​of the horizontal axis are read separately to determine the reverse scale to be offset, followed by the cumulative offset values ​​of the vertical axis, and finally the cumulative offset values ​​of the vertical axis. These three-axis reverse correction scales are then integrated together, forming the spatial position compensation. Combining this with the cumulative offset values ​​of fuselage pitch, horizontal, and lateral rotation recorded by the total spatiotemporal drift, the reverse correction scales corresponding to the three types of rotation angles are determined, forming the attitude angle compensation. The calculation of the two types of compensation scales is completed for each of the four key tower points within a single frame image. All spatial position compensation and attitude angle compensation values ​​for the four points in the same time sequence are integrated and aggregated. The completed set of compensation scale data is defined as multidimensional pose parameters.

[0036] Step 405: Based on the multi-dimensional pose parameters and the base point pose vectors corresponding to the left and right crossarm endpoints, apply the symmetry constraint conditions of the tower crossarm structure to calculate the spatial intersection of the perpendicular bisector of the line connecting the left and right crossarm endpoints with the preset flight plane, and obtain the initial hovering reference coordinates. Specifically, this includes: retrieving the multi-dimensional pose parameters, retrieving the original three-axis coordinates of the two sets of base point pose vectors for the left and right crossarm endpoints in the current frame, and using the corresponding spatial position compensation amount within the multi-dimensional pose parameters to correct the original three-axis coordinates of the two sets of endpoints, offsetting the coordinate offset error caused by the spatiotemporal drift throughout the entire process, and obtaining the true coordinates of the two endpoints after eliminating the deviation. Real-space points; connect the two corrected endpoint spatial points with a straight line, locate the spatial coordinates of the midpoint of the line, extend a spatial line segment outward from the midpoint of the line, and keep the line segment perpendicular to the line connecting the two points throughout. This perpendicular line segment is the perpendicular bisector corresponding to the line connecting the endpoints of the left and right crossarms; the preset flight plane is a standard horizontal working plane uniformly defined before the project inspection operation based on the average height of the line and the overall direction of the entire line. Continue to extend the perpendicular bisector spatial line segment until the line segment intersects with the preset flight plane, and read the complete three-axis spatial coordinates corresponding to the intersection point. These three-axis spatial coordinates are the initial hovering reference coordinates.

[0037] In this embodiment, the symmetrical solid structure of the crossarm of the transmission tower is used as the calculation constraint. The intersection of the perpendicular bisector and the standard inspection plane is obtained by using the corrected actual endpoint coordinates as the hovering reference. This avoids the problem of the reference position being biased to one side due to separate calculation of the single-sided point. The reference point naturally falls in the standard shooting area on the front of the tower, reducing the workload of multiple subsequent route corrections.

[0038] In a preferred embodiment of the present invention, step 5, which involves offset correction of the initial hovering reference coordinates based on the trajectory correction parameters and multi-dimensional pose parameters to obtain the target hovering coordinates, may include: Step 501: Based on the trajectory correction parameters and multi-dimensional pose parameters, define the three-dimensional offset tolerance range of the initial hovering reference coordinates to obtain the spatial offset constraint envelope configuration. Specifically, this includes: retrieving the trajectory correction parameters, which record the lateral and vertical pitch adjustment scales caused by local deformation of the tower, representing the spatial magnitude that needs to be corrected due to changes in the tower structure; retrieving the multi-dimensional pose parameters, which include spatial position compensation scales in the three directions of the horizontal, vertical, and longitudinal axes, as well as three types of attitude angle compensation scales for the fuselage pitch, horizontal, and lateral directions, representing the correction magnitude caused by long-term positioning drift; and retrieving the initial hovering reference coordinates and processing the three-axis values ​​of these coordinates... The central reference point of the entire three-dimensional space is used as the starting point. The lateral adjustment scale and pitch adjustment scale are extracted from the trajectory correction parameters, and the three-axis position compensation scale of the multi-dimensional pose parameters is extracted. Taking the central reference point as the starting point, the farthest movable distance is defined in the direction of increasing horizontal axis value and in the direction of decreasing horizontal axis value. The farthest movable distances in the positive and negative directions of the longitudinal axis and the positive and negative directions of the vertical axis are defined in the same way. The farthest movable distances in the six directions together form the complete range of allowable variation of each of the three axes. The three sets of ranges are spliced ​​together to form a closed cubic three-dimensional space, which is named the spatial offset constraint envelope configuration.

[0039] Step 502: Based on the vertex coordinates of the spatial offset constraint envelope configuration, calculate the extreme value projection intervals of the envelope configuration boundary on the three-dimensional spatial coordinate axes edge by edge to obtain the extreme value projection set of each coordinate axis. Specifically, this includes: retrieving all three-axis coordinates of the eight vertices of the spatial offset constraint envelope configuration; sequentially extracting each boundary line segment constituting the cube, where each boundary line segment is formed by connecting two vertices, and reading the complete three-axis coordinates of each of the two vertices; performing a vertical projection operation on the coordinates of the two vertices individually onto the horizontal axis, retaining only the values ​​corresponding to the horizontal axis after projection, comparing the two projected values, the smaller value is the minimum value of the horizontal axis projection of the boundary, and the larger value is the maximum value of the horizontal axis projection of the boundary. The two values ​​are combined to form the current... The horizontal projection interval corresponding to the front boundary; using the same projection operation, the coordinates of the two vertices are projected vertically onto the vertical and vertical axes respectively, resulting in the vertical and vertical projection intervals of a single boundary; traversing all twelve boundary segments of the cube, all minimum horizontal projection values ​​of the boundaries are collected, and the global minimum and global maximum horizontal projection values ​​are selected from them, and the two sets of values ​​are combined to form the complete horizontal axis extreme value projection interval; following the same processing logic, all boundary vertical projection data are collected to obtain the complete vertical axis extreme value projection interval, and all boundary vertical projection data are collected to obtain the complete vertical axis extreme value projection; the three sets of complete extreme value projection intervals of the horizontal, vertical, and vertical axes are integrated into the extreme value projection set of each coordinate axis.

[0040] Step 503: Based on the extreme value projection set of each coordinate axis, extract the boundary intersection points of the extreme value projection intervals of each coordinate axis, construct an extreme value orthogonal constraint shell, and obtain the spatial offset constraint boundary. Specifically, this includes: retrieving the extreme value projection set and extracting the minimum and maximum projection values ​​of the horizontal axis, the minimum and maximum projection values ​​of the vertical axis, and the minimum and maximum projection values ​​of the vertical axis; performing cross-combinations of the three-axis extreme values, with each combination containing one horizontal axis extreme value, one vertical axis extreme value, and one vertical axis extreme value, for a total of eight combinations, each combination corresponding to the three-axis coordinates of a spatial intersection point; sequentially connecting the eight spatial intersection points in a cubic spatial arrangement order, and splicing all the connecting lines together to form a complete closed three-dimensional shell, which is defined as the extreme value orthogonal constraint shell; the extreme value orthogonal constraint shell contains six three-dimensional dividing planes, and all the dividing planes together form the spatial offset constraint boundary.

[0041] Step 504: Based on the spatial offset constraint boundary, extract the candidate offset coordinate set inside the boundary, and calculate the instantaneous change rate of position deviation and attitude deviation in three-dimensional space point by point to obtain the set of instantaneous change rates of spatial deviation. Specifically, this includes: retrieving the complete three-axis extreme value range of the extreme value orthogonal constraint shell; uniformly distributing a large number of independent three-dimensional points at a uniform fixed spatial interval within the three-dimensional space enclosed by the constraint shell, with each point representing a set of selectable hovering adjustment coordinates; integrating all the three-dimensional points distributed inside the shell to form a candidate offset coordinate set; sequentially extracting individual candidate offset coordinates within the set, comparing the three-axis values ​​of the coordinate with the initial hovering reference coordinate three-axis values ​​item by item, and calculating the difference in the horizontal axis coordinate. The system calculates the pitch angle difference, horizontal rotation angle difference, and lateral rotation angle difference of the body by combining multi-dimensional pose parameters with the vertical axis coordinate difference. A minor change in the value of any axis of the current candidate coordinate is observed, and the rate of increase or decrease of the three-axis position difference and the three types of attitude difference are simultaneously monitored. A rapid change in the difference after a value change indicates a large instantaneous rate, while a small change indicates a small instantaneous rate. The degree of increase or decrease of this difference with coordinate changes is defined as the instantaneous rate of change of spatial deviation corresponding to the current point. Each candidate coordinate is associated with a complete rate record. All points within the set are traversed, and all point coordinates and corresponding rates are summarized to form a set of instantaneous rates of change of spatial deviation.

[0042] Step 505: Based on the set of instantaneous changes in spatial deviation, calculate the cumulative total deviation of each candidate offset coordinate along the three-dimensional spatial coordinate axes, and directly solve for the three-dimensional spatial coordinates corresponding to the minimum state of the cumulative total deviation to obtain the target hovering coordinates. Specifically, this includes: retrieving the set of instantaneous changes in spatial deviation, sequentially retrieving individual candidate offset coordinates within the set, and simultaneously retrieving the corresponding three-axis instantaneous change rate data for that point; continuously accumulating the total change amplitude formed by the superposition of all position and attitude deviations along the complete numerical interval of the horizontal axis for that point, then accumulating the corresponding deviation change amplitude along the complete numerical interval of the vertical axis, and finally accumulating the corresponding deviation change amplitude along the complete numerical interval of the vertical axis; combining and adding the amplitude values ​​obtained from the three-axis accumulations, and the sum is the cumulative total deviation corresponding to the current candidate offset coordinate. Traverse all points within the candidate offset coordinate set, and independently perform three-axis deviation accumulation calculations for each point to obtain its corresponding cumulative total deviation; compare the total values ​​calculated for all points, and select the group with the lowest amplitude among all values. The candidate offset three-axis coordinates corresponding to the lowest amplitude value are the target hovering coordinates.

[0043] In this embodiment, all selectable hovering coordinates are evenly distributed inside the safety housing. For each point, the rate of change of deviation in the three-axis position and three types of attitudes is observed. This can distinguish the deviation-sensitive areas in space and improve the stability of imaging.

[0044] In a preferred embodiment of the present invention, step 6, calculating the desired flight altitude based on the target hovering coordinates and in conjunction with the preset safety distance and gimbal field of view parameters, may include: Step 601: Based on the target hovering coordinates, extract the vertical spatial distance between the UAV and the tower components to obtain the initial vertical spacing; based on the initial vertical spacing and the preset safety distance, calculate the minimum flight altitude that meets the safety constraints to obtain the safe flight altitude. Specifically, this includes: retrieving the complete three-axis spatial values ​​of the target hovering coordinates, which accurately calibrates the three-dimensional spatial points where the UAV plans to hover; retrieving all three-axis spatial coordinates of the four types of key tower inspection components (left crossarm endpoint, right crossarm endpoint, left insulator attachment point, and right insulator attachment point) located in the previous visual recognition process, and uniformly collecting the vertical axis spatial position values ​​corresponding to all key tower components; using the vertical spatial position of all tower components as a reference, comparing the spatial difference between the vertical position of the target hovering coordinates and the vertical position of each tower component, and calculating the minimum vertical spatial interval between the UAV and all charged metal components of the tower. The minimum vertical interval distance obtained by this measurement is the initial vertical spacing.

[0045] The preset safety distance used in this step is a fixed vertical safety interval parameter that is pre-fixed into the airborne system based on the actual voltage level of the transmission line during the early stage of the operation. This parameter is used to mitigate the risk of mechanical collisions and high-voltage electrical gap discharges during close-range inspections by UAVs. Fixed values ​​are assigned for commonly used transmission lines on site: 10 meters for low-voltage distribution lines; 15 meters for conventional medium-voltage transmission lines; 25 meters for high-voltage transmission lines; and 50 meters for ultra-high-voltage transmission lines. This parameter is a rigid vertical safety margin and will not dynamically change with the flight position of the UAV or the shape of the tower structure.

[0046] After obtaining the initial measured value of the vertical spacing, a vertical safety margin corresponding to the preset safety distance is added on the basis of the original measured vertical spacing between the UAV and the tower components. The superposition calculation of the two vertical distances is completed to eliminate the safety hazards caused by the insufficient original vertical spacing. After the superposition calculation, the minimum vertical flight height that the current inspection point of the UAV can meet the dual requirements of mechanical collision prevention and electrical insulation is obtained. This height is the safe flight height.

[0047] Step 602: Based on the safe flight altitude and gimbal field of view parameters, calculate the vertical field of view span required for the gimbal to fully cover the tower components, obtaining the field of view compensation height; extract the maximum value between the safe flight altitude and the field of view compensation height to obtain the desired flight altitude. Specifically, this includes: retrieving the vertical value of the safe flight altitude, and simultaneously retrieving the gimbal field of view parameters that have been permanently stored after the airborne gimbal's factory optical calibration. This parameter is an inherent optical parameter of the gimbal lens, specifically referring to the vertical field of view opening angle of the lens, which directly determines the vertical spatial span that the lens can capture. This parameter uniformly ranges from 45 degrees to 90 degrees. The field of view parameters of the onboard gimbal remain fixed, requiring no manual modification throughout the operation. A full-coverage verification is conducted by matching the gimbal's fixed vertical field of view with the drone's vertical position corresponding to the safe flight altitude, and the complete vertical spatial extension of the tower from its base to its top crossarm and insulator strings. The verification checks whether the gimbal's vertical field of view at the current safe flight altitude can completely encompass all key components of the tower to be inspected, determining if any top or bottom components are outside the lens's field of view. If the camera position at the current safe flight altitude can completely cover all tower components, there is no need to raise the drone's flight altitude. If the current camera position cannot completely cover the entire structure of the tower, the drone's vertical position needs to be raised to expand the lens's vertical field of view.

[0048] Based on the fixed gimbal's field of view and the overall vertical dimensions of the tower, the additional vertical space required for the drone to be raised to achieve full-area imaging without blind spots is calculated. This raised height, used to compensate for the insufficient field of view of the lens, is called the field-of-view compensation height. The safe flight altitude and the field-of-view compensation height are compared numerically. The safe flight altitude is the minimum flight elevation required for safe operation, while the field-of-view compensation height is the actual flight elevation required to meet the full imaging requirements. The larger of the two height values ​​is selected as the final execution altitude, which is the desired flight altitude.

[0049] This embodiment incorporates the inherent imaging field of view parameters of the equipment into the altitude calculation process, no longer relying solely on safety standards to determine the flight altitude. It can automatically determine the field of view coverage based on the overall vertical volume of the tower itself, and add targeted vertical lifting dimensions to supplement the shooting field of view. This ensures that the drone is always within the safe operating range and that the inspection image can fully capture all key inspection components, avoiding situations where the upper components of the tower are out of the shooting field of view or the lower components cannot be fully imaged.

[0050] In a preferred embodiment of the present invention, step 7, outputting a patrol route control command including a heading angle and a gimbal pitch angle based on the desired flight altitude and the target hovering coordinates, may include: Step 701: Based on the target hovering coordinates, extract the projected coordinates of the target hovering coordinates onto the horizontal plane to obtain the horizontal hovering position; based on the horizontal hovering position and the preset inspection reference direction, calculate the horizontal deflection angle of the horizontal hovering position relative to the inspection reference direction to obtain the desired heading angle. Specifically, this includes: retrieving the complete three-axis spatial position data of the target hovering coordinates. This set of coordinates contains three types of complete spatial information: the UAV's horizontal axis spatial position, vertical axis spatial position, and vertical axis spatial position, corresponding to the three-dimensional stopping point of the UAV's final fixed-point inspection; stripping and discarding the vertical axis spatial position data representing vertical height within the target hovering coordinates, retaining only the two types of planar position data: horizontal axis spatial position and vertical axis spatial position; performing vertical projection processing on the target hovering coordinates along the vertical direction to completely map the three-dimensional spatial point onto the standard horizontal inspection plane on the ground; the two-dimensional planar coordinates obtained after removing the position interference caused by vertical height are the horizontal hovering position, which can reflect the actual stopping position of the UAV's hovering point within the horizontal inspection plane.

[0051] The preset inspection reference direction set in this step is a fixed reference orientation that is pre-marked and fixed in the airborne control system along the entire transmission line conductor before the inspection operation begins. This reference direction is uniformly set to a northward deviation from the line's forward direction, with a fixed value range of 0 to 30 degrees. The orientation will not change throughout the entire transmission line inspection operation. It serves as a unified reference zero line for all UAV fuselage heading adjustments, eliminating orientation reference deviations caused by different tower positions. The overall geometric center of the tower is used as the reference origin for horizontal angle measurement, and the preset inspection reference direction is used as the reference point. Using the baseline as the starting point for the horizontal angle, starting from the baseline position, the azimuth deviation between the horizontal hovering position and the baseline is gradually compared along the horizontal inspection plane in a clockwise direction. The azimuth deviation is accumulated segment by segment to fully calculate the overall angle of the drone's current horizontal hovering position deviating from the standard inspection baseline. By adjusting the drone's horizontal orientation using this angle, the front of the drone can be aligned with the standard inspection field of view area of ​​the tower crossarm center, reducing the shooting deviation caused by the left and right tilt of the drone. The final calculated horizontal azimuth deviation angle is the desired heading angle.

[0052] Step 702: Based on the desired flight altitude and the preset tower component elevation, calculate the vertical height difference between the UAV and the tower component to obtain the target observation height difference. Specifically, this includes: retrieving the desired flight altitude, which is the fixed vertical flight altitude of the UAV determined after taking into account the safe operating distance and the complete field of view requirements of the gimbal, and is also the constant vertical height value that the UAV maintains after reaching the hovering point; the preset tower component elevation used in this step is a fixed vertical elevation parameter of the airborne system that is pre-entered according to the general tower design standard for transmission lines. This parameter specifically corresponds to the standard vertical height of the crossarm in the middle of the tower and the core inspection area of ​​the insulator string. For conventional towers, the value range of this parameter is fixed between 20 meters and 40 meters. When carrying out the entire inspection operation of the same tower, this elevation parameter remains fixed and does not require the airborne equipment to identify the tower height in real time for dynamic updates.

[0053] The system reads the real-time vertical hovering height of the UAV fuselage corresponding to the desired flight altitude, and the vertical height of the core inspection area of ​​the tower corresponding to the preset tower component elevation. Using the vertical plane of the tower insulator and crossarm assembly as the observation reference plane, the system compares the vertical position of the UAV fuselage with the vertical position of the tower components to determine the height difference between the overall UAV fuselage and the tower components. The system compares the vertical position differences layer by layer throughout the process to fully calculate the actual vertical spatial difference between the UAV lens installation position and the target inspection component of the tower. This difference is the vertical observation height difference that the gimbal lens naturally needs to compensate for, without the need for additional safety height margin or field of view compensation margin. The final calculated vertical spatial difference is the target observation height difference.

[0054] Step 703: Based on the target observation elevation difference and the horizontal distance from the horizontal hovering position to the tower component, calculate the required downward angle for the gimbal optical axis to point towards the tower component, and obtain the desired gimbal pitch angle; based on the desired heading angle and the desired gimbal pitch angle, encapsulate and generate a patrol flight path control command containing the heading angle and gimbal pitch angle, specifically including: retrieving the target observation elevation difference, and simultaneously calculating the straight-line interval length between the two points in the same horizontal patrol plane based on the two-dimensional plane coordinates of the horizontal hovering position and the center plane coordinates of the tower component. This straight-line interval length is the horizontal straight-line distance from the horizontal hovering position to the tower component. This distance is not affected by changes in vertical height and only reflects the spatial interval in the horizontal plane; combined with the UAV gimbal output... The initial horizontal installation angle is fixed at the factory. The initial horizontal forward orientation of the gimbal is used as the angle calculation benchmark. Combined with the straight-line interval distance in the horizontal plane and the target observation height difference in the vertical direction, the angle range of downward rotation of the gimbal lens is calculated step by step. First, the lens foundation alignment position is determined by combining the horizontal distance. Then, the downward angle compensation of the lens is supplemented by combining the vertical height difference. The downward angle of the lens is gradually fine-tuned so that the central optical axis of the gimbal can be aligned with the core inspection area of ​​the tower crossarm and insulator. This alleviates the problem that the target component is below the screen due to the downward angle of view being too small, or that the target component is above the screen due to the downward angle of view being too large. The final calculated downward rotation angle of the gimbal is the desired gimbal pitch angle.

[0055] The desired heading angle and the desired gimbal pitch angle calculated in the previous step are retrieved. Combined with the target hovering coordinates and desired flight altitude determined earlier, all spatial position parameters and attitude angle parameters are systematically integrated and uniformly packaged according to the communication data format preset by the UAV onboard flight control system. The fuselage horizontal heading control parameters, gimbal pitch attitude control parameters, and fixed-point hovering spatial coordinate parameters are synchronously integrated into a complete set of control data packets. This standardized data packet is the inspection route control command.

[0056] This embodiment combines two types of spatial parameters—horizontal plane interval distance and vertical observation height difference—to collaboratively solve the gimbal pitch angle. It takes into account both horizontal spatial orientation and vertical height difference, closely aligning with the actual observation space geometry of the UAV. This allows the gimbal optical axis to be better aligned with the core components of the tower, ensuring that most critical inspection components are located in the center of the inspection image, thus reducing the preprocessing difficulty of image defect identification and analysis at the back end.

[0057] like Figure 2 As shown, embodiments of the present invention also provide an automatic flight path planning system for unmanned aerial vehicles (UAVs) that integrates multi-source perception, comprising: The acquisition module is used to acquire binocular image sequences, inertial navigation measurement data, and satellite positioning data during UAV flight. The processing module is used to identify the left and right crossarm endpoints and left and right insulator hanging points of the tower based on the binocular image sequence, and extract the pose vectors of four base points; construct the spatial configuration base domain based on the base point pose vectors and obtain an initial point set, and use each point in the point set as a candidate point; construct a detection vector starting from the candidate point, calculate the net number of direction flips across the boundary line segment of the spatial configuration base domain, determine the spatial inclusion relationship based on the net number of direction flips, screen the effective reference points and divide them into micro-area geometric elements; The analysis module is used to analyze the spatial deformation gradient between adjacent elements point by point based on the vertex coordinate distribution of multiple micro-region geometric elements, and obtain the trajectory correction parameters. The derivation module is used to accumulate the total spatiotemporal drift of the pose deviation frame by frame based on the temporal changes of inertial navigation measurement data, satellite positioning data and base point pose vectors to obtain multidimensional pose parameters; and to derive the initial hovering reference coordinates by selecting the base point pose vectors corresponding to the endpoints of the left and right crossarms and the preset symmetry constraints. The correction module is used to correct the initial hovering reference coordinates based on the trajectory correction parameters and multi-dimensional pose parameters to obtain the target hovering coordinates. The calculation module is used to calculate the desired flight altitude based on the target hovering coordinates, combined with the preset safety distance and gimbal field of view parameters; The output module is used to output inspection route control commands, including heading angle and gimbal pitch angle, based on the desired flight altitude and target hovering coordinates.

[0058] It should be noted that this system is a system corresponding to the above method. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0059] Embodiments of the present invention also provide a computing device, including: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0060] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for automatic flight path planning of unmanned aerial vehicles (UAVs) integrating multi-source sensing, characterized in that, The method includes: Step 1: Acquire binocular image sequences, inertial navigation measurement data, and satellite positioning data during UAV flight; Step 2: Based on the binocular image sequence, identify the left and right crossarm endpoints and left and right insulator hanging points of the tower, and extract the pose vectors of four base points; construct the spatial configuration base domain based on the base point pose vectors and obtain the initial point set, and use each point in the point set as a candidate point; construct the detection vector with the candidate point as the starting point, calculate the net number of direction flips across the boundary line segment of the spatial configuration base domain, determine the spatial inclusion relationship based on the net number of direction flips, screen the effective reference points and divide them to obtain micro-area geometric elements; Step 3: Based on the vertex coordinate distribution of multiple micro-region geometric elements, analyze the spatial deformation gradient between adjacent elements point by point to obtain the trajectory correction parameters. Step 4: Based on the temporal changes of inertial navigation measurement data, satellite positioning data, and base point pose vector, accumulate the total spatiotemporal drift of pose deviation frame by frame to obtain multidimensional pose parameters; select the base point pose vectors corresponding to the endpoints of the left and right crossarms and the preset symmetric constraints to derive the initial hovering reference coordinates. Step 5: Correct the initial hovering reference coordinates based on the trajectory correction parameters and multi-dimensional pose parameters to obtain the target hovering coordinates; Step 6: Calculate the desired flight altitude based on the target hovering coordinates, combined with the preset safety distance and gimbal field of view parameters; Step 7: Based on the desired flight altitude and target hovering coordinates, output inspection route control commands including heading angle and gimbal pitch angle.

2. The automatic flight path planning method for unmanned aerial vehicles (UAVs) based on multi-source sensing as described in claim 1, characterized in that, Based on the binocular image sequence, the left and right crossarm endpoints and left and right insulator hanging points of the tower are identified, and four base point pose vectors are extracted. A spatial configuration basis domain is constructed based on the base point pose vectors to obtain an initial point set. Each point in the point set is used as a candidate point location, including: For each frame of the binocular image sequence, tower components are detected, and the left crossarm end point, right crossarm end point, left insulator hanging point, and right insulator hanging point of the tower are identified to obtain four key points; Based on the pixel disparity of the four key points in the left and right views of the binocular image, the three-dimensional spatial coordinates of each key point relative to the UAV are calculated, and the pose vectors of the four base points are extracted. Based on the positions of the four base point pose vectors in three-dimensional space, the adjacent base point pose vectors are connected in sequence to form a quadrilateral region, thus constructing the spatial configuration base domain. Within the spatial configuration base domain, uniformly distributed three-dimensional spatial points are generated according to a preset spatial sampling step size to obtain an initial point set; each spatial point in the initial point set is taken as a candidate point.

3. The automatic flight path planning method for unmanned aerial vehicles (UAVs) based on multi-source sensing as described in claim 2, characterized in that, A probe vector is constructed starting from candidate points. The net number of direction flips across the boundary segments of the spatial configuration base domain is calculated. The spatial inclusion relationship is determined based on the net number of direction flips. Valid reference points are selected and the micro-region geometric elements are obtained, including: Starting from each candidate point, a detection vector is constructed in the direction of each boundary line segment of the spatial configuration base domain; For each candidate point, calculate the sign change of the normal component when the probe vector crosses the boundary segment of the spatial configuration base domain, and accumulate the number of sign changes of all boundary segments to obtain the net number of direction flips corresponding to the candidate point. The spatial inclusion relationship of candidate points relative to the spatial configuration base domain is determined based on the net number of orientation flips. Candidate points with an odd number of orientation flips are determined to be located inside the spatial configuration base domain and marked as valid reference points, while candidate points with an even number of orientation flips are determined to be located outside the spatial configuration base domain and are eliminated. All valid reference points are combined into a valid point set. The valid point set is then divided by connecting adjacent points to obtain multiple micro-region geometric elements enclosed by adjacent valid reference points.

4. The automatic flight path planning method for unmanned aerial vehicles (UAVs) based on multi-source sensing as described in claim 3, characterized in that, Based on the vertex coordinate distribution of multiple micro-region geometric elements, the spatial deformation gradient between adjacent elements is analyzed point by point to obtain the trajectory correction parameters, including: Based on multiple micro-region geometric elements, extract the common connecting edges of adjacent micro-region geometric elements to obtain a common edge set; based on the vertex coordinates of the common edge set and the micro-region geometric elements, calculate the angle deviation between the normals of adjacent micro-region geometric elements to obtain the element normal deflection. Based on the continuous distribution of the normal deflection of the surface element along the common edge set, the spatial variation trend of the normal deflection of the surface element is deduced segment by segment to obtain the spatial deformation gradient. Based on the spatial deformation gradient and combined with the preset flight path safety tolerance range, the lateral compensation value and longitudinal pitch compensation value of the UAV flight path are calculated to obtain the flight path correction parameters.

5. The automatic flight path planning method for unmanned aerial vehicles (UAVs) based on multi-source sensing according to claim 4, characterized in that, Based on the temporal changes of inertial navigation measurement data, satellite positioning data, and base point pose vector, the total spatiotemporal drift of pose deviation is accumulated frame by frame to obtain multidimensional pose parameters. By selecting the pose vectors of the base points corresponding to the endpoints of the left and right crossarms and the preset symmetry constraints, the initial hovering reference coordinates are derived, including: Based on inertial navigation measurement data and satellite positioning data, the sampling timestamps are aligned and the UAV flight pose sequence is extracted to obtain the pose time sequence; Based on the temporal change between the pose temporal sequence and the base point pose vector, the pose offset difference between adjacent frames is compared frame by frame to obtain the pose deviation of a single frame. Based on the continuous distribution of single-frame pose deviation along the flight time axis, the cumulative offset of single-frame pose deviation is calculated across the entire domain to obtain the total spatiotemporal drift. Based on the relationship between the total spatiotemporal drift and the spatial position of the base point pose vector, the compensation distribution of pose error is calculated point by point to obtain multidimensional pose parameters that include spatial position compensation and attitude angle compensation. Based on the multidimensional pose parameters and the base point pose vectors corresponding to the endpoints of the left and right crossarms, the symmetric constraints of the tower crossarm structure are applied to calculate the spatial intersection of the perpendicular bisector of the line connecting the endpoints of the left and right crossarms with the preset flight plane, thus obtaining the initial hovering reference coordinates.

6. The automatic flight path planning method for unmanned aerial vehicles (UAVs) based on multi-source sensing as described in claim 5, characterized in that, The initial hovering reference coordinates are offset-corrected based on the trajectory correction parameters and multi-dimensional pose parameters to obtain the target hovering coordinates, including: Based on the trajectory correction parameters and multi-dimensional pose parameters, the three-dimensional offset tolerance range of the initial hovering reference coordinates is defined, and the spatial offset constraint envelope configuration is obtained. Based on the vertex coordinates of the spatial offset constraint envelope configuration, the extreme value projection intervals of the envelope configuration boundary on the three-dimensional spatial coordinate axes are calculated edge by edge to obtain the extreme value projection set of each coordinate axis. Based on the extreme value projection set of each coordinate axis, extract the boundary intersection points of the extreme value projection intervals of each coordinate axis, construct the extreme value orthogonal constraint shell, and obtain the spatial offset constraint boundary; Based on the spatial offset constraint boundary, extract the candidate offset coordinate set inside the boundary, and calculate the instantaneous change rate of position deviation and attitude deviation in three-dimensional space point by point to obtain the set of instantaneous change rates of spatial deviation. Based on the set of instantaneous change rates of spatial deviation, the cumulative total deviation of each candidate offset coordinate is calculated along the three-dimensional spatial coordinate axis. The three-dimensional spatial coordinates corresponding to the minimum state of the cumulative total deviation are directly solved to obtain the target hovering coordinates.

7. The automatic flight path planning method for unmanned aerial vehicles (UAVs) based on multi-source sensing according to claim 6, characterized in that, Based on the target hovering coordinates, combined with the preset safety distance and gimbal field of view parameters, the desired flight altitude is calculated, including: Based on the target hovering coordinates, the vertical spatial distance between the UAV and the tower components is extracted to obtain the initial vertical spacing; based on the initial vertical spacing and the preset safety distance, the minimum flight altitude that meets the safety constraints is calculated to obtain the safe flight altitude; Based on the safe flight altitude and the gimbal's field of view parameters, the vertical field of view span required for the gimbal to fully cover the tower components is calculated, and the field of view compensation height is obtained. Based on the safe flight altitude and the field of view compensation height, the maximum value between the two is extracted to obtain the desired flight altitude.

8. The automatic flight path planning method for unmanned aerial vehicles (UAVs) based on multi-source sensing according to claim 7, characterized in that, Based on the desired flight altitude and target hovering coordinates, output inspection flight path control commands including heading angle and gimbal pitch angle, including: Based on the target hovering coordinates, extract the projected coordinates of the target hovering coordinates on the horizontal plane to obtain the horizontal hovering position; based on the horizontal hovering position and the preset inspection reference direction, calculate the horizontal deflection angle of the horizontal hovering position relative to the inspection reference direction to obtain the desired heading angle; Based on the desired flight altitude and the preset elevation of the tower components, the vertical height difference between the UAV and the tower components is calculated to obtain the target observation height difference; Based on the target observation elevation difference and the horizontal distance from the horizontal hovering position to the tower component, the required downward angle for the gimbal optical axis to point to the tower component is calculated to obtain the desired gimbal pitch angle; based on the desired heading angle and the desired gimbal pitch angle, the inspection route control command containing the heading angle and gimbal pitch angle is encapsulated and generated.

9. An automatic flight path planning system for unmanned aerial vehicles (UAVs) integrating multi-source sensing, wherein the system implements the method as described in any one of claims 1 to 8, characterized in that, include: The acquisition module is used to acquire binocular image sequences, inertial navigation measurement data, and satellite positioning data during UAV flight. The processing module is used to identify the left and right crossarm endpoints and left and right insulator hanging points of the tower based on the binocular image sequence, and extract the pose vectors of four base points; construct the spatial configuration base domain based on the base point pose vectors and obtain an initial point set, and use each point in the point set as a candidate point; construct a detection vector starting from the candidate point, calculate the net number of direction flips across the boundary line segment of the spatial configuration base domain, determine the spatial inclusion relationship based on the net number of direction flips, screen the effective reference points and divide them into micro-area geometric elements; The analysis module is used to analyze the spatial deformation gradient between adjacent elements point by point based on the vertex coordinate distribution of multiple micro-region geometric elements, and obtain the trajectory correction parameters. The derivation module is used to accumulate the total spatiotemporal drift of the pose deviation frame by frame based on the temporal changes of inertial navigation measurement data, satellite positioning data and base point pose vector, and obtain multidimensional pose parameters. Select the pose vectors of the base points corresponding to the endpoints of the left and right crossarms and the preset symmetric constraints to derive the initial hovering reference coordinates. The correction module is used to correct the initial hovering reference coordinates based on the trajectory correction parameters and multi-dimensional pose parameters to obtain the target hovering coordinates. The calculation module is used to calculate the desired flight altitude based on the target hovering coordinates, combined with the preset safety distance and gimbal field of view parameters; The output module is used to output inspection route control commands, including heading angle and gimbal pitch angle, based on the desired flight altitude and target hovering coordinates.

10. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the method as described in any one of claims 1 to 8.

Images