Radar and photoelectricity-based inspection map construction system

Abstract

The invention discloses an inspection map construction system based on radar and photoelectricity. The system comprises a three-dimensional laser radar module, a photoelectric positioning module, a data processing module and a map reconstruction module. The three-dimensional laser radar module collects environment data, constructs a global three-dimensional point cloud map and outputs the initial pose of the robot. The photoelectric positioning module corrects parking deviation through photoelectric sensors which are arranged in a crossed manner, and obtains an axle image through an up-looking camera; after the data processing module preprocesses the map, multi-source data are fused, the optimal pose is output through Kalman filtering, and secondary pose correction is completed by combining the comparison result of the axle image and the standard template; and the map reconstruction module realizes dynamic optimization and reconstruction of an inspection map, through multi-module cooperation and hierarchical correction, the positioning precision and scene adaptability are improved, the problems of positioning drift, weak map dynamic updating capability and the like in the traditional technology are effectively solved, and the high-precision requirement of rail transit underbody inspection is met.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent inspection of rail transit, in particular to a kind of inspection map construction system based on radar and photoelectricity. BACKGROUND

[0002] With the rapid development of rail transit industry, train operation mileage and vehicle ownership continue to grow, and train maintenance demand is increasingly urgent. Traditional train inspection relies on manual completion, which has problems such as low efficiency, high labor intensity, and detection results easily affected by personnel experience and state, which is difficult to meet the safety protection demand under high-density operation. Therefore, intelligent inspection robots are gradually applied in the field of rail transit, and the positioning accuracy and map adaptation ability of the robot are put forward with high requirements due to the characteristics of closed environment, dense equipment and narrow space in train inspection.

[0003] The existing map construction technology of inspection robot mainly relies on single sensor or general multi-sensor fusion scheme, which has the following disadvantages: low positioning progress, traditional two-dimensional laser radar map construction has high dependence on fixed reference in environment, and there are temporary obstacles and dynamic changes of equipment in rail transit section, which easily leads to positioning drift and cannot meet the centimeter-level positioning demand of train inspection; poor scene adaptability, general robot map construction scheme does not optimize the position relationship of train and robot, ignores individual characteristics such as train carriage length difference and train head position fine adjustment, which leads to deviation of robot and detection point, and affects the accuracy of fault detection; low multi-sensor data fusion efficiency, although some schemes combine laser and visual sensors, there is lack of cooperative strategy for rail transit scene, and problems such as laser point cloud noise and visual image reflection easily lead to accumulation of fusion error and affect the reliability of map.

[0004] Therefore, in order to solve the problems existing in the prior art, the present application provides an inspection map construction system based on radar and photoelectricity. SUMMARY

[0005] In view of the deficiencies in the prior art, the purpose of the present application is to provide an inspection map construction system based on radar and photoelectricity.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical scheme: An inspection map construction system based on radar and photoelectricity, comprising: A three-dimensional laser radar module for collecting three-dimensional point cloud data, inertial measurement data and mileage data, constructing a global three-dimensional point cloud map according to the collected data and outputting an initial pose of the robot; An optical positioning module comprising a cross-arranged photoelectric sensor and an upward-looking camera, the photoelectric sensor being used to detect the position of the train head and correct the parking deviation of the robot, and the upward-looking camera being used to obtain train wheel shaft images; The data processing module is in communication connection with the photoelectric positioning module and the three-dimensional laser radar module, pre-processes the three-dimensional point cloud map, combines the inertial measurement data, the mileage data and the initial pose to obtain the optimal pose of the robot through Kalman filtering optimization, and outputs the real-time positioning result; calculates the position deviation according to the matching result of the wheel shaft image and the standard wheel shaft template, and performs secondary correction on the pose of the robot relative to the train according to the position deviation to obtain the pose correction result; The map reconstruction module reconstructs the inspection map according to the three-dimensional point cloud map and the pose correction result.

[0007] As a further improvement of the application, the three-dimensional laser radar module comprises a map acquisition and construction sub-module, which is used for acquiring three-dimensional laser radar data, inertial measurement unit data and odometer data through robot closed-loop path scanning, generating a three-dimensional point cloud map composed of multiple frames of movement data, performing rasterization processing and denoising point processing on the three-dimensional point cloud map, and setting an initial positioning value of the robot relative to the three-dimensional point cloud map coordinate system.

[0008] As a further improvement of the application, the data processing module comprises a data acquisition and matching sub-module, which is used for obtaining an initial pose of the robot according to the initial positioning value through the inertial measurement data and the odometer data analysis, matching a current frame of laser radar with a pre-processed three-dimensional point cloud map frame to output a six-degree-of-freedom pose of the robot, the six-degree-of-freedom pose comprising an X-axis position, a Y-axis position, a Z-axis position, a heading angle, a pitch angle and a roll angle, and performing data fusion on the six-degree-of-freedom pose as an observation value and the inertial measurement unit data and the odometer data to output an optimal pose of the robot to the inspection target position.

[0009] As a further improvement of the application, the secondary correction of the pose in the data processing module comprises: capturing a wheel shaft image of each car of the train through the upward-looking camera, extracting a wheel shaft feature to analyze the actual position of the wheel shaft, comparing the actual position of the wheel shaft with a preset standard wheel shaft template position to calculate a wheel shaft position deviation, the wheel shaft position deviation being a pose deviation of the robot relative to the current car, adjusting the pose of the robot relative to the train according to the wheel shaft position deviation, and outputting a secondary corrected pose correction result, the pose correction result comprising a six-degree-of-freedom correction value of the robot relative to the current car, a relative coordinate relationship between the robot and a detection point of the train and a pose calibration mark for map reconstruction, the pose calibration mark being used for associating the coordinate mapping relationship between the global three-dimensional point cloud map and the local features of the train, and being output to the map reconstruction module.

[0010] As a further improvement of the application, the map reconstruction module comprises a coordinate mapping submodule for establishing a dynamic mapping relationship between the global three-dimensional point cloud map coordinate system and the local feature coordinate system of the train based on the pose calibration marker; a pose correction submodule for inversely superimposing the six-degree-of-freedom correction value to the historical positioning trajectory of the robot in the global three-dimensional point cloud map to correct the cumulative error of the robot pose in the global map; and a feature updating submodule for updating the coordinate information of the train detection points in the global three-dimensional point cloud map according to the relative coordinate relationship between the robot and the train detection points in combination with the dynamic mapping relationship, and completing dynamic optimization and reconstruction of the inspection map.

[0011] As a further improvement of the application, the photoelectric positioning module corrects the parking deviation by, during the driving of the robot to the preset theoretical parking point based on the optimal pose, real-time scanning of the train head area by the cross-arranged photoelectric sensors, identification of the train head position by the preset head edge feature model, prediction of the parking deviation generated when reaching the theoretical parking point based on the current driving speed, the distance to the theoretical parking point, and historical driving error data, adjustment of the driving trajectory of the robot according to the predicted parking deviation, and completion of the correction of the parking deviation by the photoelectric sensors after reaching the theoretical parking point to detect the deviation between the actual train head position and the theoretical position, and compensation adjustment in combination with the predicted parking deviation.

[0012] As a further improvement of the application, the map reconstruction module further comprises calling the pose calibration marker to analyze the marker point coordinates in the global three-dimensional point cloud map coordinate system and the marker point coordinates in the local feature coordinate system of the train, solving the conversion matrix of the two coordinate systems by the least square method to establish the dynamic mapping relationship, and updating the dynamic mapping relationship in real time as the train position changes; extracting the historical positioning trajectory data of the robot in the global three-dimensional point cloud map, inversely superimposing the six-degree-of-freedom correction value to the historical pose at the corresponding time according to the time sequence, and calculating the correction amount of each historical pose, wherein the X-axis, Y-axis, and Z-axis correction amounts are distributed to the historical trajectory segment by linear interpolation, and the heading angle, pitch angle, and roll angle correction amounts compensate for the rotation error by spherical linear interpolation; converting the local coordinates of the detection points into absolute coordinates in the global three-dimensional point cloud map coordinate system according to the relative coordinate relationship between the robot and the train detection points in combination with the conversion matrix, comparing the converted coordinates with the originally stored detection point coordinates in the map, updating the map data with the converted coordinates if the deviation exceeds the preset threshold, and recording the coordinate update time stamp and the corresponding pose calibration marker version; and integrating the corrected historical trajectory data and the updated detection point coordinate information to generate an inspection map including the global environmental features, the robot driving trajectory, and the dynamic features of the train.

[0013] As a further improvement of the application, the map reconstruction module further comprises, after establishing the dynamic mapping relationship, optimizing the conversion matrix through the random sample consensus algorithm, eliminating abnormal mapping data caused by marker point recognition errors, and retaining conversion matrix parameters with a confidence level higher than a preset threshold; after generating the inspection map, calculating the matching degree of the corrected historical trajectory and the static features in the global three-dimensional point cloud map, and if the matching degree is lower than a preset threshold, re-establishing the coordinate mapping relationship and the subsequent steps until the matching degree is higher than the preset threshold, the matching degree being calculated by the mean value of the Euclidean distance of the point cloud overlap area.

[0014] As a further improvement of the application, the steps of the map construction method of the system include collecting environment data through closed-loop path scanning, constructing an initial global three-dimensional point cloud map and performing gridization, denoising and pretreatment, and setting an initial positioning value of the robot; based on the initial positioning value, fusing inertial measurement data, mileage data and laser radar matching results, outputting an optimal pose of the robot through Kalman filtering, controlling the robot to autonomously navigate to a target lane, predicting parking deviation during the robot driving process through the photoelectric positioning module and adjusting the trajectory in advance, detecting the vehicle head position through the cross photoelectric sensor after reaching the target lane, and completing correction of the parking deviation; shooting a train wheel axle image, calculating the pose deviation through comparison between the wheel axle image and a standard template, and outputting a pose correction result containing a six-degree-of-freedom correction value, a relative coordinate relationship and a pose calibration mark; based on the pose correction result, establishing coordinate mapping, correcting global pose error and updating detection point coordinates, and completing dynamic reconstruction and storage of the inspection map.

[0015] The application has the following advantages: (1) The positioning accuracy and stability are improved. Through the cooperation of the three-dimensional laser radar module and the Kalman filtering algorithm, the inertial measurement data and the mileage data are fused to realize the output of the optimal pose of the robot, and the positioning error can be controlled within centimeters. At the same time, the cross photoelectric sensor of the photoelectric positioning module corrects the parking deviation of the robot through a correction mechanism, avoiding the false triggering problem of single-point detection. On this basis, the wheel axle secondary correction further eliminates the local deviation caused by the length difference of the carriages through the six-degree-of-freedom correction value, finally realizes the millimeter-level alignment of the robot and the detection points of the train, effectively prevents the mechanical arm from colliding with the undercarriage equipment, and ensures the safety of the inspection.

[0016] (2) Enhance the adaptability of complex scenes, three-dimensional laser radar module through closed-loop path scanning and point cloud preprocessing technology, reduce the dependence on the environment fixed reference, solve the traditional two-dimensional laser radar in the field of temporary obstacles, equipment dynamic change scene positioning drift problem. At the same time, the global coordinate system and the dynamic mapping relationship between the local coordinate system of the train established by the pose calibration mark and the coordinate mapping sub module can be updated in real time with the change of the train parking position, adapt to the individual characteristics of different marshalling trains, and break through the strong dependence of the general map construction scheme on the static environment.

[0017] (3) Improve the anti-interference ability and improve the reliability of map construction. Through the hierarchical fusion of multi-sensor data by the data processing module, noise and environmental interference are effectively filtered out, the measurement deviation problem of single sensor in complex car bottom environment is solved, and the robustness of map construction is improved. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 is a system block diagram of a kind of inspection map construction system based on radar and photoelectricity of the present application; Figure 2 is the schematic diagram of map reconstruction module of the system of the present application; Figure 3 is the flow chart of deviation correction of the photoelectric positioning module of the system of the present application; Figure 4 is the running flow chart of map reconstruction module of the system of the present application; Figure 5 is the inspection map method construction flow chart of a kind of inspection map construction system based on radar and photoelectricity of the present application. DETAILED DESCRIPTION

[0019] The present application is further described in detail below in conjunction with the drawings and examples. Wherein the same parts are indicated by the same reference signs. It should be noted that the words "front", "back", "left", "right", "up" and "down" used in the following description refer to the directions in the drawings, and the words "bottom surface" and "top surface", "inner" and "outer" refer to the direction towards or away from the geometric center of a particular component.

[0020] The present application proposes a kind of inspection map construction system based on radar and photoelectricity, as shown in Figure 1 It includes: three-dimensional laser radar module, photoelectric positioning module, data processing module, map reconstruction module.

[0021] Three-dimensional laser radar module is used to collect environment three-dimensional point cloud data, inertial measurement data and mileage data, constructs global three-dimensional point cloud map according to the acquisition data and exports robot initial pose; The three-dimensional laser radar module is a core unit of the system to obtain spatial information of the environment, and is mainly responsible for collecting three-dimensional point cloud data, inertial measurement data and mileage data of the inspection environment, and constructing a global three-dimensional point cloud map based on the data, and outputting an initial pose of the robot.

[0022] The three-dimensional point cloud data is obtained by scanning the surrounding environment through laser beams emitted by the laser radar, and can accurately reflect the three-dimensional coordinates and morphological characteristics of static objects in the environment. The inertial measurement data is collected through an integrated inertial measurement unit, and is used to capture the acceleration and angular velocity changes in the movement process of the robot, and assist in judging the movement state of the robot. The mileage data is recorded by the odometer to record the distance and direction information of the robot, and provides a basic reference for pose calculation.

[0023] In the process of map construction, the module completes data collection through closed-loop path scanning of the robot, that is, the robot starts from the starting point, travels along the preset path and finally returns to the starting point, ensuring that the collected data forms a complete closed loop. Based on the closed-loop data, the module fuses and splices the point cloud data collected in the movement process to generate a global three-dimensional point cloud map. The initial pose serves as a starting reference for the robot in the map coordinate system, and provides a coordinate reference for subsequent positioning and navigation.

[0024] Specifically, as shown in Figures 1 to 5 The three-dimensional laser radar module includes a map acquisition and construction submodule, which is used to collect three-dimensional laser radar data, inertial measurement unit data and odometer data through closed-loop path scanning of the robot, generate a three-dimensional point cloud map composed of multiple moving data, perform rasterization processing and denoising point processing on the three-dimensional point cloud map, and set an initial positioning value of the robot relative to the three-dimensional point cloud map coordinate system.

[0025] The map acquisition and construction submodule completes environmental data collection through closed-loop path scanning of the robot. Closed-loop path scanning refers to that the robot starts from a preset starting point, travels along a planned inspection path, covers the target lane and the surrounding environment, and then returns to the starting point to form a complete closed trajectory. The core advantage of this path design is to provide an error calibration reference for subsequent data fusion by overlapping the first and last trajectories, and to reduce the cumulative error caused by long-distance travel.

[0026] During the scanning process, the sub-module synchronously collects three types of key data: three-dimensional laser radar data, inertial measurement unit data, and odometry data. The three-dimensional laser radar data generates point cloud data containing spatial coordinates and reflectivity by scanning the environment with high-frequency laser pulses, accurately capturing the shape and location of static features such as tracks, walls, and equipment. The inertial measurement unit data records the acceleration and angular velocity changes during the robot's movement, reflecting the robot's dynamic attitude. The odometry data records the distance and direction of the robot's travel, assisting in tracking the robot's movement trajectory. The three types of data are time-stamped and synchronized to ensure that each frame of point cloud data matches the corresponding motion state parameters, providing a consistent basis for subsequent map stitching.

[0027] Based on the collected multi-frame movement data, the sub-module generates a three-dimensional point cloud map through fusion. Multi-frame data fusion is achieved through coordinate transformation: each frame of point cloud data is converted to the global coordinate system based on the current pose of the robot, and then the feature matching algorithm is used to overlap the adjacent frame data in the overlapping area, eliminating the frame-to-frame deviation. For the repeated structures commonly found in rail transit scenes, such as track spacing and equipment supports, the sub-module identifies stable feature points through a feature extraction algorithm, and uses these feature points as a reference to optimize the frame-to-frame registration accuracy, ensuring that the multi-frame data stitching forms a continuous and complete global point cloud map.

[0028] To improve the usability of the map and the efficiency of subsequent positioning, the sub-module performs rasterization processing and noise removal processing on the generated three-dimensional point cloud map.

[0029] Rasterization processing divides the three-dimensional space into regular grids of fixed size, and each grid stores the point cloud data in that area as an independent spatial unit. This processing method reduces data redundancy and speeds up the point cloud matching efficiency in the subsequent positioning process. At the same time, through the feature statistics of the grid unit, such as point cloud density and average reflectivity, the environmental feature recognizability is enhanced.

[0030] Noise removal processing filters out the noise introduced during the collection process. Noise mainly comes from laser reflection interference, environmental interference, and sensor errors. The sub-module identifies and removes isolated abnormal points in a single frame through multi-frame comparison and analysis; removes false point clouds caused by reflections through a reflectivity threshold filter; and filters out discrete points that do not conform to the physical form through spatial continuity verification. After noise removal, the map only retains stable static feature point clouds in the environment, providing a reliable feature reference for subsequent positioning.

[0031] After the map generation is completed, the map acquisition and construction sub-module needs to set the initial positioning value of the robot relative to the three-dimensional point cloud map coordinate system. The initial positioning value is the reference coordinate and attitude parameter of the robot in the map, which directly determines the starting accuracy of subsequent navigation and positioning.

[0032] In the setting process, the sub-module first takes the starting point of the closed-loop path as the physical reference point, and determines the three-dimensional coordinates of the point in the map coordinate system through the matching of the point cloud data collected by the laser radar at the point with the surrounding environment features such as the right angle of the wall body and the edge of the fixed equipment. At the same time, combined with the attitude data of the inertial measurement unit at the starting point, the initial heading angle, pitch angle and roll angle of the robot are determined to ensure that the initial attitude is consistent with the actual physical state.

[0033] For rail transit scenes, the setting of the initial positioning value will also be calibrated in combination with fixed scene features such as track number and track spacing. For example, by identifying the rail edge features on both sides of the track, the X-axis and Y-axis parameters of the initial coordinates are fine-tuned to ensure that the map coordinate system is consistent with the actual track direction, laying a foundation for the subsequent accurate travel of the robot along the track.

[0034] Through the above process, the three-dimensional point cloud map generated by the map acquisition and construction sub-module can not only fully reflect the spatial features of the inspection environment, but also effectively offset environmental interference and equipment errors through closed-loop path design, multi-source data fusion and fine preprocessing, providing high-precision basic map support for the positioning and navigation of the entire system.

[0035] The photoelectric positioning module includes a photoelectric sensor and an upward-looking camera arranged in a cross manner, the photoelectric sensor is used for detecting the position of the train head and correcting the parking deviation of the robot, and the upward-looking camera is used for acquiring train wheel shaft images. The photoelectric positioning module adopts a multi-sensor cooperative design, including a photoelectric sensor and an upward-looking camera arranged in a cross manner, and is mainly used for realizing accurate alignment of the robot and the train.

[0036] The photoelectric sensor arranged in a cross manner is installed at the front end of the robot in an X-shaped structure layout, and its core function is to detect the position of the train head and correct the parking deviation of the robot. This cross layout can expand the detection range, reduce the misjudgment of a single sensor caused by shielding or reflection, and improve the stability of the train head edge recognition. During the approach of the robot to the train, the sensor scans the train head area in real time, judges the relative position relationship by recognizing the train head edge features, and provides a basis for parking position adjustment.

[0037] The upward-looking camera is installed on the top of the robot, and the lens is directed upward to align with the bottom of the train, and is used for acquiring wheel shaft images of each train carriage. The wheel shaft, as a fixed feature of the train, has a stable corresponding relationship with the train carriage, and the actual position information of the wheel shaft can be extracted by shooting the wheel shaft image, providing a key reference for subsequent pose correction.

[0038] Specifically, as Figures 1 to 5As shown, the photoelectric positioning module corrects the parking deviation, including that during the robot driving to the preset theoretical parking point based on the optimal pose, the cross-arranged photoelectric sensors scan the train head area in real time, the train head position is identified through the preset head edge feature model, and the parking deviation generated when reaching the theoretical parking point is predicted based on the current driving speed, the distance to the theoretical parking point, and historical driving error data, the robot driving track is adjusted according to the predicted parking deviation, and after reaching the theoretical parking point, the deviation of the actual train head position from the theoretical position is detected again through the photoelectric sensor, and the predicted parking deviation is compensated and adjusted to complete the correction of the parking deviation.

[0039] The cross-arranged photoelectric sensors are the core sensing unit of the parking deviation correction, which are distributed at the front end of the robot in an X-shaped structure to form a multi-angle and full-coverage detection area. This layout can reduce the recognition blind area of a single direction sensor caused by head edge shielding, light reflection or environmental stray light, and improve the stability of head position detection. During the robot driving to the preset theoretical parking point based on the optimal pose, the sensor continuously scans the train head area with high-frequency pulse signals, the emitted probe light beam is reflected when encountering the head edge, the sensor receives the reflected signal and converts it into an electrical signal to form the contour sensing data of the head area.

[0040] To accurately identify the head position, the module is built-in with a preset head edge feature model. The model is constructed based on the typical geometric features of the train head, including edge straightness, corner angle, contour gradient change and other key parameters. By comparing the contour data collected by the sensor in real time with the model features, the actual edge position of the train head can be quickly located, and the train head and the surrounding environment such as the equipment beside the track and the boundary of the wall can be distinguished, ensuring the accuracy of the train head position identification.

[0041] Based on the identification of the real-time position of the train head, the module predicts the parking deviation in combination with the current driving state parameters of the robot. The prediction process integrates three types of key data: the current driving speed is obtained through the odometer and the driving motor feedback, reflecting the speed of the robot; the distance to the theoretical parking point is calculated through the laser radar positioning data and the theoretical coordinates, quantifying the remaining driving distance; and the historical driving error data is a statistical analysis of the parking deviation of the robot in the same track and similar working conditions in the past, including typical error patterns at different speed segments and different distances.

[0042] The module fuses the three types of data through an error prediction model to calculate the position deviation that the robot may generate when driving along the current trajectory to the theoretical parking point. For example, when the historical data shows that the robot is prone to forward parking error due to inertia at a certain speed segment, and the current distance to the theoretical parking point is close, the model will predict the corresponding forward deviation. Based on the predicted parking deviation, the module sends a trajectory adjustment instruction to the robot drive system: if there is a lateral deviation, the left and right wheel speed difference is adjusted to achieve lateral displacement compensation; if there is a front and back deviation, the driving speed is adjusted to adjust the longitudinal position, ensuring that the robot has eliminated most of the deviation in advance before reaching the theoretical parking point.

[0043] After the robot reaches the theoretical parking point, the photoelectric positioning module starts a secondary detection process, which scans the vehicle head area again through a cross photoelectric sensor to obtain the deviation data between the actual vehicle head position and the theoretical vehicle head position corresponding to the theoretical parking point. The core purpose of the secondary detection is to correct the difference between the predicted deviation and the actual scene - due to uncontrollable factors such as changes in ground friction and slight bumps during driving, the predicted deviation may not match the actual deviation, and the secondary detection can capture this real-time difference.

[0044] The module fuses the three types of data through an error prediction model to calculate the position deviation that the robot may generate when driving along the current trajectory to the theoretical parking point. For example, when the historical data shows that the robot is prone to forward parking error due to inertia at a certain speed segment, and the current distance to the theoretical parking point is close, the model will predict the corresponding forward deviation. Based on the predicted parking deviation, the module sends a trajectory adjustment instruction to the robot drive system: if there is a lateral deviation, the left and right wheel speed difference is adjusted to achieve lateral displacement compensation; if there is a front and back deviation, the driving speed is adjusted to adjust the longitudinal position, ensuring that the robot has eliminated most of the deviation in advance before reaching the theoretical parking point.

[0045] Through the coherent process of real-time scanning and identification, dynamic prediction and pre-adjustment, and secondary detection compensation, the photoelectric positioning module effectively offsets the cumulative error, environmental interference, and uncontrollable factors during the robot's driving, ensuring that the deviation between the robot's final parking position and the theoretical parking point is controlled within a very small range, providing a stable initial position reference for subsequent wheel image acquisition, secondary pose correction, and other links.

[0046] The data processing module is in communication connection with the photoelectric positioning module and the three-dimensional laser radar module, pre-processes the three-dimensional point cloud map, optimizes the robot's optimal pose through Kalman filtering based on the inertial measurement data, mileage data, and initial pose, and outputs real-time positioning results; calculates the position deviation based on the matching results of the wheel image and the standard wheel template, and performs secondary correction on the robot's pose relative to the train based on the position deviation to obtain a pose correction result; The data processing module is the core computing unit of the system, which maintains real-time communication with the three-dimensional laser radar module and the photoelectric positioning module, is responsible for data fusion, optimization, and correction calculation, and specifically implements the following functions: The data processing module pre-processes the global three-dimensional point cloud map output by the three-dimensional laser radar module, divides the point cloud data by spatial grid through rasterization processing, removes noise points and redundant data at the same time, and improves the clarity and usability of the map. The data processing module combines the pre-processed map data, inertial measurement data, odometer data, and initial pose, and performs data fusion optimization through Kalman filtering algorithm. Kalman filtering can effectively offset the measurement error of a single sensor, such as the cumulative drift of inertial measurement data and the deviation of odometer data due to ground skidding, through prediction and update of sensor data. Finally, the optimal pose and real-time positioning result of the robot are output, ensuring the positioning stability of the robot in the global range. The data processing module compares and analyzes the wheel shaft image obtained based on the upward-looking camera with the preset standard wheel shaft template. The standard wheel shaft template contains the standard position and shape characteristics of the wheel shaft. By extracting the wheel shaft features in the actual wheel shaft image and comparing them with the template, the position deviation of the wheel shaft can be calculated. This deviation directly reflects the pose deviation of the robot relative to the current carriage. Based on this deviation, the module performs secondary correction on the pose of the robot, generates a pose correction result, and ensures the accurate alignment of the robot and the train detection point.

[0047] Specifically, as shown in Figures 1 to 5 The data acquisition and matching sub-module is used to obtain the initial pose of the robot through analysis of the inertial measurement data and odometer data based on the initial positioning value, and to match the current frame of the laser radar with the pre-processed three-dimensional point cloud map frame to output the six-degree-of-freedom pose of the robot, which includes X-axis position, Y-axis position, Z-axis position, heading angle, pitch angle, and roll angle. The six-degree-of-freedom pose is used as an observation value for data fusion with the inertial measurement unit data and odometer data, and the optimal pose of the robot reaching the inspection target position is output.

[0048] The data acquisition and matching sub-module calculates the initial pose of the robot based on the initial positioning value set by the three-dimensional laser radar module, combined with the inertial measurement data and odometer data. The initial positioning value is the starting coordinate and attitude reference of the robot in the map coordinate system, while the inertial measurement data and odometer data are used to refine the dynamic accuracy of the initial pose.

[0049] The inertial measurement data contains acceleration and angular velocity information of the robot motion, which can reflect the attitude change of the robot in the dynamic process of starting and turning; the odometry data records the cumulative distance and direction deflection angle of the robot, which quantifies the displacement trajectory of the robot. The sub-module fuses the two types of data through the kinematic model: taking the initial positioning value as the starting point, the displacement increment of the robot is calculated using the odometry data, and the attitude deviation in the motion process is corrected combined with the inertial measurement data, and finally the initial pose containing the position and attitude parameters is obtained. This calculation method supplements the static reference with dynamic data, effectively reduces the error of the initial positioning value caused by environmental interference, and lays a foundation for subsequent positioning.

[0050] In order to obtain the real-time pose of the robot in the global map, the sub-module matches the current frame of the laser radar with the preprocessed three-dimensional point cloud map frame. The preprocessed three-dimensional point cloud map frame has completed gridding and denoising processing, and retains stable static features in the environment, such as track edges, equipment supports, etc., which are used as a map template to provide a reference for matching.

[0051] The current frame of the laser radar is generated by real-time scanning and contains the point cloud features of the environment within the current field of view of the robot. The sub-module identifies key features such as planes, corners, and other geometric structures from the current frame through a feature extraction algorithm, and then calculates the spatial transformation relationship between the current frame and the map frame through a point cloud registration algorithm. During the registration process, the sub-module preferentially matches high-stability features such as the intersection of walls and floors, and the fixed outline of equipment, and determines the six-degree-of-freedom pose of the robot relative to the map by minimizing the spatial distance error between feature points.

[0052] The six-degree-of-freedom pose completely describes the spatial state of the robot: the X-axis position and the Y-axis position reflect the coordinates of the robot in the horizontal plane, and the Z-axis position reflects the height change; the heading angle determines the driving direction of the robot, and the pitch angle and roll angle reflect the attitude tilt of the robot due to uneven or bumpy ground. Accurate acquisition of these parameters ensures that the robot can clearly determine its spatial position in a complex environment.

[0053] The six-degree-of-freedom pose obtained by matching the laser radar frame is used as an observation value, which is fused with the inertial measurement unit data and the odometry data to output the optimal pose of the robot reaching the inspection target position. This fusion achieves precision improvement by complementing the characteristics of different sensors: the observation value obtained by matching the laser radar frame has high positioning accuracy, but is limited by the scanning frequency and has a time interval in the output; the inertial measurement unit data and the odometry data have high output frequency and can reflect dynamic changes in real time, but long-term accumulation will produce drift error.

[0054] The sub-module realizes data fusion through a filtering algorithm: taking the laser radar observation value as a high-precision benchmark, the recursive results of the inertial measurement unit and the odometer are calibrated. When the laser radar outputs a new observation value, the sub-module compares it with the recursive result, calculates the deviation and corrects the subsequent recursive process; in the interval between two observation value outputs, the continuity of the pose output is maintained through the dynamic supplement of the inertial measurement data and the odometer data. This fusion strategy not only retains the high-precision advantage of the laser radar, but also makes up for the lack of observation interval through high-frequency dynamic data. The final output of the optimal pose has both high static precision and high dynamic response characteristics, ensuring that the robot always maintains accurate alignment with the target lane during driving.

[0055] By establishing a benchmark through initial pose calculation, obtaining real-time coordinates through laser radar frame matching, and optimizing dynamic precision through multi-source data fusion, the data acquisition and matching sub-module constructs a complete positioning chain of benchmark calibration, real-time observation, and dynamic optimization, providing reliable pose support for the robot to autonomously navigate to the inspection target position, effectively adapting to the complexity of the rail transit section environment and the high-precision requirements of the inspection task.

[0056] Specifically, as shown in Figures 1 to 5 The secondary correction of the pose in the data processing module includes: capturing the wheel shaft image of each car of the train by the upward-looking camera, extracting the wheel shaft feature to analyze the actual position of the wheel shaft, comparing the actual position of the wheel shaft with the preset standard wheel shaft template position, calculating the wheel shaft position deviation, the wheel shaft position deviation being the pose deviation of the robot relative to the current car, adjusting the pose of the robot relative to the train according to the wheel shaft position deviation, and outputting the secondary corrected pose correction result, the pose correction result including the six-degree-of-freedom correction value of the robot relative to the current car, the relative coordinate relationship between the robot and the train detection point, and the pose calibration mark for map reconstruction, the pose calibration mark being used to associate the coordinate mapping relationship between the global three-dimensional point cloud map and the local features of the train, and output to the map reconstruction module.

[0057] The starting link of the secondary correction is wheel shaft image acquisition. The upward-looking camera is installed on the top of the robot, and the lens is vertically upward aligned with the bottom of the train. During the driving of the robot along the length direction of the train, the camera is triggered to capture high-definition images of the wheel shaft of each car at preset intervals. As a inherent structure of the train, the wheel shaft has a strict geometric correspondence with the car, and its shape is stable and not easily disturbed by the environment, making it an ideal feature carrier for calibrating the relative position of the robot and the train. During the shooting process, the camera adjusts the exposure parameters adaptively to cope with changes in the light at the bottom of the train, ensuring that key structures such as the edge of the wheel shaft and the hub are clearly imaged, providing high-quality image data for subsequent feature extraction.

[0058] After acquiring the wheelset image, the data processing module extracts wheelset features and analyzes the actual position through image recognition algorithms. Feature extraction focuses on the geometric key elements of the wheelset: first, the contour boundary of the wheelset is identified through an edge detection algorithm to distinguish the wheelset from other structures on the carriage floor; then, the hub center, rim edge, and other feature points are located through a shape matching algorithm, and the coordinates of these feature points form the feature set of the wheelset.

[0059] Based on the feature set, the module converts the image plane coordinates to three-dimensional space coordinates by combining the camera's internal parameters and the camera installation height, obtaining the actual position of the wheelset in the robot's local coordinate system. This conversion process is achieved through the principle of perspective projection, which restores two-dimensional image information to three-dimensional space position, ensuring the measurement accuracy of the actual position of the wheelset.

[0060] To quantify the relative position deviation of the robot and the train, the module compares the actual position of the wheelset with the preset standard wheelset template position. The standard wheelset template is a digital model based on the design parameters of the train, containing the theoretical position of the wheelset in the standard carriage coordinate system, the geometric relationship between feature points, and other data. During the template construction process, multiple sets of standard train wheelset data are collected for statistical optimization to ensure that it adapts to the common features of wheelsets of different marshalling trains.

[0061] The comparison process is achieved through coordinate mapping: the actual position of the wheelset is converted from the robot's local coordinate system to the standard carriage coordinate system, and then compared point by point with the theoretical position of the standard wheelset template to calculate the spatial distance difference of the corresponding feature points. The set of these differences constitutes the position deviation of the wheelset, which directly reflects the pose deviation of the robot relative to the current carriage, including translation deviation along the length of the carriage, lateral deviation perpendicular to the direction of the carriage, and angle deviation of the robot's attitude and the axis of the carriage.

[0062] According to the wheelset position deviation, the data processing module adjusts the pose of the robot relative to the train. The adjustment logic is based on the dimensional classification of the deviation: for translation deviation, the speed difference of the robot's drive wheels is controlled to achieve micro-displacement compensation in the forward or left-right direction; for angle deviation, the robot's steering mechanism is adjusted to correct the heading angle, ensuring that the robot's travel direction is parallel to the axis of the carriage. During the adjustment process, the module collects real-time motion feedback data of the robot, closes the loop to verify the adjustment effect, and continues until the deviation is reduced to within the preset threshold.

[0063] After the adjustment is completed, the module outputs the secondary corrected pose correction result, which contains three types of core information: six-degree-of-freedom correction value, quantified robot translation correction amount in X, Y and Z axis directions and rotation correction amount of heading angle, pitch angle and roll angle, which are used to accurately update the robot pose; the relative coordinate relationship between the robot and the train detection point, which clearly indicates the spatial distance and orientation between the current position of the robot and the detection targets such as bolts and pipelines in the train car, and provides a basis for the detection path planning of the mechanical arm; the pose calibration marker, which records the corresponding relationship parameters between the global three-dimensional point cloud map coordinate system and the local feature coordinate system of the train, and serves as a coordinate correlation reference for subsequent map reconstruction, ensuring the spatial consistency of the global map and the dynamic features of the train.

[0064] The pose correction result is output to the map reconstruction module through a data interface, and the pose calibration marker is used as a key correlation element to integrate the local correction data of the robot into the global map framework. This correlation mechanism enables the global three-dimensional point cloud map to not only contain environmental static features, but also integrate the position information of dynamic features of the train in real time, providing more accurate coordinate reference for the subsequent map calling of the inspection task.

[0065] Through directional capture, accurate comparison and quantitative adjustment of the wheel shaft feature, the secondary correction link effectively eliminates the pose deviation caused by individual differences of the train and cumulative errors of the robot navigation, so that the alignment accuracy of the robot and the train detection point reaches the millimeter level, providing a core guarantee for the accuracy of the underframe equipment detection.

[0066] The map reconstruction module reconstructs and generates the inspection map according to the three-dimensional point cloud map and the pose correction result.

[0067] The map reconstruction module realizes dynamic update and optimization of the inspection map based on the global three-dimensional point cloud map and the pose correction result output by the data processing module.

[0068] After receiving the pose correction result, the module fuses the coordinate mapping relationship, pose correction amount and other information in the result with the global three-dimensional point cloud map, and updates the key information such as the historical trajectory of the robot and the coordinates of the train detection points in the map. By integrating real-time correction data, the module can eliminate the cumulative errors caused by long-term driving of the robot, adapt to scenarios such as changes in train parking position and dynamic adjustment of equipment, and ensure that the map always maintains consistency with the actual environment. Finally, the module generates an inspection map containing global environmental features, robot driving trajectory and dynamic features of the train, providing accurate map support for the execution of subsequent inspection tasks.

[0069] Specifically, as Figures 1 to 5As shown, the map reconstruction module includes a coordinate mapping submodule for establishing a dynamic mapping relationship between a global three-dimensional point cloud map coordinate system and a train local feature coordinate system based on the pose calibration markers; a pose correction submodule for inversely superimposing the six-degree-of-freedom correction value onto the historical positioning trajectory of the robot in the global three-dimensional point cloud map to correct the cumulative error of the robot pose in the global map; and a feature updating submodule for updating the coordinate information of the train detection points in the global three-dimensional point cloud map according to the relative coordinate relationship between the robot and the train detection points in combination with the dynamic mapping relationship and completing dynamic optimization and reconstruction of the inspection map.

[0070] The core function of the coordinate mapping submodule is to establish a dynamic mapping relationship between a global three-dimensional point cloud map coordinate system and a train local feature coordinate system based on the pose calibration markers. The global three-dimensional point cloud map coordinate system is a spatial coordinate system established based on the initial positioning value of the robot, covering static features of the entire inspection environment, such as tracks, walls, and fixed equipment; and the train local feature coordinate system is a coordinate system based on the structure of the train itself, taking the wheel shaft and the car connection as references, reflecting the relative positions of the parts of the train.

[0071] The pose calibration markers contain two types of coordinate information: one type is the absolute coordinates of the marker points in the global three-dimensional point cloud map coordinate system, obtained by the three-dimensional laser radar module through global positioning; and the other type is the relative coordinates of the same marker points in the train local feature coordinate system, determined by the overhead camera in combination with the wheel shaft feature recognition. The submodule calculates the conversion parameters between the two coordinate systems, including the translation and rotation, by analyzing the double coordinates of these marker points, to form a conversion matrix. The conversion matrix can convert any point coordinate in the train local feature coordinate system to the global three-dimensional point cloud map coordinate system, or vice versa.

[0072] The core of the dynamic mapping relationship lies in real-time updating. Since the train stopping position may be adjusted, the submodule continuously receives new pose calibration marker data, dynamically adjusts the conversion matrix parameters by comparing the coordinate deviations of the new marker points and the historical marker points. When the train position changes, the conversion matrix is updated, ensuring that the association between the global coordinate system and the train local coordinate system is always consistent with the actual scene, providing an accurate coordinate reference for subsequent data fusion.

[0073] The pose correction submodule is used to inversely superimpose the six-degree-of-freedom correction value onto the historical positioning trajectory of the robot in the global three-dimensional point cloud map to correct the cumulative error of the robot pose in the global map. In the long-term driving process, the historical positioning trajectory of the robot may gradually deviate from the actual path due to factors such as uneven ground and wheel diameter wear, forming cumulative errors, which will cause the deviation between the global map and the actual environment to gradually expand if not corrected.

[0074] The reverse superposition refers to inversely distributing the six-degree-of-freedom correction values obtained by the secondary correction to each segment of the historical trajectory in time sequence. Specifically, the sub-module first extracts the complete historical positioning trajectory data of the robot from the starting point to the current position, including the pose parameters at each time; then, according to the time node of the secondary correction, the six-degree-of-freedom correction values are decomposed into correction components corresponding to the historical stages. For the translation correction amounts of the X-axis, Y-axis and Z-axis, the linear interpolation method is used to distribute them to each historical trajectory segment in proportion to the driving distance, ensuring that the correction amounts are evenly distributed along the trajectory length. For the rotation correction amounts of the heading angle, pitch angle and roll angle, the spherical linear interpolation method is used to distribute them in proportion to the time, avoiding sudden changes in rotation errors at the trajectory splicing points.

[0075] Through the reverse superposition, the pose parameters at each time in the historical trajectory are corrected to be consistent with the actual driving path, eliminating the long-term accumulated positioning deviation, so that the robot driving trajectory recorded in the global three-dimensional point cloud map completely coincides with the actual inspection path, providing a reliable data foundation for subsequent map analysis and backtracking.

[0076] The feature updating sub-module is used to update the coordinate information of the train detection points in the global three-dimensional point cloud map according to the relative coordinate relationship between the robot and the train detection points, in combination with the dynamic mapping relationship. The train detection points refer to key parts that need to be focused on detection, such as bolts, pipeline interfaces, shock absorbers, etc., and the accuracy of their coordinates in the global map directly affects the detection accuracy of the inspection robot.

[0077] The relative coordinate relationship between the robot and the train detection points is obtained by the data processing module through visual recognition, reflecting the distance and direction of the detection points relative to the current position of the robot. The sub-module first converts this relative coordinate to the local feature coordinate system of the train, and then further converts it to the absolute coordinate in the global three-dimensional point cloud map coordinate system through the conversion matrix established by the coordinate mapping sub-module.

[0078] After the conversion is completed, the sub-module compares the newly obtained absolute coordinate with the original stored coordinate of the detection point in the global map. If the spatial distance between the two exceeds the preset threshold, it means that the original coordinate can no longer reflect the actual position of the detection point, and the sub-module updates the global map data with the new coordinate; if the deviation is within the threshold range, the original coordinate remains unchanged. During the updating process, the sub-module synchronously records the timestamp of the coordinate update and the corresponding pose calibration marker version, realizing the traceability of the map data.

[0079] By continuously updating the global coordinates of the train detection points, the global three-dimensional point cloud map can dynamically adapt to scenarios such as changes in train stopping positions and slight displacement of equipment, ensuring that the positions of the detection points recorded in the map are always consistent with the actual positions, providing reliable map support for the accurate positioning and fault detection of the inspection robot.

[0080] The three sub-modules cooperatively realize the spatial correlation between the global and local through coordinate mapping, preserve the accuracy of the historical trajectory through pose correction, and keep the real-time of the detected points through feature updating, finally complete the dynamic optimization and reconstruction of the inspection map, so that the map always highly matches the actual scene of the rail transit vehicle inspection.

[0081] Specifically, as shown in Figures 1 to 5 The map reconstruction module further comprises: calling the pose calibration marker to analyze the marker point coordinates in the global three-dimensional point cloud map coordinate system and the marker point coordinates in the local feature coordinate system of the train, solving the conversion matrix of the two coordinate systems by the least square method, establishing a dynamic mapping relationship, and updating the dynamic mapping relationship in real time with the change of the train position; extracting the historical positioning trajectory data of the robot in the global three-dimensional point cloud map, and reversely superimposing the six-degree-of-freedom correction value to the historical pose at the corresponding time according to the time sequence, calculating the correction amount of each historical pose, wherein the X-axis, Y-axis and Z-axis correction amount is distributed to the historical trajectory segment by linear interpolation method, and the heading angle, pitch angle and roll angle correction amount compensates the rotation error by spherical linear interpolation method; according to the relative coordinate relationship between the robot and the detected points of the train, combining the conversion matrix to convert the local coordinates of the detected points into absolute coordinates in the global three-dimensional point cloud map coordinate system, comparing the converted coordinates with the original stored detected point coordinates in the map, if the deviation exceeds the preset threshold, updating the map data with the converted coordinates, and recording the coordinate update timestamp and the corresponding pose calibration marker version; integrating the corrected historical trajectory data and the updated detected point coordinate information to generate an inspection map including global environmental features, robot driving trajectory and train dynamic features.

[0082] Specifically, as shown in Figures 1 to 5 The map reconstruction module further comprises: after establishing the dynamic mapping relationship, optimizing the conversion matrix by the random sample consensus algorithm, eliminating abnormal mapping data caused by marker point recognition error, and retaining the conversion matrix parameters with a confidence higher than a preset threshold; after generating the inspection map, calculating the matching degree of the corrected historical trajectory and the static features in the global three-dimensional point cloud map, if the matching degree is lower than a preset threshold, re-establishing the coordinate mapping relationship and the subsequent steps until the matching degree is higher than the preset threshold, and the matching degree is calculated by the mean of the Euclidean distance of the point cloud overlap area.

[0083] The map reconstruction module first calls the pose calibration marker to analyze two types of key coordinates: one type is the absolute coordinates of the marker points in the global three-dimensional point cloud map coordinate system, which is continuously updated by the global positioning of the three-dimensional laser radar module, reflecting the fixed position of the marker points in the environment; the other type is the relative coordinates of the same marker points in the local feature coordinate system of the train, which is determined based on the wheel shaft feature recognition and dynamically adjusted with the change of the train position.

[0084] To establish the mathematical correlation of the two coordinate systems, the module solves the transformation matrix by the least squares method. The least squares method calculates the optimal translation and rotation parameters by minimizing the sum of the squares of the deviations of the marker points in the two coordinate systems. These parameters form a transformation matrix that can convert the coordinates of any point in the two coordinate systems. For example, when there is a deviation between the coordinates of the marker points in the global coordinate system and the coordinates in the local coordinate system of the train, the least squares method iteratively optimizes the parameters of the transformation matrix until the total deviation is minimized, ensuring the accuracy of the conversion. Based on this transformation matrix, the module establishes a dynamic mapping relationship between the global three-dimensional point cloud map coordinate system and the local feature coordinate system of the train. This relationship is updated in real time as the train's parking position is adjusted - whenever new pose calibration marker data is input, the module recalculates the transformation matrix parameters to ensure that the coordinate mapping always matches the actual scene.

[0085] The module extracts the historical positioning trajectory data of the robot in the global three-dimensional point cloud map, which records the complete driving path and corresponding pose parameters of the robot from the starting point to the current position. To eliminate the accumulated pose errors from long-term driving, the module reversely superimposes the six-degree-of-freedom correction values obtained from the secondary correction to the corresponding historical pose in time sequence, accurately calculating the correction amount of each historical pose.

[0086] The correction amount allocation adopts a differentiated strategy: for the translation correction amount of the X, Y, and Z axes, linear interpolation is used to allocate the correction amount according to the driving distance proportion of the historical trajectory segment. For example, if a certain historical trajectory segment accounts for 20% of the total trajectory, it is allocated 20% of the translation correction amount, ensuring that the correction amount is evenly distributed along the trajectory and avoiding excessive or insufficient local correction. For the rotation correction amount of the heading angle, pitch angle, and roll angle, spherical linear interpolation is used to allocate the correction amount according to the time proportion. This method smoothly transitions the rotation parameters in a three-dimensional spherical space, avoiding distortion in trajectory splicing caused by sudden changes in rotation angle, and ensuring the continuity of the historical trajectory's attitude. Through reverse superposition and differentiated allocation, the pose parameters at each time in the historical trajectory are corrected to be consistent with the actual driving path, eliminating the impact of accumulated errors on map accuracy.

[0087] Based on the relative coordinate relationship between the robot and the train detection points, the module converts the local coordinates of the detection points into absolute coordinates in the global three-dimensional point cloud map coordinate system combined with the transformation matrix. The relative coordinate relationship reflects the spatial distance and orientation of the detection points from the current position of the robot. Through the transformation matrix, it can be mapped from the local coordinate system of the train to the global coordinate system, achieving accurate positioning of the detection points in the environment.

[0088] After the transformation is completed, the module compares the newly obtained absolute coordinates with the original stored coordinates of the detection item point in the map. If the spatial deviation between the two exceeds the preset threshold, it means that the original coordinates can no longer reflect the actual position of the detection item point, and the module updates the map data with the transformed coordinates; if the deviation is within the threshold range, the original coordinates remain unchanged to reduce redundant calculations. To achieve traceability of the map data, the module simultaneously records the timestamp of the coordinate update and the corresponding pose calibration marker version - the specific time when the timestamp marker update occurs, and the version number is associated with the corresponding transformation matrix parameters, facilitating subsequent tracing of the basis for updating the map data and ensuring data reliability. Finally, the module integrates the corrected historical trajectory data and the updated detection item point coordinate information to generate an inspection map containing global environmental static features, complete robot travel trajectory, and train dynamic features.

[0089] After establishing the dynamic mapping relationship, the module optimizes the transformation matrix through the random sample consensus algorithm. The random sample consensus algorithm repeatedly randomly selects a portion of the marker point coordinate samples, calculates the corresponding transformation matrix, and checks the matching degree of the remaining marker points with the matrix, and selects the matrix parameters that cover the most marker points as the effective solution. This process can eliminate abnormal mapping data caused by coordinate misjudgment due to marker point recognition errors, such as local occlusion, and retain transformation matrix parameters with a confidence level higher than the preset threshold, further improving the stability of coordinate transformation.

[0090] After generating the inspection map, the module calculates the matching degree of the corrected historical trajectory and the static features in the global three-dimensional point cloud map. The matching degree is calculated by the average Euclidean distance of the overlapping area of the point cloud: select the static feature point cloud within a certain range around the historical trajectory (such as the track edge, fixed equipment), calculate the average spatial distance between the trajectory points and the corresponding static feature points, and the smaller the distance, the higher the matching degree. If the matching degree is lower than the preset threshold, it means that there is a deviation in the historical trajectory correction or coordinate transformation, and the module reverts to the coordinate mapping relationship establishment step, recalculates the transformation matrix and executes the subsequent correction process until the matching degree is higher than the preset threshold. This closed-loop verification mechanism ensures that the final generated inspection map is highly consistent with the actual environment, providing reliable map support for the inspection task.

[0091] Through the coherent process of coordinate transformation matrix solving, historical trajectory accurate correction, abnormal data elimination, and matching degree verification, the map reconstruction module effectively eliminates the influence of coordinate system deviation, cumulative error, and environmental interference, making the inspection map have both the stability of the global environment and the dynamic adaptability of the train features.

[0092] Specifically, as Figures 1 to 5As shown, the steps of the system's map construction method include collecting environment data through a closed-loop path scan, constructing an initial global three-dimensional point cloud map and performing rasterization and denoising preprocessing, and setting an initial positioning value of the robot; based on the initial positioning value, fusing inertial measurement data, odometry data, and laser radar matching results, outputting the optimal pose of the robot through Kalman filtering, controlling the robot to autonomously navigate to the target lane, predicting the parking deviation during the robot's travel process through the photoelectric positioning module and adjusting the trajectory in advance, detecting the vehicle head position through the cross photoelectric sensor after reaching the target lane, and completing the correction of the parking deviation; capturing the train axle image, calculating the pose deviation through the comparison of the axle image and the standard template, and outputting the pose correction result containing the six-degree-of-freedom correction value, the relative coordinate relationship, and the pose calibration mark; based on the pose correction result, establishing coordinate mapping, correcting the global pose error, and updating the detection point coordinates, and completing the dynamic reconstruction and storage of the inspection map.

[0093] The starting link of map construction is environment data collection and initial map generation. The system controls the robot to travel along the preset closed-loop path to complete the collection of environment data. The closed-loop path is designed to start from the starting point, cover the target inspection area, and return to the starting point to form a closed trajectory. This path can reduce the cumulative error of long-distance travel through consistency verification of the first and last data. During the travel process, the three-dimensional laser radar module synchronously collects three types of core data: three-dimensional point cloud data captures the spatial form of the environment static features, inertial measurement data records the acceleration and angular velocity changes of the robot motion, and odometry data quantifies the travel distance and direction.

[0094] Based on the collected multi-frame data, the system constructs an initial global three-dimensional point cloud map. During the construction process, the multi-frame point cloud data is unified to the same coordinate system through coordinate transformation, and then the frame deviation is eliminated through feature matching to form a complete environment space model. To improve the usability of the map, the system performs rasterization and denoising preprocessing on the initial map: rasterization divides the three-dimensional space into regular grids, aggregates point cloud data by grid units, reduces redundancy, and speeds up subsequent positioning matching; denoising eliminates isolated abnormal points caused by reflection and dust through multi-frame comparison and feature verification, and retains stable environment features. After preprocessing, the system sets the initial positioning value of the robot based on the starting point of the closed-loop path, combined with the surrounding fixed features (such as the lane edge and the right angle of the wall), as the origin reference of the map coordinate system.

[0095] Based on the initial positioning value, the system enters the autonomous navigation stage. The data processing module fuses three types of data to optimize positioning accuracy: inertial measurement data reflects the dynamic attitude of the robot in real time, odometry data records the incremental trajectory, and laser radar matching results provide absolute position reference through the alignment of features between the current frame and the preprocessed map frame. The three are fused through Kalman filtering algorithm: the filtering algorithm takes the laser radar matching result as the high-precision observation value, calibrates the cumulative drift of inertial measurement and odometry data, and outputs the continuous and stable optimal pose of the robot, providing accurate position and attitude reference for navigation.

[0096] Relying on the optimal pose, the system controls the robot to autonomously navigate along the planned path to the target track. During the driving process, the photoelectric positioning module starts the parking deviation prediction mechanism: the photoelectric sensors arranged across the train head area are scanned in real time, and the train head position is identified combined with the pre-set train head edge feature model, while integrating the current driving speed, remaining distance and error law under the same historical working condition to predict the deviation that may be generated when reaching the theoretical parking point. Based on the predicted deviation, the system adjusts the driving trajectory of the robot in advance, and compensates for the potential deviation through fine adjustment of the driving wheel speed. After reaching the target track, the photoelectric sensor detects the deviation between the actual train head position and the theoretical position again, calculates the final compensation amount combined with the predicted deviation, and completes the parking deviation correction through micro-displacement adjustment to ensure that the robot is parked at the pre-set detection starting position.

[0097] After parking correction, the system starts the second fine adjustment of the pose. The upward-looking camera takes pictures of the wheel shaft of each car of the train at preset intervals, and during the shooting process, the light is adjusted adaptively to ensure that the key features such as the wheel shaft edge and hub are clearly imaged. The data processing module extracts features from the wheel shaft image, identifies the wheel shaft contour boundary and key feature points, and then converts the image coordinates to three-dimensional space coordinates through perspective projection conversion to obtain the actual position of the wheel shaft.

[0098] The system compares the actual position of the wheel shaft with the standard wheel shaft template. The standard wheel shaft template is constructed based on the design parameters of the train, including the theoretical position of the wheel shaft in the standard car coordinate system and the geometric relationship of the features. By mapping the coordinates, the actual position and the template position are aligned, and the spatial deviation between the two is calculated, which directly quantifies the pose deviation of the robot relative to the current car. Based on the deviation analysis, the system outputs the pose correction result: the six-degree-of-freedom correction value clearly indicates the adjustment amount of translation and rotation direction, the relative coordinate relationship between the robot and the detection point marks the spatial orientation of the key detection part, and the pose calibration mark records the correlation parameters between the global map and the local coordinate system of the train, providing a coordinate reference for subsequent map updating.

[0099] Based on the pose correction result, the system enters the map dynamic optimization phase. The map reconstruction module first establishes coordinate mapping through the pose calibration marker: analyze the double coordinates of the marker point in the global map coordinate system and the train local coordinate system, solve the conversion matrix through the least square method, realize the accurate association of the two coordinate systems, and the mapping relationship is updated in real time with the train position fine tuning.

[0100] Subsequently, the system corrects the global pose error: the six-degree-of-freedom correction value is added to the robot historical positioning trajectory in reverse time sequence, the translation correction amount is distributed through linear interpolation, the rotation error is compensated by spherical linear interpolation, and the cumulative deviation of long-term driving is eliminated. At the same time, the feature update submodule converts the relative coordinates of the robot and the detection point into absolute coordinates in the global map coordinate system combined with the conversion matrix, compares the original stored coordinates, and if the deviation is out of limit, updates the data and records the time stamp and marker version. Finally, the system integrates the corrected historical trajectory, the updated detection point coordinates and the global environment features, generates a dynamically optimized inspection map, and completes data storage, providing accurate map support for subsequent inspection tasks.

[0101] Through the above steps, the system realizes the whole process closed loop from environment perception to map dynamic optimization, which not only ensures the global integrity of the initial map, but also adapts to the dynamic changes of the scene through real-time correction and update, and meets the high-precision positioning requirements of rail transit vehicle inspection.

[0102] The above shows and describes the basic features, principles and advantages of the present application. It should be noted that the present application is not limited by the above embodiments, and only some embodiments are provided. Without departing from the spirit and scope of the present application, several improvements and supplements are considered as the protection scope of the present application.

Claims

1. A patrol map construction system based on radar and photoelectric technology, characterized in that, include: The 3D LiDAR module is used to collect 3D point cloud data, inertial measurement data and odometer data of the environment, construct a global 3D point cloud map based on the collected data and output the robot's initial pose; The photoelectric positioning module includes a cross-arranged photoelectric sensor and a top-view camera. The photoelectric sensor is used to detect the position of the train's front end and correct the robot's parking deviation; the top-view camera is used to acquire images of the train's wheel axles. The data processing module is communicatively connected to the photoelectric positioning module and the 3D lidar module. It preprocesses the 3D point cloud map, combines the inertial measurement data, mileage data, and initial pose, and optimizes the robot's optimal pose through Kalman filtering, then outputs the real-time positioning result. It calculates the position deviation based on the matching result between the wheel axle image and the standard wheel axle template, and performs secondary correction on the robot's pose relative to the train based on the position deviation to obtain the pose correction result. The map reconstruction module reconstructs and generates the inspection map based on the 3D point cloud map and pose correction results.

2. The inspection map construction system based on radar and photoelectric technology according to claim 1, characterized in that, The three-dimensional LiDAR module includes a map acquisition and construction submodule, which is used to acquire three-dimensional LiDAR data, inertial measurement unit data and odometer data through robot closed-loop path scanning, generate a three-dimensional point cloud map composed of multiple frames of motion data, perform rasterization and noise reduction processing on the three-dimensional point cloud map, and set the initial positioning value of the robot relative to the coordinate system of the three-dimensional point cloud map.

3. The inspection map construction system based on radar and photoelectric technology according to claim 1, characterized in that, The data processing module includes a data acquisition and matching submodule. This submodule is used to analyze the initial pose of the robot based on the initial positioning value using the inertial measurement unit data and odometry data. It then matches the current frame of the lidar with the preprocessed 3D point cloud map frame to output the robot's six-degree-of-freedom pose, which includes X-axis position, Y-axis position, Z-axis position, yaw angle, pitch angle, and roll angle. Finally, it uses the six-degree-of-freedom pose as an observation value to fuse with the inertial measurement unit data and odometry data to output the optimal pose for the robot to reach the inspection target position.

4. The inspection map construction system based on radar and photoelectric technology according to claim 1, characterized in that, The secondary pose correction in the data processing module includes: capturing wheel and axle images of each train carriage using the upward-facing camera, extracting wheel and axle features to analyze the actual wheel and axle positions, comparing the actual wheel and axle positions with preset standard wheel and axle template positions, calculating the wheel and axle position deviation, which is the robot's pose deviation relative to the current carriage, adjusting the robot's pose relative to the train based on the wheel and axle position deviation, and outputting the pose correction result after secondary correction. The pose correction result includes the robot's six-degree-of-freedom correction value relative to the current carriage, the relative coordinate relationship between the robot and the train's detection points, and pose calibration markers for map reconstruction. The pose calibration markers are used to associate the coordinate mapping relationship between the global 3D point cloud map and the train's local features and are output to the map reconstruction module.

5. The inspection map construction system based on radar and photoelectric technology according to claim 4, characterized in that, The map reconstruction module includes a coordinate mapping submodule, which is used to establish a dynamic mapping relationship between the global three-dimensional point cloud map coordinate system and the train local feature coordinate system based on the pose calibration mark. The pose correction submodule is used to inversely superimpose the six-degree-of-freedom correction values ​​onto the robot's historical positioning trajectory in the global 3D point cloud map, thereby correcting the cumulative error of the robot's pose in the global map. The feature update submodule is used to update the coordinate information of the train inspection points in the global 3D point cloud map based on the relative coordinate relationship between the robot and the train inspection points and the dynamic mapping relationship, and to complete the dynamic optimization and reconstruction of the inspection map.

6. The inspection map construction system based on radar and photoelectric technology according to claim 1, characterized in that, The photoelectric positioning module corrects parking deviation by: during the robot's journey towards the preset theoretical parking point based on the optimal pose, cross-arranged photoelectric sensors scan the train's front area in real time, identify the front position through a preset front edge feature model, predict the parking deviation that will occur when reaching the theoretical parking point based on the current driving speed, the distance to the theoretical parking point, and historical driving error data, adjust the robot's driving trajectory according to the predicted parking deviation, and after reaching the theoretical parking point, use photoelectric sensors to detect the deviation between the actual front position and the theoretical position a second time, and combine the predicted parking deviation to make compensation adjustments to complete the correction of the parking deviation.

7. The inspection map construction system based on radar and photoelectric technology according to claim 5, characterized in that, The map reconstruction module also includes calling the pose calibration marker to parse the coordinates of the marker points in the global three-dimensional point cloud map coordinate system and the coordinates of the marker points in the train local feature coordinate system, solving the transformation matrix of the two coordinate systems by the least squares method, and establishing a dynamic mapping relationship. The dynamic mapping relationship is updated in real time as the train position changes. Historical positioning trajectory data of the robot is extracted from the global 3D point cloud map. The six-degree-of-freedom correction values ​​are then superimposed on the historical poses at the corresponding times in reverse order according to the time series. The correction amount for each historical pose is calculated. The correction amounts for the X-axis, Y-axis, and Z-axis are allocated to the historical trajectory segments using linear interpolation, while the correction amounts for the heading angle, pitch angle, and roll angle are compensated for rotational errors using spherical linear interpolation. Based on the relative coordinate relationship between the robot and the train detection points, the local coordinates of the detection points are converted into absolute coordinates in the global 3D point cloud map coordinate system using a transformation matrix. The converted coordinates are compared with the original stored coordinates of the detection points in the map. If the deviation exceeds a preset threshold, the map data is updated with the converted coordinates. At the same time, the coordinate update timestamp and the corresponding pose calibration mark version are recorded. The corrected historical trajectory data is integrated with the updated coordinate information of the detection points to generate an inspection map that includes global environmental features, robot trajectory, and train dynamic features.

8. The inspection map construction system based on radar and photoelectric technology according to claim 7, characterized in that, The map reconstruction module further includes, after establishing the dynamic mapping relationship, optimizing the transformation matrix using a random sampling consistency algorithm, eliminating abnormal mapping data caused by marker point identification errors, and retaining transformation matrix parameters with confidence levels higher than a preset threshold; after generating the inspection map, calculating the matching degree between the corrected historical trajectory and the static features in the global 3D point cloud map, if the matching degree is lower than a preset threshold, then re-establishing the coordinate mapping relationship and subsequent steps until the matching degree is higher than the preset threshold, wherein the matching degree is calculated using the mean Euclidean distance of the overlapping area of ​​the point cloud.

9. The inspection map construction system based on radar and photoelectric technology according to claim 1, characterized in that, The map construction method of the system includes the following steps: collecting environmental data through closed-loop path scanning, constructing an initial global 3D point cloud map and performing rasterization and noise reduction preprocessing, and setting the initial positioning value of the robot; based on the initial positioning value, fusing inertial measurement data, mileage data and lidar matching results, outputting the optimal pose of the robot through Kalman filtering, controlling the robot to autonomously navigate to the target track, the photoelectric positioning module predicts the parking deviation and adjusts the trajectory in advance during the robot's movement, and after arriving at the target track, the position of the train head is detected by the cross-type photoelectric sensor to complete the correction of the parking deviation; capturing images of the train wheel axles, calculating the pose deviation by comparing the wheel axle images with a standard template, and outputting the pose correction result including six-degree-of-freedom correction values, relative coordinate relationships and pose calibration marks; establishing coordinate mapping based on the pose correction result, correcting the global pose error and updating the coordinates of the detected items, and completing the dynamic reconstruction and storage of the inspection map.

Images