Robot positioning navigation system based on three-dimensional laser radar and cross photoelectric correction

Abstract

The invention discloses a robot positioning and navigation system based on a three-dimensional laser radar and cross photoelectric correction, and the system comprises a navigation task starting module, a real-time data collection module, a positioning deviation calculation module, and a cooperative obstacle avoidance module. The auxiliary positioning robot constructs a platform track cooperative positioning reference; a robot in the real-time data acquisition module performs real-time data acquisition and synchronization; the positioning deviation calculation module is used for calculating the positioning credibility and adjusting sensor parameters and navigation paths based on the difference between the crossed photoelectric data and the standard database and the fitting degree of the image of the auxiliary robot and the positioning of the inspection robot; the cooperative obstacle avoidance module monitors station obstacles in real time, synchronous and safe start and stop of the double robots are ensured, high-precision positioning and dynamic correction are achieved through cooperation of the double robots, the positioning error is reduced, the subway night short-time window inspection requirement is met, and the inspection precision and efficiency are improved.

Description

Technical Field

[0001] This invention relates to the field of positioning for rail train inspection robots, specifically to a robot positioning and navigation system based on three-dimensional lidar and cross-photoelectric correction. Background Technology

[0002] With the rapid development of urban rail transit, the subway, as a core mode of public transportation, directly affects passenger safety and operational efficiency through the safety and reliability of its track system. Subway tracks are subjected to train loads, environmental erosion, and vibration impacts over long periods, making them prone to malfunctions such as loose fasteners, track deformation, and broken sleepers, requiring regular inspections and maintenance. The high-density nature of subway operations dictates that inspections are mostly carried out during nighttime shutdown periods, placing extremely high demands on inspection efficiency, positioning accuracy, and automation.

[0003] Currently, subway track inspection mainly relies on two types of technical solutions: one is manual inspection combined with handheld detection equipment. This solution is inefficient, labor-intensive, and susceptible to human experience, making it prone to missed or false inspections and unable to meet the inspection needs of large-scale track networks. The other is an automated inspection solution based on a single robot, which uses LiDAR, inertial navigation, or vision sensors mounted on the track robot for positioning and navigation.

[0004] However, existing automated inspection solutions suffer from significant technical bottlenecks: positioning accuracy is greatly affected by environmental interference. Subway tunnels are dimly lit, have high dust concentrations, and are subject to strong electromagnetic interference. When relying solely on the inspection robot's own 3D LiDAR for positioning, it is prone to positioning drift due to noise superposition in the track environment point cloud data and feature matching deviations. Although cross-electro-optical sensors can collect landmark features such as track fasteners and sleepers, they rely solely on comparing their own data with a standard database, lacking external independent benchmark verification. When local track features change due to wear or dirt, misjudgments of positioning confidence can easily occur, causing the inspection robot to deviate from the target task point. In existing solutions, if the inspection robot experiences positioning deviations, it often relies on backtracking calibration of the preset path or manual intervention from the background, resulting in slow correction response and limited accuracy. Although some solutions introduce auxiliary positioning equipment, the markers are easily damaged by track construction and environmental erosion, and cannot move dynamically with the inspection robot, making it difficult to achieve continuous positioning verification and real-time correction throughout the entire inspection path.

[0005] Therefore, in order to solve the problems existing in the prior art, this application proposes a robot positioning and navigation system based on three-dimensional lidar and cross photoelectric correction. Summary of the Invention

[0006] In view of the shortcomings of the existing technology, the purpose of this invention is to provide a robot positioning and navigation system based on three-dimensional lidar and cross photoelectric correction.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A robot positioning and navigation system based on 3D LiDAR and cross-electro-optical correction includes: The navigation task initiation module allows the inspection robot to receive the inspection task, plan a navigation path based on the preset track path and three-dimensional lidar scanning data, and send a follow command to the auxiliary positioning robot. The auxiliary positioning robot completes its own positioning and establishes the platform coordinate system through the preset positioning marks on the platform. The real-time data acquisition module allows the inspection robot to collect point cloud data of the track environment via a 3D lidar and track landmark feature data via cross photoelectric sensors. The auxiliary positioning robot moves synchronously following the inspection robot, collecting the appearance features of the inspection robot and the surrounding track features, as well as the real-time straight-line distance data between the robot and the inspection robot. The positioning deviation calculation module calculates the positioning confidence level of the inspection robot based on the difference between the cross photoelectric sensor data uploaded by the inspection robot and the preset track standard feature database, and the fitting degree between the image data uploaded by the auxiliary positioning robot and the positioning data of the inspection robot itself. If it is higher than the preset threshold, the inspection robot continues to move along the current navigation path; otherwise, it calculates the deviation compensation value and performs navigation system correction.

[0008] As a further improvement of the present invention, the navigation task initiation module includes: the platform preset positioning mark is an coded label containing mileage information; the auxiliary positioning robot obtains its own coordinates in the platform coordinate system by recognizing the coded label, and at the same time establishes a mapping relationship between the coded label and the preset reflective mark on the track side to complete the construction of a collaborative positioning reference between the platform and the track; the preset reflective mark on the track side is set at intervals along the outer side of the track, and the coded label is set along the edge of the platform corresponding to the position of the reflective mark; the auxiliary positioning robot takes pictures of the reflective mark through the image acquisition module, and combines the distance measurement module to measure the fixed distance between the coded label and the reflective mark to complete the calibration of the collaborative positioning reference.

[0009] As a further improvement of the present invention, the real-time data acquisition module includes the auxiliary positioning robot dynamically adjusting its own moving speed and direction based on the real-time positioning data of the inspection robot, ensuring that the inspection robot is always within the effective field of view of the auxiliary positioning robot image acquisition module, and that the distance between the auxiliary positioning robot and the inspection robot is maintained within a preset following range.

[0010] As a further improvement of the present invention, the track marking features include track fasteners, sleepers and track mileage markers. The cross photoelectric sensor collects image data of the marking features at a preset frequency and records the timestamp of the collection time. When the auxiliary positioning robot collects data, it records the same timestamp and completes the time synchronization between the inspection robot and the auxiliary positioning robot.

[0011] As a further improvement of the present invention, the positioning deviation calculation module includes: calculating the ratio of the deviation value between the data collected by the cross photoelectric sensor and the standard feature database of the track to a preset standard deviation threshold to obtain the photoelectric data difference rate; calculating the image fitting degree by the degree of matching between the feature image of the inspection robot collected by the auxiliary positioning robot and the image features converted from the positioning data of the inspection robot itself; and calculating the positioning confidence degree based on the photoelectric data difference rate and the image fitting degree.

[0012] As a further improvement of the present invention, the calculation of the image fitting degree includes: converting the pixel coordinates of the appearance features of the inspection robot in the image collected by the assisted positioning robot into actual coordinates in the platform coordinate system by combining the positioning data of the assisted positioning robot; converting the positioning data of the inspection robot itself into coordinates in the platform coordinate system; calculating the Euclidean distance between the two sets of coordinates; and determining the image fitting degree based on the ratio of the Euclidean distance to a preset distance threshold.

[0013] As a further improvement of the present invention, the deviation compensation value calculation includes: the auxiliary positioning robot continuously collects multiple frames of image data and distance data, removes abnormal data to obtain average image feature coordinates and average distance value; calculates the displacement deviation and angle deviation of the inspection robot's self-positioning and auxiliary verification positioning based on the average image feature coordinates, and corrects the displacement deviation in combination with the average distance value to obtain the final deviation compensation value.

[0014] As a further improvement of the present invention, a collaborative obstacle avoidance module is also included. The collaborative obstacle avoidance module is communicatively connected to the inspection robot and the auxiliary positioning robot, acquires images of the subway platform and determines the distance between the auxiliary positioning robot and the obstacle. If the distance between the auxiliary positioning robot and the obstacle is less than a safety threshold, the auxiliary positioning robot stops moving and feeds back its position information, and the inspection robot stops moving synchronously.

[0015] As a further improvement of the present invention, the navigation system calibration includes the following steps: after receiving the deviation compensation value, the inspection robot decomposes the compensation value into displacement correction and angle correction through its own positioning module; updates the current positioning coordinates in the navigation system based on the displacement correction, and adjusts the scanning reference angle of the three-dimensional lidar and the acquisition field of view angle of the cross photoelectric sensor in combination with the angle correction; the navigation system calls the calibrated positioning coordinates and sensor parameters to regenerate the navigation path to the target task point, and controls the drive module to move along the new path. During the movement, the calibrated data is collected in real time and fed back to the positioning confidence calculation stage to continuously verify the navigation system calibration effect until the movement state is stable and the positioning confidence is maintained within the preset acceptable range.

[0016] The beneficial effects of this invention are: improved positioning accuracy and reliability. Through a dual-robot collaborative architecture, it solves the problem of single-robot positioning being susceptible to tunnel dust and electromagnetic interference. Dual-robot trust verification improves positioning accuracy, thereby enhancing the accuracy of rail train inspection. By collaborating between the inspection robot and the auxiliary positioning robot, and using cross-sensor data combined with the auxiliary robot's image fitting degree to calculate positioning trust, along with a benchmark constructed from platform coding labels and track reflective markings, the problem of positioning drift caused by tunnel dust and electromagnetic interference is solved, preventing the accumulation of errors over long distances and significantly improving positioning reliability. Attached Figure Description

[0017] Figure 1 This is a block diagram of the robot positioning and navigation system based on three-dimensional lidar and cross-photoelectric correction of the present invention; Figure 2 This is a flowchart of the navigation task initiation and collaborative positioning reference construction process of the present invention; Figure 3 This is a flowchart of the real-time data acquisition and time synchronization process for dual robots in this invention; Figure 4 This is a flowchart of the positioning trust calculation and deviation judgment process of the present invention; Figure 5 This is a flowchart of the deviation compensation value calculation and correction process of the present invention. Detailed Implementation

[0018] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Identical components are denoted by the same reference numerals. It should be noted that the terms "front," "rear," "left," "right," "upper," and "lower" used in the following description refer to directions in the accompanying drawings, and the terms "bottom surface," "top surface," "inner," and "outer" refer to directions toward or away from the geometric center of a specific component, respectively.

[0019] This invention proposes a robot positioning and navigation system based on three-dimensional lidar and cross-photoelectric correction, comprising: The navigation task initiation module allows the inspection robot to receive the inspection task, plan a navigation path based on the preset track path and three-dimensional lidar scanning data, and send a follow command to the auxiliary positioning robot. The auxiliary positioning robot completes its own positioning and establishes the platform coordinate system through the preset positioning marks on the platform. The navigation task initiation module is the initial trigger and benchmark establishment unit for the system to carry out inspection operations. The inspection task received by the inspection robot not only includes the scope of the target inspection section, but also specifies the specific types of track components to be inspected, such as track fasteners, concrete sleepers, and rail joints, as well as the corresponding inspection accuracy requirements, ensuring that the inspection robot clearly understands the operation objectives and quality standards. The preset track path used by the inspection robot to plan the navigation path is a digital path model constructed in advance based on subway line design drawings, actual track laying parameters, and historical inspection data. This model contains key geometric information such as track mileage coordinates, curve curvature radius, and track gradient variation, providing a basic framework for initial path planning.

[0020] During path planning, the 3D LiDAR first scans the initial track environment around the inspection robot. By continuously emitting laser beams and receiving reflected signals from the track and surrounding structures, such as tunnel walls and track support devices, it generates 3D point cloud data containing spatial coordinates and reflection intensity information. The inspection robot's path planning unit compares this point cloud data with a preset track path model, identifies subtle deviations between the initial path and the actual track, and dynamically corrects the initial path based on the comparison results. This ensures that the planned navigation path closely matches the actual track direction, avoiding navigation deviations caused by differences between the preset model and the actual environment.

[0021] Simultaneously, the inspection robot sends a follow command to the auxiliary positioning robot. This command includes the inspection robot's initial positioning coordinates, expected speed, and the platform area corresponding to the current inspection route, ensuring the auxiliary positioning robot clearly understands the spatial boundaries and speed matching benchmarks for following. The auxiliary positioning robot completes its own positioning using pre-set positioning markers on the platform. These markers are visual identifiers spaced along the platform edge, each integrating a unique mileage code and made of wear-resistant and oil-resistant polymer material, maintaining clear visibility over a long period in complex environments such as dust and humidity on subway platforms.

[0022] The image acquisition unit of the assisted positioning robot captures images of these positioning markers. Using an image decoding algorithm, it extracts mileage information from the markers and combines this information with the distance traveled recorded by its onboard odometer to calibrate the initially acquired position information, ultimately determining its precise coordinates within the platform plane. Based on these precise coordinates, the assisted positioning robot establishes a platform coordinate system. This system uses a fixed reference point at the platform's starting end as the origin, a horizontal direction parallel to the platform's edge as the X-axis, and a horizontal direction perpendicular to the platform's edge and pointing towards the track as the Y-axis. This forms a unified spatial reference system for subsequent positioning verification, providing a benchmark for coordinate transformation between the inspection robot and the assisted positioning robot.

[0023] The real-time data acquisition module allows the inspection robot to collect point cloud data of the track environment via a 3D lidar and track landmark feature data via cross photoelectric sensors. The auxiliary positioning robot moves synchronously following the inspection robot, collecting the appearance features of the inspection robot and the surrounding track features, as well as the real-time straight-line distance data between the robot and the inspection robot. The real-time data acquisition module is responsible for acquiring multi-source data required for the collaborative positioning of the two robots. Through the collaborative work of the sensors of the two types of robots, it can achieve comprehensive acquisition of key information such as the track environment and robot position.

[0024] The inspection robot, equipped with a 3D LiDAR, continuously collects point cloud data of the track environment during its movement. Its scanning range covers the track surface, the fastening system on both sides of the track, sleepers, and the lower part of the tunnel wall. The LiDAR's scanning frequency is dynamically adjusted according to the robot's movement speed, ensuring sufficient density of point cloud data is acquired at different speeds. This data not only clearly presents the overall outline of the track but also captures subtle protrusions, depressions, or missing fasteners on the track surface, providing environmental reference for subsequent track fault detection and location correction.

[0025] The inspection robot's cross-electro-optical sensor focuses on collecting data on track landmark features. This sensor consists of a light transmitter and a receiver. The transmitter emits near-infrared light of a specific wavelength, which has strong penetrating power in dusty environments, reducing environmental interference. The receiver receives the light signals reflected from these track landmark features. These features include the outline of track fasteners, the cross-sectional dimensions and spacing of sleepers, and the mileage markings on the rail sides. These features, due to their stable geometric shapes and regular positional distribution, become core elements for positioning reference. The cross-electro-optical sensor analyzes the intensity changes and phase shifts of the received light signals to extract the geometric parameters and spatial position information of these features, ensuring that the collected data accurately reflects the actual state of the track features.

[0026] The assisted positioning robot collects data by synchronously following the inspection robot. Its following process uses dynamic position closed-loop control logic: the assisted positioning robot receives positioning data sent by the inspection robot in real time, calculates the relative position relationship between the two by combining its own position information, and if it detects a change in the movement speed of the inspection robot, the drive control unit of the assisted positioning robot will immediately adjust its own speed to ensure that the relative distance between the two remains stable along the track extension direction; at the same time, the visual monitoring unit of the assisted positioning robot will continuously track the position of the inspection robot. If it detects that the inspection robot is deviating from its own image acquisition field of view, it will promptly fine-tune its movement direction to ensure that the inspection robot is always within the effective range of image acquisition and avoid the interruption of positioning verification due to loss of field of view.

[0027] Regarding data acquisition, the image acquisition unit of the assisted positioning robot simultaneously collects two types of data: one is the robot's appearance features, including pre-set high-contrast geometric patterns on the robot's body. The relative positions of these high-contrast geometric patterns are pre-defined and uniquely identifiable, allowing the robot's specific location in the image to be determined. The other is the track features surrounding the robot, including reflective markings on the track side and the location of track seams. This data complements the track features collected by the robot, reducing the limitations of single-sensor acquisition. Furthermore, the laser rangefinder on the assisted positioning robot continuously collects the real-time straight-line distance to the robot. This device emits laser signals to specific reflective areas of the robot, records the time interval between the round trip, and calculates the straight-line distance by combining this with the laser's propagation speed in air. This distance data is used for subsequent positioning deviation correction calculations, improving positioning accuracy.

[0028] The positioning deviation calculation module calculates the positioning confidence level of the inspection robot based on the difference between the cross photoelectric sensor data uploaded by the inspection robot and the preset track standard feature database, and the fitting degree between the image data uploaded by the auxiliary positioning robot and the positioning data of the inspection robot itself. If it is higher than the preset threshold, the inspection robot continues to move along the current navigation path; otherwise, it calculates the deviation compensation value and performs navigation system correction.

[0029] The positioning deviation calculation module is the core unit for judging the positioning reliability of the inspection robot and realizing dynamic correction. Its operation logic revolves around data comparison, trust quantification and deviation correction.

[0030] This module first addresses two key data discrepancies: one is the difference between the cross-sensor data uploaded by the inspection robot and the preset track standard feature database. The preset track standard feature database is a digital storage system built upon a large amount of track feature data under normal conditions. It includes standard geometric parameters for different types of track fasteners, sleepers, and mileage markers, as well as the standard coordinate information of these features at different mileage positions. The database is periodically updated based on parameter changes after track maintenance to ensure the timeliness of the standard data. The module then compares the feature parameters collected by the cross-sensor with the standard parameters at the corresponding mileage positions in the database dimension-by-dimensional, analyzing indicators such as geometric dimension deviation and spatial position offset. The comprehensive results of these indicators reflect the degree of deviation between the data collected by the inspection robot and the standard state, serving as the basis for judging the accuracy of positioning.

[0031] Another type involves calculating the fit between image data uploaded by the assisted positioning robot and the inspection robot's own positioning data. This process first preprocesses the images acquired by the assisted positioning robot, using image enhancement algorithms to mitigate the impact of dust and lighting changes on image quality. Then, feature extraction algorithms extract the pixel coordinates of the inspection robot's appearance features. Combining the assisted positioning robot's precise position in the platform coordinate system with the intrinsic parameters of the image acquisition unit, a coordinate transformation algorithm converts the pixel coordinates into actual spatial coordinates in the platform coordinate system. Simultaneously, the module also transforms the inspection robot's calculated positioning coordinates to the same platform coordinate system. The fit is obtained by calculating the degree of spatial overlap between the two sets of coordinates. The higher the fit, the more reliable the inspection robot's self-positioning.

[0032] Based on the analysis results of the two types of data mentioned above, the module calculates the positioning confidence level of the inspection robot. The positioning confidence level is a quantitative indicator that comprehensively reflects the reliability of positioning. Its calculation process does not rely solely on one type of data, but rather comprehensively calculates the degree of difference between cross-sensor data and standard database, as well as the fitting degree of image data, according to preset weights. The weight settings are based on the reliability verification results of the two types of data in different environments, ensuring that the confidence level can objectively reflect the positioning status. The final confidence level value ranges from 0 to 1, with a higher value indicating a more reliable positioning.

[0033] The preset positioning threshold in the module is determined based on the accuracy requirements of subway track inspection, error statistics of historical positioning data, and environmental characteristics of different inspection scenarios. The threshold in different scenarios can be flexibly adjusted according to actual needs to balance positioning accuracy and system response sensitivity. If the calculated positioning confidence level is higher than the threshold, it indicates that the current positioning of the inspection robot is reliable, and the navigation execution unit will control the inspection robot to continue moving along the currently planned navigation path; if the positioning confidence level is lower than the threshold, it indicates that the inspection robot has a positioning deviation, and the module will initiate the deviation compensation value calculation process.

[0034] The calculation of the deviation compensation value calls up multiple sets of image data and distance data recently collected by the assisted positioning robot, and filters out valid data samples through an outlier elimination algorithm. Based on these valid samples, the displacement deviation and angle deviation between the inspection robot's own positioning and the assisted positioning results are calculated. Combined with the real-time straight-line distance data collected by the assisted positioning robot, the displacement deviation is further corrected, and finally the deviation compensation value that can accurately reflect the positioning error of the inspection robot is obtained.

[0035] After obtaining the deviation compensation value, the module will trigger the navigation system correction process: the positioning correction unit of the inspection robot will decompose the deviation compensation value into displacement correction and angle correction. The displacement correction is used to adjust the current positioning coordinates recorded in the navigation system to ensure that the coordinates are consistent with the actual position of the inspection robot; the angle correction is used to adjust the scanning reference angle of the 3D LiDAR to make the scanning range of the LiDAR more accurately cover the track area, and at the same time adjust the acquisition field angle of the cross photoelectric sensor to ensure that it can accurately capture the track's landmark features and avoid data deviation caused by sensor angle offset.

[0036] After calibration, the navigation system will call the updated positioning coordinates and adjusted sensor parameters to regenerate the navigation path from the current position to the target task point. This path will be optimized according to the calibrated position and the actual trajectory to ensure the accuracy of the path. The drive module will control the inspection robot to move according to the new path parameters. During the movement, the system will continuously collect the calibrated sensor data and positioning data and feed them back to the positioning confidence calculation stage to monitor the positioning status in real time until the positioning confidence is stably maintained within the preset acceptable range, ensuring that the positioning accuracy always meets the operation requirements during subsequent inspections.

[0037] Specifically, such as Figures 1 to 5As shown, the navigation task initiation module includes: the platform preset positioning mark is an coded label containing mileage information; the auxiliary positioning robot obtains its own coordinates in the platform coordinate system by recognizing the coded label, and at the same time establishes a mapping relationship between the coded label and the preset reflective mark on the track side to complete the construction of a collaborative positioning reference between the platform and the track; the preset reflective mark on the track side is set at intervals along the outer side of the track, and the coded label is set along the edge of the platform corresponding to the position of the reflective mark; the auxiliary positioning robot takes pictures of the reflective mark through the image acquisition module, and combines the distance measurement module to measure the fixed distance between the coded label and the reflective mark to complete the calibration of the collaborative positioning reference.

[0038] The platform's pre-set positioning markers are in the form of coded labels. The process of the positioning robot recognizing these coded labels involves multiple steps of image processing and data analysis. First, the robot's image acquisition module continuously captures images of the platform's edge areas. During this process, the exposure parameters are automatically adjusted based on ambient light intensity. In dimly lit areas at the platform's ends, the module increases the exposure time and enhances image gain to ensure clear label images. In the brightly lit central areas, the exposure intensity is reduced to avoid overexposure caused by reflections from the label surface. After acquiring the label images, the image preprocessing unit first performs noise filtering, using a Gaussian filtering algorithm to eliminate dust particles and light interference. Then, an edge detection algorithm extracts the label's outline boundaries, separating the label area from the complex platform background and preventing background elements from interfering with the coding recognition.

[0039] The encoding and parsing unit performs grayscale processing on the separated label images, converting the color image to grayscale to simplify data calculations. Then, a binarization algorithm converts the grayscale image to a black-and-white binary image, making the black-and-white contrast of the encoded graphic units more vivid. Based on the binary image, the parsing unit identifies the binary information of the graphic units row by row and column by column according to preset decoding rules, converting the graphic signal into a digital signal, and then extracting the mileage data contained in the label. Simultaneously, the odometer of the auxiliary positioning robot records its cumulative displacement distance during movement and compares this displacement distance with the mileage data extracted from the encoded label. If the deviation is within the range set based on the odometer's accuracy level, the robot's coordinates in the platform coordinate system are directly determined using the mileage data from the encoded label. If the deviation exceeds the range, a weighted fusion algorithm combines the two types of data, using the encoded label data as the primary data and the odometer data as a secondary data source, to correct its coordinates and ensure the accuracy of the positioning result.

[0040] When establishing a mapping relationship, the assisted positioning robot first moves to the location of a specific coded tag and uses its image acquisition module to photograph the trackside reflective marker corresponding to that tag. During the photographing process, the robot adjusts its posture so that the optical axis of the image acquisition module's lens is approximately perpendicular to the center of the reflective marker, reducing image distortion caused by excessive shooting angles. Simultaneously, the robot activates its supplementary lighting device, emitting near-infrared light towards the reflective marker, making it appear as a bright spot in the image, facilitating recognition by the image processing unit. The image processing unit extracts the spot from the image containing the reflective marker, determines its pixel coordinates, and then, combining the intrinsic parameters of the image acquisition module and the robot's own coordinates, calculates the reflective marker's coordinates in the platform coordinate system using a perspective projection transformation algorithm.

[0041] The backend system retrieves the track coordinates corresponding to the reflective mark from the database and associates them with the calculated platform coordinates, establishing a one-to-one correspondence between the reflective mark and its corresponding coded label. This means that the platform coordinates of the coded label can be directly linked to the track coordinates of the reflective mark, and vice versa. To ensure the stability of the mapping relationship, the auxiliary positioning robot takes multiple photos and performs coordinate calculations on the same set of coded labels and reflective marks. After removing outlier data, the average value is taken as the final mapped coordinates, reducing the impact of single measurement errors on the mapping relationship. By associating all coded labels with their corresponding reflective marks one by one, a platform and track collaborative positioning benchmark covering the entire inspection section is ultimately formed, providing a unified spatial reference framework for subsequent positioning data conversion and verification by the two robots.

[0042] Specifically, such as Figures 1 to 5 As shown, the real-time data acquisition module includes a system in which the auxiliary positioning robot dynamically adjusts its own speed and direction based on the real-time positioning data of the inspection robot, ensuring that the inspection robot is always within the effective field of view of the auxiliary positioning robot's image acquisition module, and that the distance between the auxiliary positioning robot and the inspection robot is maintained within a preset following range.

[0043] Specifically, such as Figures 1 to 5 As shown, the track marking features include track fasteners, sleepers, and track mileage markers. The cross photoelectric sensor collects image data of the marking features at a preset frequency and records the timestamp of the collection time. When the auxiliary positioning robot collects data, it records the same timestamp and completes the time synchronization between the inspection robot and the auxiliary positioning robot.

[0044] Specifically, such as Figures 1 to 5As shown, the positioning deviation calculation module includes: calculating the ratio of the deviation value between the data collected by the cross photoelectric sensor and the standard feature database of the track to a preset standard deviation threshold to obtain the photoelectric data difference rate; calculating the image fitting degree by the degree of matching between the feature image of the inspection robot collected by the auxiliary positioning robot and the image features converted from the positioning data of the inspection robot itself; and calculating the positioning confidence level based on the photoelectric data difference rate and the image fitting degree.

[0045] Specifically, such as Figures 1 to 5 As shown, the calculation of the image fitting degree includes converting the pixel coordinates of the inspection robot's appearance features in the image collected by the assisted positioning robot into actual coordinates in the platform coordinate system, combined with the positioning data of the assisted positioning robot; converting the inspection robot's own positioning data into coordinates in the platform coordinate system; calculating the Euclidean distance between the two sets of coordinates; and determining the image fitting degree based on the ratio of the Euclidean distance to a preset distance threshold.

[0046] Specifically, such as Figures 1 to 5 As shown, the deviation compensation value calculation includes: the auxiliary positioning robot continuously collecting multiple frames of image data and distance data, removing abnormal data to obtain average image feature coordinates and average distance values; calculating the displacement deviation and angle deviation of the inspection robot's self-positioning and auxiliary verification positioning based on the average image feature coordinates; and correcting the displacement deviation by combining the average distance value to obtain the final deviation compensation value.

[0047] The raw data collected by the cross-sensor contains a large amount of environmental interference signals. Therefore, it needs to be preprocessed before comparison with the standard database. The preprocessing stage is completed collaboratively by the signal purification unit and the feature reconstruction unit: the signal purification unit adopts an adaptive median filtering algorithm, which dynamically adjusts the size of the filtering window according to the distribution density of noise points in the image. This can eliminate isolated dust noise points while preserving the edge details of the track's iconic features, avoiding over-filtering that could lead to feature blurring; the feature reconstruction unit, on the other hand, targets features obscured by local stains. It uses a neighborhood feature interpolation algorithm to complete the feature data of the obscured area based on the feature parameters of the unobscured area, ensuring the integrity of the data to be compared and avoiding misjudgments caused by local obscuration.

[0048] The track standard feature database is not a static collection of images, but a dynamic feature model library built based on track component industry standards, actual installation parameters, and full lifecycle operation and maintenance data. In this database, each track's distinctive feature contains multi-dimensional standard parameters: taking track fasteners as an example, parameters cover the length, width, and height of the fastener base, the bending curvature of elastic components, the bolt center-to-center distance and bolt head diameter, and even the normal wear parameter range for different service years. The database stores these parameters in segments according to subway line mileage, with each mileage segment corresponding to a unique set of feature parameters, ensuring that inspection robots can access standard data for comparison at the corresponding location.

[0049] During the multi-dimensional deviation extraction process, the calculation module compares the preprocessed collected data with the standard parameters corresponding to the mileage in the database one by one, extracting the deviation value for each dimension. Taking railway sleepers as an example, it calculates the deviations between the actual cross-sectional width and standard width, the actual spacing and standard spacing, and the actual height and standard height of the sleepers in the collected data. For railway fasteners, it calculates the deviations between the actual center spacing of the bolts and the standard spacing, and the deviations between the actual length of the fastener base and the standard length. Subsequently, a weighted average algorithm is used to fuse the multi-dimensional deviation values. Based on the weight of the influence of different dimension parameters on positioning accuracy, such as the fastener bolt spacing having a higher reference value for positioning than the fastener base height, a higher weight is assigned to it, and a comprehensive deviation value is calculated. This value can comprehensively reflect the overall deviation degree between the collected data and the standard state.

[0050] The preset standard deviation threshold is a dynamic threshold determined based on the accuracy requirements of subway track inspection and historical data statistics. The threshold setting varies depending on the distinctive characteristics of different tracks: for example, track mileage markers are directly associated with mileage coordinates and have the greatest impact on positioning accuracy, so their threshold setting is more stringent; sleeper spacing has a relatively weak auxiliary reference effect on positioning, so the threshold can be appropriately relaxed. At the same time, the threshold is dynamically adjusted in conjunction with the service life of the track. Newly laid tracks have high stability of characteristic parameters, so the threshold setting is smaller; for tracks with a longer service life, due to natural wear and tear, the characteristic parameters fluctuate more widely, so the threshold is appropriately increased to avoid misjudging positioning deviations due to normal wear and tear.

[0051] The formula for calculating the photoelectric data difference rate is: Difference Rate = Overall Deviation Value / Preset Standard Deviation Threshold. The difference rate ranges from 0 to 1. When the overall deviation value is 0, the difference rate is 0, indicating that the collected data is completely consistent with the standard data. When the overall deviation value equals the preset standard deviation threshold, the difference rate is 1, indicating that the collected data has reached the upper limit of the allowable deviation. If the overall deviation value exceeds the threshold, the difference rate will be greater than 1. In this case, it is directly determined that the current photoelectric data cannot support reliable positioning, and the deviation correction process must be triggered first.

[0052] Image fitting degree is a core indicator for measuring the spatial consistency between the images acquired by the assisted positioning robot and the positioning data of the inspection robot itself, ensuring that the results can accurately reflect the synergy of the two robots' positioning.

[0053] The feature images of the inspection robot collected by the assisted positioning robot need to undergo feature enhancement and target segmentation processing before effective matching features can be extracted. The feature enhancement stage uses the Retinex image enhancement algorithm, which separates the illumination and reflection components of the image to eliminate the impact of uneven lighting within the tunnel on image quality, significantly improving the contrast of the inspection robot's appearance features. The target segmentation stage uses a semantic segmentation algorithm to separate the inspection robot region in the image from the track and tunnel background, avoiding interference from background elements in feature extraction. After segmentation, the feature extraction unit uses the ORB feature detection algorithm to extract key points of the inspection robot's appearance features, such as the vertices of geometric markers and edge turning points, and calculates a descriptor for each key point. The descriptor contains information such as the grayscale distribution of the key point and the gradient direction of neighboring pixels, which can be used for subsequent feature matching.

[0054] The conversion of the inspection robot's self-positioning data relies on the previously established platform and track collaborative positioning benchmark. The robot's self-positioning data is generated based on the track coordinate system, including longitudinal coordinates along the track mileage direction, lateral coordinates perpendicular to the track direction, and its own attitude angles. During the conversion process, the calculation module first calls the mapping relationship between track coordinates and platform coordinates in the collaborative positioning benchmark to convert the inspection robot's track coordinates into theoretical coordinates in the platform coordinate system. Simultaneously, combined with the inspection robot's attitude angles, the theoretical positions of its key external features in the platform coordinate system are calculated. For example, if the inspection robot has an attitude deviation, the actual position of its geometric markings will differ from when there is no deviation. The theoretical coordinates need to be corrected using attitude angles to ensure that the theoretical position matches the inspection robot's actual attitude.

[0055] The feature matching quantification process is not simply about determining whether features overlap, but rather about achieving quantitative evaluation through keypoint matching, coordinate deviation calculation, and goodness-of-fit conversion. First, the feature keypoint descriptors extracted by the assisted positioning robot are matched with the theoretical keypoint descriptors derived from the inspection robot's own positioning data. By calculating the Hamming distance between the descriptors, keypoint pairs with high matching degrees are selected. Then, the RANSAC algorithm is used to eliminate mismatched keypoint pairs, ensuring the reliability of the matching results.

[0056] Subsequently, the Euclidean distance of each pair of matched keypoints in the platform coordinate system is calculated. The Euclidean distance reflects the spatial deviation between the actual acquired keypoint positions and the theoretical keypoint positions. To avoid misjudgments of fit due to deviations in a single keypoint, the average Euclidean distance of all valid matched keypoint pairs is calculated. This average distance comprehensively reflects the overall deviation of the dual-robot positioning data. The preset distance threshold is set based on the accuracy of the auxiliary positioning robot's image acquisition module and the positioning verification requirements.

[0057] The formula for calculating the image fit is: Fit = 1 - (Average Euclidean Distance / Preset Distance Threshold), where the fit ranges from 0 to 1. When the average Euclidean distance is 0, the fit is 1, indicating that the feature positions collected by the assisted positioning robot are completely consistent with the positioning of the inspection robot itself. When the average Euclidean distance equals the preset distance threshold, the fit is 0, indicating that the positioning deviation has reached the allowable upper limit. If the average Euclidean distance exceeds the threshold, the fit will be less than 0, at which point it is directly determined that there is a significant deviation in the dual-robot positioning data, and a correction process needs to be initiated.

[0058] The location confidence score is not simply the sum of the photoelectric data difference rate and the image fitting degree, but a weighted fusion based on the reliability weights of the two. Its calculation process needs to combine environmental adaptability analysis and dynamic weight adjustment to ensure that the confidence score can objectively reflect the location status under different environments.

[0059] After calculating the positioning confidence level, the module compares it with a preset confidence threshold. This preset threshold is also set based on the accuracy requirements of the inspection scenario: in inspection scenarios requiring high-precision positioning, such as track fasteners and rail joints, the threshold is set higher to ensure that inspection continues only when the positioning is extremely reliable; in scenarios with lower accuracy requirements, such as overall track contour detection, the threshold can be appropriately lowered to balance positioning accuracy and inspection efficiency. If the confidence level is higher than the threshold, it indicates that the current positioning status meets the inspection requirements, and the calculation module sends a continue navigation command to the inspection robot's navigation execution unit; if the confidence level is lower than the threshold, the calculation module immediately initiates the deviation compensation value calculation process and sends a pause navigation command to the navigation execution unit to prevent inspection data from becoming invalid or the robot from deviating from its track due to positioning deviation.

[0060] In addition, the calculation module performs continuous trend analysis on the positioning confidence level. If the confidence level is higher than the threshold but shows a continuous downward trend, it indicates that the positioning status of the inspection robot is gradually deteriorating, and there may be potential deviation risks. At this time, the module will send a positioning warning signal to the backend system in advance. The backend system can proactively adjust the inspection robot's data collection frequency or assist the positioning robot's following strategy to prevent the positioning confidence level from falling further below the threshold, thereby achieving early control of positioning risks and improving the stability of system operation.

[0061] Specifically, such as Figures 1 to 5As shown, it also includes a collaborative obstacle avoidance module. The collaborative obstacle avoidance module is communicatively connected to the inspection robot and the auxiliary positioning robot. It acquires images of the subway platform and determines the distance between the auxiliary positioning robot and the obstacle. If the distance between the auxiliary positioning robot and the obstacle is less than a safety threshold, the auxiliary positioning robot stops moving and feeds back its position information. The inspection robot stops moving synchronously.

[0062] Specifically, such as Figures 1 to 5 As shown, the navigation system calibration includes the following steps: After receiving the deviation compensation value, the inspection robot decomposes the compensation value into displacement correction and angle correction through its own positioning module; based on the displacement correction, it updates the current positioning coordinates in the navigation system, and adjusts the scanning reference angle of the three-dimensional lidar and the acquisition field of view angle of the cross photoelectric sensor in combination with the angle correction; the navigation system calls the calibrated positioning coordinates and sensor parameters to regenerate the navigation path to the target task point, and controls the drive module to move along the new path. During the movement, the calibrated data is collected in real time and fed back to the positioning confidence calculation stage to continuously verify the navigation system calibration effect until the movement state is stable and the positioning confidence is maintained within the preset acceptable range.

[0063] The foregoing has illustrated and described the basic features, principles, and advantages of the present invention. It should be noted that the present invention is not limited to the above embodiments, but only to some embodiments. Any improvements and additions made without departing from the spirit and scope of the present invention are considered to be within the scope of protection of the present invention.

Claims

1. A robot positioning and navigation system based on three-dimensional lidar and cross-photoelectric correction, characterized in that, include: The navigation task initiation module allows the inspection robot to receive the inspection task, plan a navigation path based on the preset track path and three-dimensional lidar scanning data, and send a follow command to the auxiliary positioning robot. The auxiliary positioning robot completes its own positioning and establishes the platform coordinate system through the preset positioning marks on the platform. The real-time data acquisition module allows the inspection robot to collect point cloud data of the track environment via a 3D lidar and track landmark feature data via cross photoelectric sensors. The auxiliary positioning robot moves synchronously following the inspection robot, collecting the appearance features of the inspection robot and the surrounding track features, as well as the real-time straight-line distance data between the robot and the inspection robot. The positioning deviation calculation module calculates the positioning confidence level of the inspection robot based on the difference between the cross photoelectric sensor data uploaded by the inspection robot and the preset track standard feature database, and the fitting degree between the image data uploaded by the auxiliary positioning robot and the positioning data of the inspection robot itself. If it is higher than the preset threshold, the inspection robot continues to move along the current navigation path; otherwise, it calculates the deviation compensation value and performs navigation system correction.

2. The robot positioning and navigation system based on three-dimensional lidar and cross-photoelectric correction according to claim 1, characterized in that, The navigation task initiation module includes: the platform preset positioning mark is an coded label containing mileage information; the auxiliary positioning robot obtains its own coordinates in the platform coordinate system by recognizing the coded label, and at the same time establishes a mapping relationship between the coded label and the preset reflective mark on the track side to complete the construction of a collaborative positioning benchmark between the platform and the track; the preset reflective mark on the track side is set at intervals along the outer side of the track, and the coded label is set along the edge of the platform corresponding to the position of the reflective mark; the auxiliary positioning robot takes pictures of the reflective mark through the image acquisition module, and combines the distance measurement module to measure the fixed distance between the coded label and the reflective mark to complete the calibration of the collaborative positioning benchmark.

3. The robot positioning and navigation system based on three-dimensional lidar and cross-photoelectric correction according to claim 1, characterized in that, The real-time data acquisition module includes a system in which the auxiliary positioning robot dynamically adjusts its own speed and direction based on the real-time positioning data of the inspection robot, ensuring that the inspection robot is always within the effective field of view of the auxiliary positioning robot's image acquisition module, and that the distance between the auxiliary positioning robot and the inspection robot is maintained within a preset following range.

4. The robot positioning and navigation system based on three-dimensional lidar and cross-photoelectric correction according to claim 1, characterized in that, The track marker features include track fasteners, sleepers, and track mileage markers. The cross photoelectric sensor collects image data of the marker features at a preset frequency and records the timestamp of the collection time. When the auxiliary positioning robot collects data, it records the same timestamp and completes the time synchronization between the inspection robot and the auxiliary positioning robot.

5. The robot positioning and navigation system based on three-dimensional lidar and cross-photoelectric correction according to claim 1, characterized in that, The positioning deviation calculation module includes: calculating the ratio of the deviation value between the data collected by the cross photoelectric sensor and the standard feature database of the track to a preset standard deviation threshold to obtain the photoelectric data difference rate; calculating the image fitting degree by the degree of matching between the feature image of the inspection robot collected by the auxiliary positioning robot and the image features converted from the positioning data of the inspection robot itself; and calculating the positioning confidence level based on the photoelectric data difference rate and the image fitting degree.

6. The robot positioning and navigation system based on three-dimensional lidar and cross-photoelectric correction according to claim 5, characterized in that, The calculation of the image fitting degree includes converting the pixel coordinates of the appearance features of the inspection robot in the image acquired by the assisted positioning robot into actual coordinates in the platform coordinate system, combined with the positioning data of the assisted positioning robot; converting the positioning data of the inspection robot itself into coordinates in the platform coordinate system; calculating the Euclidean distance between the two sets of coordinates; and determining the image fitting degree based on the ratio of the Euclidean distance to a preset distance threshold.

7. The robot positioning and navigation system based on three-dimensional lidar and cross-photoelectric correction according to claim 1, characterized in that, The deviation compensation value calculation includes: the auxiliary positioning robot continuously collecting multiple frames of image data and distance data, removing abnormal data to obtain average image feature coordinates and average distance values; calculating the displacement deviation and angle deviation of the inspection robot's self-positioning and auxiliary verification positioning based on the average image feature coordinates, and correcting the displacement deviation in combination with the average distance value to obtain the final deviation compensation value.

8. The robot positioning and navigation system based on three-dimensional lidar and cross-photoelectric correction according to claim 1, characterized in that, It also includes a collaborative obstacle avoidance module, which is communicatively connected to the inspection robot and the auxiliary positioning robot. It acquires images of the subway platform and determines the distance between the auxiliary positioning robot and the obstacle. If the distance between the auxiliary positioning robot and the obstacle is less than a safety threshold, the auxiliary positioning robot stops moving and reports its position information, and the inspection robot stops moving synchronously.

9. The robot positioning and navigation system based on three-dimensional lidar and cross-photoelectric correction according to claim 1, characterized in that, The navigation system calibration includes the following steps: after receiving the deviation compensation value, the inspection robot decomposes the compensation value into displacement correction and angle correction through its own positioning module; updates the current positioning coordinates in the navigation system based on the displacement correction; and adjusts the scanning reference angle of the three-dimensional lidar and the acquisition field of view angle of the cross photoelectric sensor in combination with the angle correction. The navigation system calls the corrected positioning coordinates and sensor parameters to regenerate the navigation path to the target task point, and controls the drive module to move along the new path. During the movement, the corrected data is collected in real time and fed back to the positioning confidence calculation stage to continuously verify the correction effect of the navigation system until the movement is stable and the positioning confidence is maintained within the preset acceptable range.

Images