Laser navigation self-positioning method of intelligent inspection robot for booster station outdoor equipment

Abstract

The invention aims to provide a booster station outdoor equipment intelligent inspection robot laser navigation self-positioning method, and relates to the technical field of automatic control, and the method comprises the following steps: S1, building a reflection intensity hot area, and initializing a GNSS-RTK weak anchor; s2, online detection and triggering are carried out; s3, intensity alignment and local fingerprint relocation are carried out; and S4, intelligent track playback smoothing and long-term drift suppression are carried out. According to the method, long-term drift can be remarkably inhibited, error accumulation caused by GNSS weak signals and inertial calculation is eliminated, and the stability and geometric consistency of a positioning result are improved; meanwhile, no extra marker or external base station support is needed, the maintenance cost is low, the environmental adaptability is high, continuous autonomous inspection of the booster station outdoor equipment can be achieved, and the reliability and the intelligent level of the whole system are improved.

Description

Technical Field

[0001] This invention relates to the field of automatic control technology, and in particular to a laser navigation autonomous positioning method for an intelligent inspection robot for outdoor equipment in a substation. Background Technology

[0002] With the continuous expansion of the power system and the deepening of intelligent transformation, substations, as a key link in the power transmission and transformation system, undertake the task of converting electrical energy from medium voltage to high voltage levels. Their operational safety is directly related to the stability of the entire power grid. To ensure the healthy operation of equipment within substations, maintenance units generally adopt a patrol inspection system to conduct periodic inspections of outdoor equipment. Traditional inspection methods rely on manual handheld infrared thermometers, cameras, or stethoscopes, which are affected by factors such as weather, personnel fatigue, and limited field of vision, resulting in low stability and coverage of inspection results. In recent years, intelligent inspection robots based on mobile platforms have been gradually introduced into substation sites. These devices are typically equipped with multi-line lidar, cameras, GNSS-RTK positioning systems, and inertial measurement units. By automatically driving and collecting data, they replace manual inspections, significantly improving safety and efficiency.

[0003] Current intelligent inspection robots primarily rely on a combined LiDAR mapping and GNSS-RTK localization approach for positioning and navigation. A common practice is to initially construct an environmental point cloud map using LiDAR, and then use LiDAR matching (e.g., based on ICP or NDT algorithms) and GNSS-RTK fusion for localization during subsequent inspections. However, this approach has several limitations in the outdoor environment of substations. Firstly, the substation area is relatively open but sparsely distributed with high point cloud feature redundancy, leading to potential false convergence in LiDAR matching. Secondly, GNSS-RTK signals are affected by steel structures, electromagnetic interference, and obstructions, frequently resulting in unstable positioning or frequent switching of the calculation state. When GNSS-RTK loses lock, the robot relies on inertial calculations to maintain positioning, but inertial systems have accumulated errors, easily causing attitude and position drift during long-distance inspections. Especially in areas with dense metal equipment or under high-voltage lines, the positioning system often experiences drift superposition, leading to decreased accuracy in subsequent tasks such as infrared thermography, visible light image alignment, and equipment identification. To reduce drift, some studies have attempted to introduce local loop closure detection mechanisms, which identify point cloud segments similar to historical paths during inspections to perform relocalization. However, in the open environment of substations, loop closure features are scarce and geometrically repetitive, making misjudgments easy. Furthermore, vision-based feature matching schemes exhibit poor stability in strong light, shadow, or rain and snow. Some manufacturers have proposed assisted localization methods based on artificial markers (such as QR codes and reflective signs), but these solutions require the deployment of numerous markers on-site, resulting in high maintenance costs. These markers are also prone to contamination or damage, failing to meet the requirements for long-term unattended operation. Summary of the Invention

[0004] In view of this, the main objective of this invention is to provide a laser navigation autonomous positioning method for an intelligent inspection robot of outdoor equipment in a substation. Firstly, during the initialization phase, a reflection intensity hotspot is constructed and a reflection intensity fingerprint is generated to provide a stable intensity feature benchmark for subsequent positioning. During formal inspection, a GNSS-RTK weak anchor is used to limit the candidate range, and local repositioning is triggered based on the reflection intensity features, thereby maintaining continuous positioning under conditions of partial occlusion or signal fluctuation. After triggering, intensity alignment and local fingerprint repositioning are performed, and the posture is optimized through layer-by-layer search to obtain a high-precision hotspot anchor positioning posture. Finally, the posture sequence is reconstructed within the intelligent trajectory playback window, and multi-level corrections are performed to ensure that the trajectory remains continuous and smooth before and after the hotspot.

[0005] To achieve the above objectives, the technical solution of the present invention is implemented as follows: The laser navigation autonomous positioning method for intelligent inspection robots of outdoor equipment in substations includes the following steps: Step S1, Reflection Intensity Hot Zone Construction and GNSS-RTK Weak Anchor Initialization, includes: during initial acquisition, processing the echo reflection intensity data of the multi-line lidar, constructing a reflection intensity hot zone containing reflection intensity fingerprints, and activating the GNSS-RTK weak anchor to provide absolute reference for planar position and heading; Step S2, online entry detection and triggering, includes: during formal inspection, fusing continuous laser intensity data to form the current intensity segment, determining candidate reflection intensity hotspots based on the planar position given by the GNSS-RTK weak anchor, performing entry judgment on the candidate reflection intensity hotspots, and triggering the reflection intensity hotspot event and setting the intelligent trajectory playback window when the judgment is passed and the GNSS-RTK weak anchor meets the robustness standard. Step S3, intensity alignment and local fingerprint relocation, includes: after triggering the reflection intensity hotspot event, aligning the current intensity segment with the reflection intensity fingerprint of the triggered hotspot, obtaining the local fingerprint relocation pose through coarse alignment and layer-by-layer search, and setting the obtained local fingerprint relocation pose as the hotspot anchor pose. Step S4, intelligent trajectory playback smoothing and long-term drift suppression, includes: within the intelligent trajectory playback window, replacing the trigger point pose with the hot zone anchor pose, and applying the motion increment within the window to reconstruct the pose sequence, performing corrections during the reconstruction process, and finally generating the final positioning result through boundary stitching.

[0006] Furthermore, the construction of the reflection intensity hot zone in step S1 specifically includes: dividing the echo reflection intensity output by the multi-line lidar in a single scanning cycle into a grayscale raster image by dividing it into two-dimensional grids with 0.20m × 0.20m as the grid unit; performing threshold segmentation on the grayscale raster image, where the threshold is equal to the local mean of the 11×11 grid window plus 10, to obtain a binary image; performing one 3×3 square dilation and one 3×3 square erosion on the binary image to complete the closure operation; extracting several connected components according to 4 adjacent connected components; selecting connected components that meet all of the following conditions as the reflection intensity hot zone: the number of grid units contained in the connected component is between 200 and 5000, the aspect ratio of the bounding box is between 0.30 and 3.50, and the average polygonal corner density of the connected component boundary is between 3 times per 10m and 30 times per 10m.

[0007] Furthermore, the reflection intensity fingerprint generated for each reflection intensity hot zone in step S1 consists of the following four data structures: 1) Grid intensity template: the grayscale raster image is cropped according to the hot zone bounding box and bilinearly interpolated to a 128×128 array; 2) Outer contour polyline: the closed boundary is extracted on the binary image using contour tracing and the vertex coordinate sequence is recorded according to the grid side length; 3) Equipment edge polyline: the intensity gradient magnitude is calculated on the grayscale raster image using the Sobel operator and binarized with a threshold of 30, and then the straight line segment with a length of not less than 5 grids is extracted using probabilistic Hough line detection and the endpoint coordinates are recorded; 4) Intensity sequence code: the grayscale value is sampled every 0.20 meters on the polyline segment where the center line of the inspection route crosses the reflection intensity hot zone and encoded as 0, 1, 2, 3, 4 according to the intervals [0,50), [50,100), [100,150), [150,200), respectively, to form a sequence.

[0008] Furthermore, the GNSS-RTK weak anchor activated in step S1 is used in step S2 to limit the search range of the candidate reflection intensity hot zone to an area with a radius of 25 meters centered on the planar position given by the GNSS-RTK weak anchor, and in step S4 to perform GNSS-RTK fitting correction on the reconstructed pose.

[0009] Furthermore, the entry judgment performed on the candidate reflection intensity hot zone in step S2 is a three-level entry judgment performed sequentially, including: Level 1 is template overlap judgment: the current intensity segment slides on the grid intensity template of the candidate reflection intensity hot zone with a grid step size of 1 and calculates the sum of the absolute differences of each grid. When the proportion of grids with absolute differences exceeding the threshold is not greater than 35%, it is judged as passing; Level 2 is edge polyline alignment judgment: the edge polylines in the current intensity segment are used to vote for 8 orientation candidates using histogram, and each of the 8 orientations is rotated in the grid and then aligned with the candidate reflection intensity. The equipment edge polyline set in the hot zone is matched end-to-end. The matching rules are that the distance between the endpoints is no more than 2 grids, the angle difference is no more than 10 degrees, and the number of matched straight line segments is no less than 3. If the matching rules are met, it is judged as passing. The third level is the intensity sequence code consistency judgment: the intensity sequence code is extracted from the current intensity segment in the direction of the center line of the inspection route and compared with the intensity sequence code of the candidate reflection intensity hot zone. When the length of the longest common subsequence of the two sequences is no less than 80% of the length of the intensity sequence code of the candidate reflection intensity hot zone, it is judged as passing.

[0010] Furthermore, the conditions for triggering the reflection intensity hot zone event in step S2 are: all three levels of entry judgment are passed, and the robustness of the GNSS-RTK weak anchor is verified at the same time. The verification criteria are: the solution state is fixed or floating, the standard deviation of the horizontal position is not greater than 0.30 meters, the standard deviation of the heading is not greater than 5 degrees, and the data interval within the last 3 seconds is not greater than 1 second, which meets the verification criteria; the range of the intelligent trajectory playback window set in step S2 is: with the trigger point as the reference, the starting point is 10 meters in front of the trigger point, and the ending point is 15 meters behind the trigger point.

[0011] Furthermore, the intensity alignment and local fingerprint relocalization in step S3 specifically include: constructing an orientation set with 16 orientation candidates and constructing a translation initial value set with the planar position of the GNSS-RTK weak anchor; generating a pose candidate set from the Cartesian combination of the orientation set and the translation initial value set; applying each pose candidate to the current intensity segment and comparing it grid-by-grid with the mesh intensity template of the trigger hot zone, selecting the pose with the fewest inconsistent grids as the coarse alignment pose; performing a layer-by-layer search around the translation and orientation of the coarse alignment pose with a staggered step size of 0.10 meters and 1 degree, respectively; calculating the number of endpoint matches with the device edge polyline after each search and selecting the pose with the largest number of matches as the new center, until the number of matches in two consecutive layers of search does not increase, and then stopping the search and outputting the local fingerprint relocalization pose.

[0012] Furthermore, the process of reconstructing the pose sequence in step S4 specifically includes: freezing the motion increments and time sequence from the starting point to the ending point within the intelligent trajectory playback window; replacing the pose corresponding to the trigger point with the hot zone anchor pose obtained in step S3; and applying the wheel speed and inertia increments within the intelligent trajectory playback window one by one along the time sequence to reconstruct the pose sequence.

[0013] Furthermore, the corrections performed sequentially after each application of wheel speed and inertia increment in step S4 include the following three correction processes: 1) Corridor correction: Based on the engineering layout drawing or pre-imported route design data, a corridor with a width of 1.50 meters is set on both sides of the centerline of the inspection route. When the plane position of the reconstructed pose is located outside the corridor, the plane position of the reconstructed pose is projected onto the boundary of the corridor according to the normal vector direction and the orientation remains unchanged; 2) Orientation change correction: When the orientation change amplitude between any two adjacent moments obtained by reconstruction is greater than 10 degrees, the orientation change amplitude is replaced with 10 degrees and assigned with the same steering sign; 3) GNSS-RTK fit correction: When playing back to the time point containing the GNSS-RTK weak anchor stability segment, the plane position of the time point is moved along a straight line to the position provided by GNSS-RTK and the orientation and height values ​​are kept continuous along the original sequence.

[0014] Furthermore, the boundary stitching in step S4 specifically includes: achieving directional continuity between the starting point of the intelligent trajectory playback window and the upstream pose outside the window within a 2-meter range using linear interpolation; and achieving positional continuity between the ending point of the intelligent trajectory playback window and the downstream pose outside the window within a 2-meter range using linear interpolation.

[0015] By employing the above technical solution, this invention achieves the following beneficial effects: Utilizing a collaborative mechanism of reflection intensity hotspots, GNSS-RTK weak anchors, and intelligent trajectory playback, high-precision, low-drift positioning of inspection robots is realized in long-distance, open, and GNSS signal-unstable scenarios. This method constructs reflection intensity hotspots and generates reflection intensity fingerprints during the initialization phase, enabling the inspection system to reproduce stable spatial characteristics under different times and lighting conditions. During formal inspection, GNSS-RTK weak anchors serve as global constraints, and local triggered repositioning is performed using intensity features, effectively compensating for the shortcomings of traditional GNSS and laser matching schemes in environments with occlusion, repetitive structures, or sparse features. Especially in the intensity alignment and local fingerprint repositioning stages, a layer-by-layer search strategy is adopted, with dual constraints of grayscale consistency and edge geometry matching, significantly improving the stability of attitude estimation and ensuring the deterministic and reproducible positioning results within the hotspot. The intelligent trajectory playback smoothing mechanism reconstructs the motion sequence based on the hot zone anchor positioning posture. It introduces multiple corrections within the window, including passageways, orientation changes, and GNSS-RTK alignment, gradually absorbing and spreading local errors. The trajectory seamlessly connects before and after the hot zone, thus suppressing inertial drift in the long term. This scheme demonstrates excellent performance in terms of inspection task continuity, geometric smoothness of positioning results, and environmental adaptability. Compared to traditional pure GNSS-RTK or laser point cloud matching systems, this invention achieves dynamic switching from continuous absolute reference to local relative repositioning without relying on additional markers or external signals, significantly reducing deployment and maintenance costs. By fusing reflection intensity characteristics with weak anchor positioning, the inspection robot maintains high stability even in scenarios with dense metal structures, complex lighting, or repetitive equipment. The overlap between the positioning trajectory and the actual passageway is significantly improved. The overall system is feasible for engineering implementation and has long-term operational reliability, effectively enhancing the autonomy and safety of the intelligent inspection system for substations. Attached Figure Description

[0016] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 The grayscale raster image of reflection intensity provided in the embodiments of the present invention; Figure 2 This is a schematic diagram of extracting the outer contour polyline and the device edge polyline provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of edge polyline endpoint matching provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the longest common subsequence alignment of strength sequence codes provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the lateral offset curve for the passageway correction provided in an embodiment of the present invention. Detailed Implementation

[0017] The method of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0018] Example 1: Laser navigation autonomous positioning method for intelligent inspection robot of outdoor equipment in booster station, including the following steps: Step S1, reflection intensity hot zone construction and GNSS-RTK weak anchor initialization; Step S2, online entry detection and triggering; Step S3, intensity alignment and local fingerprint relocation; and Step S4, intelligent trajectory playback smoothing and long-term drift suppression.

[0019] Step S1, the construction of the reflection intensity hot zone and the initialization of the GNSS-RTK weak anchor, includes: during the initialization acquisition, processing the echo reflection intensity data of the multi-line lidar, constructing a reflection intensity hot zone containing the reflection intensity fingerprint, and starting the GNSS-RTK weak anchor to provide an absolute reference for the plane position and heading.

[0020] In one specific implementation, the inspection robot travels along the target inspection route at a constant linear velocity, and the multi-line lidar outputs echo reflection intensity at a scanning frequency of 10 to 20 times per second. To ensure that all processing stages are based on the same time base, the laser timestamp, wheel speedometer timestamp, and GNSS-RTK weak anchor timestamp are time-calibrated before the start of data acquisition. The calibration method involves triggering simultaneous timing with a network time source and aligning the three time series within a millisecond range using linear interpolation. This calibration ensures a one-to-one correspondence between each subsequent frame of intensity data, mileage information, and the absolute reference, facilitating the formation of consistent data segments in space and time. The echo reflection intensity data is plotted using a two-dimensional grid. An in-station planar coordinate system is established with the starting point of the inspection route as the origin, using a fixed grid size of 0.20 meters × 0.20 meters, and the grid is constructed according to row and column indices. All echo intensities within a single frame of point cloud are linearly mapped to the range of 0 to 255, and each echo is mapped to a single grid cell according to its planar position. When a single grid cell contains multiple echoes within the same frame, a maximum value strategy is used to write the grayscale value, forming a grayscale raster image for that frame. This maximum value strategy causes highly reflective components to form stable bright areas on the grayscale image, while weak echoes are covered by stronger echoes, thereby improving the separability of subsequent hotspot extraction.

[0021] The grayscale raster image is segmented using an adaptive threshold with an 11×11 grid as the sliding window. The mean grayscale value within the window is calculated, and a threshold of 10 is set at the center of each window. This threshold selection ensures that highly reflective areas such as metal surfaces and ceramic housings remain in the bright region under different lighting and angle conditions, while low-reflective ground and vegetation are segmented into the dark region. After generating a binary image based on the above threshold, a morphological closing operation is performed. First, a 3×3 square structuring element is applied for dilation, followed by an erosion operation using the same structuring element. The closing operation fills the small holes and closes the narrow slits within the bright regions, improving continuity and enhancing the stability and reproducibility of subsequent connected component analysis.

[0022] Connected component extraction based on 4-adjacency is performed on the binary graph to obtain several connected components. For each connected component, the following three geometric quantities are calculated: number of grid cells, aspect ratio of the minimum bounding rectangle, and corner density of the boundary polyline. The boundary polyline is obtained through pixel-by-pixel boundary tracing, and the corner density is calculated by counting the number of inflection points per 10 meters. Connected components that simultaneously meet the following conditions are identified as reflection intensity hotspots: number of grid cells between 200 and 5000, aspect ratio between 0.30 and 3.50, and corner density between 3 and 30 times per 10 meters. The number of grid cells ensures that the hotspot corresponds to the scale of the equipment group; the aspect ratio condition excludes extremely long or short or extremely flat shapes; and the corner density condition prioritizes regions with regular inflection points on the boundary polyline, which are commonly found in the combined boundaries of rows of structures and groups of equipment. After filtering, a reflection intensity fingerprint is generated for each reflection intensity hotspot and bound to the spatial range of that hotspot.

[0023] The reflection intensity fingerprint consists of four data structures, recording the intensity template, appearance contour, main internal linear features, and intensity sequence features along the route for the reflection intensity hotspot. The first item is the grid intensity template. The grayscale raster image is cropped using the circumscribed rectangle of the reflection intensity hotspot, and bilinear interpolation is used to normalize the cropped area to a 128×128 array. This normalization unifies different scales and small attitude deviations to a fixed resolution, ensuring a fixed computational load for subsequent sliding matching and overlap calculations, and facilitating consistent comparisons between different hotspots. The second item is the outer contour polyline. Closed boundaries are extracted along the light-dark boundary using a contour tracing algorithm in the binary image, and the vertex coordinate sequence is recorded sequentially according to the grid side length unit. The outer contour polyline provides overall shape and orientation information. Typically, the framework and fence form a near-rectangular polyline, while electrical equipment groups form a multi-vertical composite polyline. This polyline is used for rapid alignment in subsequent entry determination and attitude discretization search. The third item is the equipment edge polyline. The Sobel operator is used to calculate the gradient magnitude in both the horizontal and vertical directions on a grayscale raster image, and a threshold of 30 is used to filter the gradient magnitude. Figure 2The data is valued, and then probabilistic Hough line detection is used to extract straight line segments with a length of no less than 5 grids, which are stored as endpoint coordinates. The polyline of the equipment edge reflects the arrangement relationship of linear components such as busbars, beams, and fence posts. The endpoint matching of the straight line segments provides reliable geometric constraints for attitude refinement search and remains stable under changes in illumination and surface contamination. The fourth item is the intensity sequence code. Gray values ​​are sampled every 0.20 meters on the polyline segment where the centerline of the inspection route crosses the hot zone of reflection intensity. Each sampled gray value is mapped to a level code from 0 to 4 according to the previously defined interval coding rules and formed into a sequence in spatial order. The intensity sequence code records the order of intensity fluctuations when the inspection route is crossed. The repetition and interval of the equipment group in the path direction present a stable segment structure in this sequence. The subsequent sequence consistency comparison for judgment is calculated based on this structure.

[0024] refer to Figure 1 This image shows a grayscale raster map of the echo reflection intensity from a multi-line lidar radar after plotting. An internal plane coordinate system is established with the starting point of the inspection route as the origin. The horizontal axis represents the X-coordinate in meters, ranging from 0 to 100 meters; the vertical axis represents the Y-coordinate in meters, also ranging from 0 to 100 meters. The entire internal plane is divided into a fixed grid of 0.20 meters × 0.20 meters. The image shows several areas of high reflectivity, including: a highly reflective metal frame (located at approximately 15-30 meters X-coordinate and 15-25 meters Y-coordinate), which appears as a bright area with a grayscale value of 200-255 on the grayscale map due to the high reflectivity of the metal surface; high-voltage electrical equipment (located at approximately 45-60 meters X-coordinate and 20-30 meters Y-coordinate), with a grayscale value of 180-225; porcelain insulators (located at approximately 25-35 meters X-coordinate and 40-55 meters Y-coordinate), with a grayscale value of 190-225; transformers (located at approximately 60-75 meters X-coordinate and 45-60 meters Y-coordinate), with a grayscale value of 195-225; and switchgear (located at approximately 40-50 meters X-coordinate and 10-20 meters Y-coordinate), with a grayscale value of 185-225. The low-reflectivity ground areas in the image are distributed between and around the aforementioned equipment groups, with grayscale values ​​ranging from 25 to 70, appearing as dark areas on the grayscale image. Each grid cell records the echo reflection intensity at that location. For cases where multiple echoes exist within a single frame for the same grid cell, a maximum value strategy is used to write the grayscale value, thereby enabling high-reflectivity components to form stable bright areas on the grayscale image, suppressing the low-reflectivity background, and improving the separability of subsequent hot spot extraction.

[0025] The four data structures above respectively provide the template, boundary shape, internal linear features, and intensity order of the bright region along the path direction, and these four elements complement each other. The mesh intensity template provides coarse positioning capability under large attitude errors, the outer contour polyline fixes the shape and appearance orientation within the template candidate, the device edge polyline improves alignment accuracy at nearby positions, and the intensity order code ensures that the order remains consistent along the path direction. This hierarchical representation can form a progressively converging matching link in scenarios where strong reflective components recur, thereby improving the stability of entry determination and the determinism of relocation in subsequent steps.

[0026] like Figure 2 As shown in the figure, this diagram illustrates the results of extracting the outer contour polylines and equipment edge polylines of the reflective intensity hot zone after binarization. A grid coordinate system is established in the figure, with both the horizontal and vertical coordinates expressed in grid units, ranging from 0 to 200 grid units. The figure presents a complex equipment group layout, including main equipment area 1 (X coordinate approximately 15-40 grids, Y coordinate approximately 13-33 grids), main equipment area 2 (X coordinate approximately 45-65 grids, Y coordinate approximately 18-35 grids), and the connecting area between the two (X coordinate approximately 35-48 grids, Y coordinate approximately 23-30 grids). These areas form bright areas after binarization, with a grayscale value of white (255) and a black (0) background. The outer contour polylines (thick black solid lines) in the figure are formed by extracting closed boundaries along the light-dark boundary using a contour tracing algorithm. These polylines constitute a complex polygonal contour, containing multiple turning points, reflecting the overall shape and orientation information of the equipment group. The vertex coordinates of the outer contour polylines are recorded sequentially in grid side length units, forming a vertex coordinate sequence. The polygonal lines (thin black solid lines) at the equipment edge in the figure were extracted by calculating the gradient magnitudes in the horizontal and vertical directions using the Sobel operator. Specifically, they include: 10 horizontal busbars, located at different Y-coordinate heights, with lengths ranging from 200 to 350 grid units; 7 vertical frameworks, located at different X-coordinate positions, with lengths ranging from 300 to 580 grid units; 4 diagonal supports connecting the main components at a certain angle; and 4 intersecting components forming the internal support structure of the equipment. These straight line segments were extracted using probabilistic Hough line detection, and their length is no less than 5 grid units. The two endpoints of each straight line segment are marked with a solid black circle with a radius of 6 pixels. The polygonal lines at the equipment edge reflect the arrangement of linear components such as busbars, beams, and fence posts. The endpoint matching of the straight line segments provides reliable geometric constraints for attitude refinement search. The endpoint-to-endpoint matching rules are marked in the figure: endpoint distance ≤ 2 grid units, and the angle difference between straight line segments ≤ 10 degrees. This extraction method remains stable under changes in lighting and surface contamination.

[0027] After generating the reflection intensity fingerprint, a bidirectional index is established between the reflection intensity hotspots and the inspection route coordinates. Specifically, the coordinates of the four corner points of the circumscribed rectangle of each reflection intensity hotspot, the coordinates of its geometric center, and the coordinates of its intersection with the inspection route are recorded in the station's planar coordinate system. A unique identifier is assigned to each reflection intensity hotspot. This index is used to quickly prune the candidate set based on the planar position of the GNSS-RTK weak anchor during subsequent entry determination, and also to directly locate the section of the route that the hotspot crosses upon triggering.

[0028] The GNSS-RTK weak anchor is activated synchronously at the start of initial data acquisition. The GNSS-RTK weak anchor receives correction information from the base station via a differential data link and outputs its planar position and heading at a frequency of 1 Hz. The output planar position is transformed to the in-station planar coordinate system, and the heading is unified using the difference angle between the geographic azimuth and the vehicle coordinates. The GNSS-RTK weak anchor serves two purposes in this phase: first, it establishes a common absolute reference for the reflection intensity hotspot and the inspection route during initial data acquisition, establishing a fixed association between the reflection intensity fingerprint and spatial position; second, it limits the candidate range and performs alignment correction in the subsequent decision-making and trajectory playback phases. The reason for adopting the weak anchor strategy is that the planar position of GNSS-RTK in open scenes has a stability characteristic ranging from meters to decimeters. This absolute reference is sufficient to limit the candidate range and suppress large-scale drift. Simultaneously, attitude and fine-grained positioning are handled by intensity information and geometric features, avoiding excessive constraints in structural occlusion and multipath environments, thus maintaining control over intensity alignment and sequence consistency.

[0029] In another optional implementation, to improve adaptability to local glare and shadows, the threshold segmentation window can be set to 9×9 or 15×15, and the threshold increment can be set to 8 or 12 respectively. Other steps remain consistent to complete the hotspot screening and reflection intensity fingerprint generation. In another implementation, the extraction of device edge polylines can be achieved by using Canny edge detection followed by least-squares line fitting to obtain straight line segments. The threshold and length conditions remain consistent with the aforementioned to ensure compatibility with the matching rules subsequently used in the judgment process.

[0030] Example 2: Step S2, online entry detection and triggering, includes: during formal inspection, fusing continuous laser intensity data to form the current intensity segment, determining candidate reflection intensity hotspots based on the planar position given by the GNSS-RTK weak anchor, performing entry judgment on the candidate reflection intensity hotspots, and triggering the reflection intensity hotspot event and setting the intelligent trajectory playback window when the judgment is passed and the GNSS-RTK weak anchor meets the robustness standard.

[0031] In one specific implementation, the formation of the current intensity segment employs a combination of temporal micro-fusion and motion compensation. The inspection robot moves forward at a stable linear velocity, and a multi-line lidar continuously outputs echo reflection intensity. A fusion batch is selected with consecutive frames covering 0.50 seconds. At the generation of each frame, a planar rigid transformation is performed on the point cloud between the previous and current frames based on the relative pose given by the wheel speedometer and inertial information, aligning the echoes of each frame to the current moment within the same station's planar coordinate system. The aligned echoes of each frame are plotted on a 0.20m × 0.20m grid. Each grid may receive multiple writes within the fusion batch, and the grayscale value of the grid is obtained using an arithmetic average to generate the grayscale raster image of the current intensity segment. The combination of motion compensation and averaging fusion reduces the instantaneous fluctuations of high-brightness areas on the metal surface caused by viewing angle jitter, allowing the stable, highly reflective structure of the equipment group to form smoother, more distinguishable bright areas within a short timescale, thereby improving the robustness of subsequent template overlap and edge polygon matching.

[0032] The determination of candidate reflection intensity hotspots relies on the planar position provided by the GNSS-RTK weak anchor. This planar position is transformed into the in-station planar coordinate system, and a circular search area is constructed with this position as the center and a radius of 25 meters. Using the bidirectional index established for each reflection intensity hotspot in step S1, reflection intensity hotspots whose geometric center or circumscribed rectangle intersects with this circular area are searched, resulting in a candidate set. Limiting the search range to a 25-meter radius significantly reduces the computational load for entry determination while maintaining coverage of actual entry events. In open scenes, the planar position provided by the GNSS-RTK weak anchor has a stable absolute reference. Using this reference as a spatial clipping basis allows the determination focus to be placed on device groups adjacent to the current location, avoiding interference from distant similar structures.

[0033] Template overlap determination involves a sliding comparison of the grid intensity templates for candidate reflective intensity hot zones. First, the grayscale raster image of the current intensity segment is translated and slid across the coordinates of the candidate grid intensity template in steps of one grid. Each sliding position constitutes a comparison. The absolute value of the grayscale difference between each grid in the template area covered by the current intensity segment is taken and compared with a fixed difference threshold. The number of grids whose absolute difference exceeds this threshold is counted and divided by the total number of grids participating in the comparison to obtain the proportion. The minimum proportion among all sliding positions is selected as the template overlap index for the current candidate. If this index is within 35, the candidate is deemed to have passed this level of determination. The sliding comparison method can find the best overlapping translation position under small attitude and position deviations, ensuring maximum consistency in the grayscale space of the high-reflectivity main area of ​​the equipment group. Setting the proportion threshold within 35 can accommodate local occlusion and illumination changes, while excluding weak correlation matching caused by the background ground and vegetation.

[0034] Edge polyline alignment is performed after template overlap is checked. First, the horizontal and vertical gradient magnitudes are calculated on the grayscale raster image of the current intensity segment using the Sobel operator. Binarization is performed using a threshold of 30. Probabilistic Hough line detection is used to extract line segments with a length of at least 5 grids, resulting in the set of device edge polylines for the current intensity segment. Then, eight orientation candidates are generated, with center angles of 0 degrees, 22.5 degrees, 45 degrees, 67.5 degrees, 90 degrees, 112.5 degrees, 135 degrees, and 157.5 degrees. The grid intensity template of the current intensity segment relative to the candidate reflection intensity hot zone is rotated sequentially around the template center to the eight orientation candidates within the raster. Endpoint-to-endpoint matching is performed on the device edge polyline sets on both sides under each orientation. The matching rules are: the Euclidean distance between endpoints must be within 2 grids, and the angle difference between corresponding line segments must be within 10 degrees. The number of matching pairs satisfying the rules is counted. If any orientation candidate has 3 or more matching pairs, it is determined to pass this level of judgment. By using directional quantization to provide 8 orientation candidates, common equipment layout orientations can be covered within a limited rotation enumeration. The constraints of endpoint distance and angle difference ensure that the matching is for the same physical straight line rather than a random short line segment. The number of matching pairs reaches 3, which has stable feasibility in combined structures such as frames, fences and busbars.

[0035] The intensity sequence code consistency determination is performed after edge-line alignment. First, based on the bidirectional index established in step S1, the crossing line segment between the centerline of the inspection route and the candidate reflection intensity hot zone is obtained. Every 0.20 meters along this crossing line segment, a grayscale value is sampled on the grayscale raster image of the current intensity segment. The sampled values ​​are encoded as 0, 1, 2, 3, and 4 according to the intervals [0,50), [50,100), [100,150), [150,200), and [200,255], respectively, forming the intensity sequence code of the current intensity segment. The stored intensity sequence code is retrieved from the reflection intensity fingerprint of the candidate reflection intensity hot zone as a reference. A two-dimensional array is used for sequence alignment; the row index of the two-dimensional array traverses the intensity sequence code of the current intensity segment, and the column index traverses the intensity sequence code of the candidate reflection intensity hot zone. The values ​​in the two-dimensional array are filled in from left to right and from top to bottom: when the row index and column index codes are equal, the value of the upper left cell is incremented and written to the current cell; when the codes are unequal, the larger value between the left and upper cells is written to the current cell. The value in the lower right corner of the two-dimensional array is the length of the longest common subsequence of the two sequences. The ratio of this length to the length of the candidate reflection intensity thermal zone intensity sequence code is calculated. If the ratio reaches 80 or above, the candidate is judged to have passed this level of judgment. The intensity sequence code along the route direction can faithfully reflect the repetition, interval, and occlusion relationship between the equipment group and the passageway in the route direction. Using the longest common subsequence instead of point-by-point equality comparison can accommodate the coding deviation caused by slight changes in the posture of individual sampling points, while maintaining sensitivity to the overall sequence structure.

[0036] like Figure 4As shown, this figure illustrates the process of calculating the longest common subsequence (LCS) of two intensity sequence codes using a dynamic programming algorithm. The upper part of the figure labels the current segment intensity sequence code "432014321431204" (length 15) and the hotspot template intensity sequence code "4320132143" (length 10), along with the encoding rules: grayscale values ​​[0, 50) are encoded as 0, [50, 100) as 1, [100, 150) as 2, [150, 200) as 3, and [200, 255] as 4. The main body of the figure is an (11×16) dynamic programming matrix. The row indices correspond to the hotspot template sequence (11 rows including the empty character ε), and the column indices correspond to the current segment sequence (16 columns including the empty character ε). Each cell is 70×70 pixels in size and displays the dp value at that position, using bold Arial font size 26. The first row and first column of the matrix are both 0, indicating that the LCS length of an empty sequence and any sequence is 0. Starting from the second row and second column, the matrix is ​​filled according to a dynamic programming recursive formula: when the characters at row index i and column index j are equal, dp[i][j] = dp[i-1][j-1] + 1; when the characters are unequal, dp[i][j] = max(dp[i-1][j], dp[i][j-1]). The positions of character matches in the diagram are marked with a green background (rgba(100,200,100,0.3)), and there are 10 character matching positions scattered throughout the matrix. The value in the lower right corner of the matrix is ​​9, indicating that the length of the longest common subsequence (LCS) of the two sequences is 9. The backtracking path of the LCS is marked with a thick red line (4 pixels wide) in the diagram. Backtracking starts from the bottom right corner of the matrix and proceeds according to the following rules: if the character at the current position matches, move to the top left; if not, move in the direction with the larger dp value (left or top). Cells traversed by the backtracking path are marked with a light red background (rgba(255,200,200,0.4)). The lower labeled area of ​​the graph displays the calculation results: longest common subsequence length 9, template sequence length 10, consistency ratio 90.0%. Since this ratio reaches 80% or higher, it is judged as "judgment passed (≥80%)" and highlighted with a green background and green text.

[0037] When a candidate reflection intensity hotspot meets the aforementioned criteria in template overlap determination, edge polyline alignment determination, and intensity sequence code consistency determination, a robustness verification of the GNSS-RTK weak anchor is performed simultaneously. The robustness verification includes four indicators: the solution state is fixed or floating, the standard deviation of the horizontal position is within 0.30 meters, the standard deviation of the heading is within 5 degrees, and the time interval between data points within the last 3 seconds is within 1 second. Meeting all four indicators confirms that the GNSS-RTK weak anchor is in a reliable state at that moment. Passing the Level 3 entry determination and passing the GNSS-RTK weak anchor robustness verification are considered as successful entry. Upon successful entry, a reflection intensity hotspot event is immediately triggered, and an intelligent trajectory playback window is set on the centerline coordinates of the inspection route, with the trigger point as the reference. The window starts 10 meters in front of the trigger point and ends 15 meters behind the trigger point. This window covers both the section before and after entry, allowing subsequent rollbacks and playbacks to propagate the constraints of the hot zone anchor pose to both the entry and exit points, forming a continuous, smooth interval.

[0038] In one alternative implementation, the temporal micro-fusion batch of the current intensity segment comprises 5 frames, covering 0.50 seconds, with an absolute difference threshold of 30. This combination can suppress isolated weak noise points without introducing significant motion blur when there is fog or vapor interference on the device surface, and the template overlap ratio is more stable. In another implementation, to accommodate higher vehicle speeds or denser device arrangements, the temporal micro-fusion batch can comprise 3 or 7 frames, corresponding to coverage times of approximately 0.30 seconds or 0.70 seconds, respectively. Using 3 frames is suitable for reducing the accumulation of motion compensation errors when quickly passing through narrow passages; using 7 frames is suitable for improving the stability of strongly reflective structures in open road sections. The above variations can make trade-offs based on field speed and environmental transparency without changing the three judgment criteria of template overlap, edge polyline alignment, and intensity sequence code consistency, as well as the GNSS-RTK weak anchor robustness standard.

[0039] In another alternative implementation, endpoint matching for edge-polyline alignment can first perform endpoint clustering on the rotated current intensity segment, with a cluster radius of 1 grid. The cluster center is then matched one-to-one with the endpoints of candidate reflection intensity hot zones, still using the criteria of endpoint distance within 2 grids, angle difference within 10 degrees, and at least 3 matching pairs. Endpoint clustering reduces multiple counting caused by repeated straight segments, allowing matching to concentrate on structurally strong long straight segments, thus improving the reliability of orientation candidate determination.

[0040] Example 3: Step S3, intensity alignment and local fingerprint relocation, includes: after triggering the reflection intensity hot zone event, aligning the current intensity segment with the reflection intensity fingerprint of the triggered hot zone, obtaining the local fingerprint relocation pose through coarse alignment and layer-by-layer search, and setting the obtained local fingerprint relocation pose as the hot zone anchor pose.

[0041] To ensure alignment of the two types of data within the same spatial scale, the current intensity segment is projected onto a 0.20m × 0.20m grid according to the station's planar coordinate system and clipped within the bounded rectangle of the trigger hot zone. The reflection intensity fingerprint of the trigger hot zone includes data structures such as the grid intensity template, the outer contour polyline, and the device edge polyline. The alignment process is primarily driven by the grid intensity template, while the outer contour polyline and the device edge polyline provide geometric consistency verification and orientation refinement criteria after coarse alignment. The reason for choosing the grid intensity template as the primary method is that strong reflective components are less sensitive to attitude changes in a short timescale than edge features. By first using the intensity template to absorb most of the translation and orientation errors, and then using edge features to improve the geometric fit of the final solution, convergence stability can be maintained under conditions of repetitive metal structures and local occlusion.

[0042] like Figure 3As shown, this figure illustrates the process of edge polyline endpoint matching in one of eight orientation candidates (22.5 degrees). The figure shows the matching relationship between the template edge polyline and the current fragment edge polyline after rotation transformation. The template edge polyline in the figure is represented by a black dashed line and includes 10 straight line segments distributed within the region of X coordinates 300-600 and Y coordinates 350-900. These straight line segments originate from the device edge polyline data structure in the reflection intensity fingerprint stored in step S1. The current fragment edge polyline in the figure is represented by a black solid line and also includes 10 straight line segments. These straight line segments are obtained by calculating the gradient magnitude on the grayscale raster image of the current intensity fragment using the Sobel operator, binarizing it with a threshold of 30, and extracting straight line segments with a length of no less than 5 grids using probabilistic Hough line detection. The current fragment is rotated 22.5 degrees relative to the template around the center point (coordinates 600, 600), and this rotation is achieved through a planar rigid transformation. Ten pairs of matching endpoints are marked with black solid circles (radius 8 pixels) in the figure. Each pair of endpoints is connected by a gray dashed line, indicating a matching relationship. The distance between endpoints is calculated using Euclidean distance; for example, "d=15.8" means the distance between the endpoints is 15.8 pixels (equivalent to 1.58 grid units), satisfying the matching condition of endpoint distance ≤ 2 grids. The right-hand labeled area of ​​the figure displays matching statistics: 10 template edge polylines, 10 current segment edge polylines, 10 matching pairs, and ≥ 3 successful matches. The criteria for successful matching are highlighted with a green background. The figure also lists 8 complete orientation candidates: 0 degrees, 22.5 degrees, 45 degrees, 67.5 degrees, 90 degrees, 112.5 degrees, 135 degrees, and 157.5 degrees, with 22.5 degrees (the currently used orientation) marked with a gray background. The endpoint matching conditions are clearly marked as: endpoint Euclidean distance ≤ 2 grids, corresponding line segment angle difference ≤ 10 degrees, and at least 3 successful matches.

[0043] Coarse alignment employs a finite candidate method. First, 16 orientation candidates are generated, with center angles of 0°, 22.5°, 45°, 67.5°, 90°, 112.5°, 135°, 157.5°, 180°, 202.5°, 225°, 247.5°, 270°, 292.5°, 315°, and 337.5° respectively. These orientation candidates cover a full circumference, and a uniform initial attitude set is formed through equally spaced sampling. Then, using the planar position given by the GNSS-RTK weak anchor as the initial translation value, the current intensity segment is rotated around its geometric center to each of the 16 orientation candidates, and a grid-by-grid comparison is performed with the mesh intensity template under each orientation. The comparison method is as follows: For the overlapping area of ​​the two in the template coordinates, calculate the absolute difference in grayscale for each grid, and count the proportion of grids with an absolute difference exceeding 30; select the orientation corresponding to the minimum value of this proportion from 16 orientations, and use the planar position of the GNSS-RTK weak anchor as the translation to obtain the coarse alignment pose. The reason for using this method is that by first covering the entire area with uniform orientation candidates and then selecting the best candidate based on the intensity difference proportion, the consistency of the main structure can be preserved even in the presence of small-scale occlusion and local glare; using the planar position of the GNSS-RTK weak anchor as the initial translation value allows the translation search to revolve around a reliable absolute reference, reducing the scope and number of subsequent refinement searches.

[0044] Layer-by-layer search refines translation and orientation near the coarse-aligned pose. A neighborhood set of level 1 is constructed centered on the coarse-aligned pose. The neighborhood set uses a 0.10-meter offset step size in the translation dimension, combining the translation and orientation candidates along the forward direction and laterally to generate eight-neighbor translation candidates; and uses a 1-degree offset step size in the orientation dimension, generating 1-degree orientation candidates to the left and right. These translation and orientation candidates are then combined to obtain the pose candidate set for that layer. For each pose candidate, first calculate the ratio of the absolute grayscale difference with the mesh intensity template, and select candidates with a ratio not greater than 35 for geometric verification. For candidates entering geometric verification, extract the device edge polyline of the current intensity segment under the candidate pose. The method is to calculate the gradient magnitude on the grayscale raster image using the Sobel operator, set the threshold to 30 for binarization, and use probabilistic Hough line detection to extract line segments with a length of not less than 5 grids. Then, perform endpoint-to-endpoint matching with the device edge polyline in the triggered hot zone reflection intensity fingerprint. The matching conditions are that the endpoint spacing is not greater than 2 grids and the angle difference between the corresponding line segments is not greater than 10 degrees. Count the number of matching pairs that meet the conditions. Use the candidate with the largest number of matching pairs as the new center pose, construct a neighborhood set with the level increased by 1, and repeat the above process. The search is terminated when the number of matching pairs in two consecutive levels does not increase. The center pose at the time of termination is the local fingerprint relocation pose. Layer-by-layer search uses the grayscale absolute difference ratio as the threshold for geometric verification, which can mask occasional local noise. Using the number of endpoint matches as the update criterion guides the search to the position and orientation most consistent with linear components. Linear components are stable in the long term in the substation scenario; the larger the number of matches, the stronger the physical consistency of the alignment. Using offset step sizes of 0.10 meters and 1 degree, sufficient search accuracy can be achieved under common equipment spacing and structural orientation resolution, while ensuring that the computational load remains within a controllable range.

[0045] To improve the ability to distinguish critical attitudes, pure translations along the forward and lateral directions are evaluated first within the eight-neighborhood of translation candidates, followed by diagonal translations combining both. This evaluation order allows the search to converge to candidates closer to the center first along the two principal orthogonal directions, and then fine-tunes with diagonal offsets, reducing error cancellation caused by adjusting both directions simultaneously. In the evaluation of orientation candidates, the side with the same sign and coarse alignment is evaluated first, followed by the other side. This order can find the endpoints of edge polygons with consistent orientations more quickly in the relatively fixed main orientation of the road environment within the station, making it easier to increase the number of matching pairs.

[0046] After the local fingerprint relocation pose is determined, this pose is set as the hotspot anchor pose and recorded in the form of in-station planar coordinates and orientation angle. The record includes the decimal part of the planar position and the decimal part of the orientation angle, so that subsequent intelligent trajectory playback smoothness and long-term drift suppression have precise starting point constraints within the window. In order to maintain consistent positioning quality across different triggering hotspots, the hotspot anchor pose, trigger timestamp, unique identifier of the triggering hotspot, and hierarchical number used to terminate the search are archived together for easy subsequent statistics and on-site verification.

[0047] In one alternative implementation, the maximum search level is set to 5. When the number of matching pairs still increases at level 5, the center pose of level 5 is output as the local fingerprint relocation pose. This upper limit setting ensures sufficient search depth in hotspots with complex device layouts and achieves convergence at levels 3 or 4 in most hotspots. This embodiment exhibits high orientation stability and small translational residuals when traversing large groups of devices on open roads.

[0048] In another alternative implementation, the evaluation order of the 16 orientation candidates during the coarse alignment stage is rearranged according to the peak value of the edge direction histogram of the current intensity segment. The four orientations closest to the peak value are evaluated first, followed by the remaining orientations. This utilizes the most significant linear direction in the current intensity segment, accelerating convergence and reducing the number of invalid comparisons. The edge direction histogram uses a 22.5-degree bucket width to statistically analyze the directional distribution of the device edge polygons. The peak position is usually consistent with the main direction of the architecture, so prioritizing evaluation can more quickly find candidates with a smaller absolute grayscale difference ratio.

[0049] In another alternative implementation, endpoint-to-endpoint matching employs a grid-based nearest neighbor search. First, the endpoint coordinates of the edge polyline of the triggering hot zone device are quantized to integer grids with a grid step size of 0.20 meters. Then, a hash map indexed by this grid is constructed in the current intensity segment coordinates. The matching process searches for corresponding endpoints within the same grid and its eight neighboring grids. This method aligns with the condition that the endpoint spacing is no greater than two grids, enabling constant-time nearest neighbor queries without building a complex tree structure, and is suitable for maintaining stable computational latency when multiple triggers occur within the same hot zone.

[0050] In another alternative implementation, when multiple candidates with the same number of matching pairs appear, the consistency length of the intensity sequence code is used as the sorting criterion for parallel candidates when selecting the local fingerprint relocation pose. The longest common subsequence length is calculated along the centerline of the inspection route between the current intensity segment and the intensity sequence code of the trigger hot zone, with candidates having longer lengths given priority. This allows for differentiation based on the order of intensity fluctuations along the path direction, even when edge geometry is equally consistent, resulting in a pose more consistent with the actual traversal pattern.

[0051] Step S4, intelligent trajectory playback smoothing and long-term drift suppression, includes: within the intelligent trajectory playback window, replacing the trigger point pose with the hot zone anchor pose, and applying the motion increment within the window to reconstruct the pose sequence, performing corrections during the reconstruction process, and finally generating the final positioning result through boundary stitching.

[0052] In the specific implementation process, the preparation of the intelligent trajectory playback window is based on the range set in step S2, with the trigger point as the reference. The starting point of the window is located 10 meters in front of the trigger point, and the ending point of the window is located 15 meters behind the trigger point. The wheel speed and inertia increment within this range are extracted according to the timestamp order at the time of acquisition, while maintaining the original time order. The pose corresponding to the trigger point is replaced with the hot zone anchor pose obtained in step S3 to form the starting reference for reconstruction within the window. Choosing the hot zone anchor pose as the only starting point within the window is beneficial for propagating the high-reliability spatial alignment result within the reflection intensity hot zone to both ends of the window, so that the pose before and after entry is continuous under the same coordinate reference. The pose sequence reconstruction applies the wheel speed and inertia increment sequentially in the positive time direction. Each time advance starts from the pose of the previous moment. First, the travel distance of the time step is calculated by the wheel speed increment, and then the orientation change of the time step is calculated by the inertia increment. The travel distance is updated along the current orientation direction to update the planar position, and the orientation is updated according to the orientation change to obtain the reconstructed pose of the time step. This process is equivalent to progressively unfolding a continuous trajectory within a window based on the motion data provided by the sensors, so that the final trajectory uses the actual motion data as its framework. The advantage of reconstructing in this way is that the sensor increments directly reflect the microscopic motion of the chassis, and the trajectory formed after continuous superposition naturally exhibits the smooth characteristics of the vehicle body. Subsequent corrections can achieve the target constraints by making minor adjustments based on this trajectory.

[0053] After each application of wheel speed and inertia increment, three correction processes are performed sequentially: passageway correction, orientation change correction, and GNSS-RTK alignment correction. Each of the three corrections targets a different source of error and works synergistically.

[0054] The passageway correction addresses the geometric constraints of the passageway within the window. Based on the engineering layout drawing or pre-imported route design data, a 1.50-meter-wide passageway is established on both sides of the centerline of the inspection route. The geometric relationship between the current reconstructed pose planar position and the boundary of the passageway is calculated. When the planar position is inside the passageway, it remains unchanged; when the planar position is outside the passageway, it is projected along the centerline normal direction to the nearest passageway boundary point, keeping the current orientation unchanged. This correction is introduced because the inspection robot operates within the predetermined route of the substation, and the passageway reflects the actual passage area of ​​the maintenance path. Limiting the reconstructed position to within the passageway eliminates the out-of-bounds offset caused by lateral drift accumulation, ensuring that the positioning results consistently match the passable area, and subsequent adjustments to the orientation to GNSS-RTK alignment are also performed within the legal space.

[0055] Orientation change correction is based on chassis kinematic constraints, calculating the magnitude of the orientation change between the current reconstructed pose and the previous reconstructed pose. When the orientation change is within 10 degrees, the change is maintained; when the change exceeds 10 degrees, the magnitude is replaced with 10 degrees, while retaining the same steering sign as the original orientation change. This correction is constrained by the chassis's achievable steering capability within a short time step, avoiding excessively large steering angles in a very short time. Using 10 degrees as the upper limit for a single step allows for a smooth orientation curve to be formed under conditions of 10 to 20 laser outputs per second and wheel speed and inertia updates at the 100 Hz level, reducing severe oscillations caused by instantaneous sensor noise and stabilizing attitude updates within the vehicle's achievable turning range.

[0056] GNSS-RTK alignment correction performs a one-time position alignment when encountering a GNSS-RTK weak anchor stabilization segment. It determines whether the current time point belongs to a GNSS-RTK weak anchor stabilization segment, based on four criteria: the solution state is fixed or floating, the standard deviation of the horizontal position is within 0.30 meters, the standard deviation of the heading is within 5 degrees, and the time interval between data points within the last 3 seconds is within 1 second. If these four criteria are met, the current reconstructed attitude plane position is moved along a straight line to the plane position provided by the GNSS-RTK weak anchor, maintaining continuity between the current orientation and altitude and the values ​​of the preceding and following time points. Using a one-time alignment rather than a gradual approximation allows for the immediate absorption of accumulated translational errors upon the arrival of the stabilization segment, introducing an absolute reference into the window so that the remaining time periods of the subsequent window are based on this absolute reference. Maintaining orientation and altitude continuity avoids introducing abrupt attitude changes during alignment, ensuring that the trajectory curve exhibits a continuous tangential direction at the alignment point.

[0057] After reconstructing and performing three corrections for all time points within the window, boundary stitching is performed to ensure continuity between the trajectory within and outside the window in the transition region. directional continuity stitching is then applied to the starting point of the window and the upstream pose outside the window. Trajectory segments within a 2-meter radius on both sides of the starting point are taken, and the angle difference between the upstream and starting directions is calculated. This angle difference is gradually transitioned linearly according to spatial distance, ensuring a smooth transition of direction to the starting direction within the 2-meter radius. Directional continuity stitching addresses potential directional steps that may occur after replacing the trigger point pose with a hot zone anchor pose. By distributing the angle difference across the distance dimension, the vehicle completes the directional transition within 2 meters of entering the window, resulting in continuous trajectory tangents. Position continuity stitching is then applied to the ending point of the window and the downstream pose outside the window. Trajectory segments within a 2-meter radius on both sides of the ending point are taken, and the planar offset vectors at both ends are calculated. This offset is gradually distributed linearly according to time sequence, ensuring a smooth transition of position to the downstream position within the 2-meter radius. Position continuity stitching addresses the absolute position offset formed after GNSS-RTK bonding within the window. By distributing the offset over a 2-meter length, the spatial position of the trajectory remains continuous, avoiding breakpoints at the ending point. After completing the boundary stitching at both ends, the final positioning result is output, covering the window range and smoothly transitioning at both ends.

[0058] In one alternative implementation, wheel speed and inertia increments within the intelligent trajectory playback window are applied with update frequencies of 100 Hz and 200 Hz. Corridor correction is performed once per time step, orientation change correction is performed once per time step, and GNSS-RTK alignment correction is performed once at the point where a stable segment is identified. In this embodiment, when passing through large equipment groups in open sections, the lateral offset within the window is promptly projected to the boundary by corridor correction. The orientation change curve exhibits subtle changes over consecutive time steps. Upon encountering a stable segment, the planar position aligns with the position provided by the GNSS-RTK weak anchor within one time step, and the positional offset at the end of the window smoothly transitions within a 2-meter range. For narrow passageways in densely populated equipment areas, the width of the corridor can be preset to 1.20 meters before entering the section, allowing lateral offsets to be projected to the boundary earlier. Continuous orientation stitching and continuous position stitching maintain a linear interpolation setting within a 2-meter range. This method creates stronger lateral constraints in compact corridors, resulting in smaller fluctuations in the reconstructed trajectory along both sides of the centerline within the window.

[0059] In another optional implementation, the trigger selection for GNSS-RTK alignment correction employs two strategies: the first time point within the stable segment or the median time point within the stable segment. Selecting the first time point allows for the earliest introduction of an absolute reference, suitable for short windows or situations where the equipment group boundary has just passed. Selecting the median time point can offset local deviations in individual measurements within the stable segment, suitable for situations where the stable segment length is greater than 2 seconds and the window covers most of the stable segment's duration. Both strategies are executed under the premise of one-time alignment and maintaining orientation and altitude continuity. Through the above implementation, step S4 uses the hot zone anchor positioning posture as the starting point of the window, unfolds the posture sequence according to wheel speed and inertia increments within the intelligent trajectory playback window, and introduces constraints for corridor correction, orientation change correction, and GNSS-RTK alignment correction at each step. Finally, continuous transition of direction and position is completed at both ends of the window through boundary stitching. This process forms a stable and consistent reference and geometric constraints in open and long-distance inspection routes, suppressing cumulative drift inside and outside the window, and providing reproducible positioning results for subsequent inspection cycles.

[0060] like Figure 5As shown in the figure, this diagram illustrates the effect of corridor correction on the reconstructed pose within the intelligent trajectory playback window. A two-dimensional coordinate system is established in the figure. The horizontal axis represents the route distance in meters, ranging from 0 to 50 meters; the vertical axis represents the lateral offset in meters, ranging from -2.0 to +2.0 meters. The 0-meter line represents the centerline of the inspection route. The two boundaries of the corridor are marked with black dashed lines (3 pixels wide, dashed style [16,8]): the upper boundary is located at +0.75 meters, and the lower boundary is located at -0.75 meters. The area between the two boundary lines is filled with light blue semi-transparent material (rgba(200,220,255,0.2)), representing a corridor with a width of 1.50 meters. This corridor width is set on both sides of the centerline of the inspection route based on the engineering layout drawing or pre-imported route design data. The gray dashed line (3 pixels wide, dashed style [8,4]) in the figure represents the trajectory before correction, which includes cumulative drift and instantaneous noise. In the 0-20 meter range, the lateral offset fluctuates within ±0.5 meters; in the 20-30 meter range, the lateral offset exceeds the upper boundary, with a maximum offset of approximately +1.5 meters; in the 35-42 meter range, the lateral offset exceeds the lower boundary, with a maximum offset of approximately -1.4 meters. These areas exceeding the boundaries are marked with a red semi-transparent shading (rgba(255,0,0,0.15)), clearly showing the boundary violations of the trajectory before correction. The black solid line (4 pixels wide) in the figure represents the trajectory after correction, which is strictly confined within the passageway. When the trajectory before correction is located inside the passageway, the corrected position remains unchanged; when the trajectory before correction is located outside the passageway, the position is projected onto the nearest passageway boundary point along the centerline normal direction. The figure marks two typical projection locations with red dashed lines (2 pixels wide, [4,4]): at a distance of approximately 25 meters from the route, the original position was about +1.3 meters outside the upper boundary, corrected to +0.75 meters above the upper boundary via vertical projection; at a distance of approximately 38 meters from the route, the original position was about -1.2 meters outside the lower boundary, corrected to -0.75 meters above the lower boundary via vertical projection. The text "Projected to Boundary" is labeled next to the projection lines. The upper left corner of the figure illustrates the difference between the original trajectory (gray dashed line, including drift) and the corrected trajectory (black solid line, confined within the corridor) using a legend. This figure effectively demonstrates the suppressive effect of corridor correction on lateral drift, ensuring that the reconstructed pose conforms to the geometric constraints of the passable area over the long term, providing a legitimate spatial basis for subsequent orientation correction and GNSS-RTK alignment correction.

[0061] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be embraced within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims. Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices recited in the apparatus claims may also be implemented by a single unit or device in software or hardware. The terms "first," "second," etc., are used to indicate names and do not indicate any particular order.

Claims

1. A method for autonomous positioning of a laser navigation of an intelligent inspection robot for outdoor equipment of a booster station, characterized in that, Comprising the following steps: Step S1, reflection intensity hot zone construction and GNSS-RTK weak anchor initialization, comprising: in the initialization sampling, processing the echo reflection intensity data of the multi-line laser radar, constructing the reflection intensity hot zone containing the reflection intensity fingerprint, and starting the GNSS-RTK weak anchor to provide the absolute reference of the plane position and the heading; Step S2, online entry detection and triggering, comprising: in the formal inspection, fusing the continuous laser intensity data to form the current intensity segment, determining the candidate reflection intensity hot zone based on the plane position given by the GNSS-RTK weak anchor, performing the entry judgment on the candidate reflection intensity hot zone, when the judgment passes and the GNSS-RTK weak anchor meets the robustness standard, triggering the reflection intensity hot zone event and setting the intelligent trajectory playback window; Step S3, intensity alignment and local fingerprint repositioning, comprising: after triggering the reflection intensity hot zone event, aligning the current intensity segment with the reflection intensity fingerprint of the triggered hot zone, obtaining the local fingerprint repositioning pose through coarse alignment and layer-by-layer search, and setting the obtained local fingerprint repositioning pose as the hot zone anchor pose; Step S4, intelligent trajectory playback smoothing and long-term drift suppression, comprising: in the intelligent trajectory playback window, replacing the trigger point pose with the hot zone anchor pose, and applying the motion increment in the window to reconstruct the pose sequence, performing correction in the reconstruction process, and finally generating the final positioning result through boundary splicing.

2. The method of claim 1, wherein, The reflection intensity hot zone construction in step S1 specifically comprises: dividing the echo reflection intensity output by the multi-line laser radar in a single scanning period into a grid with 0.20 meters x 0.20 meters as the grid unit to form a gray-scale grid graph; performing threshold segmentation on the gray-scale grid graph, wherein the threshold value is equal to the local mean value with a 11x11 grid window plus 10, to obtain a binary graph; performing 1 time 3x3 square dilation and 1 time 3x3 square erosion on the binary graph in turn to complete the closing operation; extracting a plurality of connected domains according to the 4-adjacent connected domain; selecting the connected domain meeting all the following conditions as the reflection intensity hot zone: the number of grid units contained by the connected domain is between 200 and 5000, the bounding box aspect ratio is between 0.30 and 3.50, and the average polyline corner density of the connected domain boundary is between 3 times per 10 meters and 30 times per 10 meters.

3. The method of claim 2, wherein, The reflection intensity fingerprint generated in step S1 for each reflection intensity hot zone consists of the following four data structures: 1) grid intensity template: the gray scale raster map is cropped according to the hot zone bounding box and bilinearly interpolated to a 128x128 array; 2) outer contour polyline: the closed boundary is extracted on the binary map by contour tracing and the vertex coordinate sequence is recorded according to the grid length; 3) device edge polyline: the intensity gradient amplitude is calculated on the gray scale raster map by the Sobel operator and binarized by a threshold of 30, then the straight line segment with a length not less than 5 grids is extracted by the probabilistic Hough line detection and the endpoint coordinates are recorded; 4) intensity order code: the gray scale value is sampled every 0.20 meters on the polyline segment of the inspection route center line passing through the reflection intensity hot zone, and the interval [0, 50), [50, 100), [100, 150), [150, 200) is coded as 0, 1, 2, 3, 4 respectively to form a sequence.

4. The method of claim 1, wherein, The GNSS-RTK weak anchor started in step S1 is used to limit the search range of the candidate reflection intensity hot zone to the area with the plane position given by the GNSS-RTK weak anchor as the center and 25 meters as the radius in step S2, and is used for GNSS-RTK fitting correction of the reconstructed pose in step S4.

5. The method of claim 1, wherein, The entry determination performed on the candidate reflection intensity hot zone in step S2 is a three-level entry determination performed in sequence, including: the first level is template overlap degree determination: the current intensity fragment is slid on the grid intensity template of the candidate reflection intensity hot zone by 1 grid step and the sum of the absolute difference of each grid is calculated, when the proportion of the grid with an absolute difference exceeding the threshold is not more than 35%, it is determined to pass; the second level is edge polyline alignment determination: the edge polyline in the current intensity fragment is voted by histogram to get 8 orientation candidates, the 8 orientations are rotated in the grid and then matched with the device edge polyline set of the candidate reflection intensity hot zone by endpoint-to-endpoint matching, the matching rule is that the endpoint distance is not more than 2 grids and the included angle difference is not more than 10 degrees, and the number of matched straight line segments is not less than 3, when the matching rule is satisfied, it is determined to pass; the third level is intensity order code consistency determination: the intensity order code is extracted from the current intensity fragment in the direction of the inspection route center line and compared with the intensity order code of the candidate reflection intensity hot zone, when the length of the longest common subsequence of the two sequences accounts for not less than 80% of the length of the intensity order code of the candidate reflection intensity hot zone, it is determined to pass.

6. The method of claim 1 or 5, wherein, The condition for triggering the reflection intensity hot zone event in step S2 is that all three levels of entry determination pass, and at the same time, the robustness of the GNSS-RTK weak anchor is verified, the verification standard is that the solution state is fixed or floating, the horizontal position standard deviation is not more than 0.30 meters, the heading standard deviation is not more than 5 degrees, the data interval in the last 3 seconds is not more than 1 second, and the verification standard is met; the range of the intelligent trajectory playback window set in step S2 is: the starting point is 10 meters in front of the trigger point, and the end point is 15 meters behind the trigger point.

7. The method of claim 1, wherein, The intensity alignment in step S3 and the local fingerprint repositioning specifically include: forming an orientation set with 16 orientation candidates and forming a translation initial value set with the planar position of the GNSS-RTK weak anchor, generating a pose candidate set from the Cartesian combination of the orientation set and the translation initial value set; applying each pose candidate to the current intensity segment and performing a grid-by-grid comparison with the grid intensity template of the trigger hot area, selecting the pose with the least number of inconsistent grids as the coarse alignment pose; performing a layer-by-layer search around the translation and orientation of the coarse alignment pose with a discrete step length of 0.10 meters and 1 degree respectively, calculating the number of matching endpoints of the device edge polyline after each search and selecting the pose with the maximum number of matches as the new center until the number of matches of two consecutive layers of search does not increase, and outputting the local fingerprint repositioning pose.

8. The method of claim 1, wherein, The process of reconstructing the pose sequence in step S4 specifically includes: freezing the motion increments and time sequence of the smart trajectory playback window from the start point to the end point; replacing the pose corresponding to the trigger point with the hot area anchor pose obtained in step S3; and applying the wheel speed and inertial increment in the smart trajectory playback window in time sequence to reconstruct the pose sequence.

9. The method of claim 8, wherein, The correction performed after each application of the wheel speed and inertial increment in step S4 includes the following three correction processes: 1) passage corridor correction: setting a passage corridor with a width of 1.50 meters on both sides of the center line of the inspection route according to the engineering layout or the route design data imported in advance, and when the planar position of the reconstructed pose is outside the passage corridor, projecting the planar position of the reconstructed pose to the passage corridor boundary in the normal vector direction and keeping the orientation unchanged; 2) orientation change correction: when the orientation change amplitude of any two adjacent time points obtained by reconstruction is greater than 10 degrees, replacing the orientation change amplitude with 10 degrees and assigning the same turning sign; 3) GNSS-RTK fitting correction: when playing back to a time point containing a GNSS-RTK weak anchor stable section, moving the planar position of the time point along a straight line to the position provided by the GNSS-RTK and keeping the orientation and height value continuous along the original sequence.

10. The method of claim 1 or 8, wherein, The boundary splicing in step S4 specifically includes: linearly interpolating the start point of the smart trajectory playback window and the upstream pose outside the window within a range of 2 meters to realize direction continuity; and linearly interpolating the end point of the smart trajectory playback window and the downstream pose outside the window within a range of 2 meters to realize position continuity.

Images