A multi-equipment inspection platform for rail transit

Abstract

This invention relates to the field of rail transit inspection technology and discloses a multi-equipment inspection platform for rail transit, comprising a central server, relay nodes, and inspection equipment. The relay nodes generate simulated equipment locations and schedule tasks. The inspection equipment achieves spatial continuity through fusion positioning and image comparison fine-tuning. The central server stitches together image data to form an inspection map. This application solves the positioning deviation problem in multi-equipment collaborative operations through dynamic fusion positioning, task priority evaluation, and a rapid succession strategy, improving the spatial continuity of inspection data and task scheduling efficiency, making it suitable for efficient inspection operations in complex environments.

Description

Technical Field

[0001] This invention relates to the field of rail transit inspection technology, and in particular to a multi-equipment inspection platform for rail transit. Background Technology

[0002] With the acceleration of urbanization, the scale of urban rail transit systems is constantly expanding, making their operational safety and reliability increasingly important. The safe operation of rail transit systems relies on regular inspections and timely maintenance to detect potential faults and anomalies in infrastructure such as tracks, signaling equipment, and power supply systems. Traditional manual inspection methods suffer from low efficiency, high missed detection rates, and poor real-time performance, making them unsuitable for the large-scale and complex inspection needs of modern rail transit. The current application of inspection robots enables rapid and accurate track inspection, with track inspection platforms playing a crucial role. The operational quality of these platforms determines the efficiency and accuracy of track inspections.

[0003] In existing systems, when multiple devices collaborate, the positioning mechanism is typically based on independent relative coordinate systems, lacking a unified spatial reference system for real-time calibration and adjustment. This design performs adequately for continuous single-device operation, but in dynamic multi-device collaborative scenarios, when one inspection device is scheduled to perform other tasks or experiences a power outage or offline event, its task is interrupted. In this case, the task needs to be reassigned to another inspection device. Since the main data collected by the inspection devices is images, ensuring the consistency of positioning between preceding and succeeding devices is crucial. However, during operation, accumulated positioning errors can lead to spatial deviations during task succession. This results in data misalignment between the preceding and succeeding devices, creating data gaps or misalignments, affecting the continuity and completeness of subsequent analysis. Summary of the Invention

[0004] The purpose of this invention is to overcome the aforementioned deficiencies of existing technologies and provide a multi-equipment inspection platform for rail transit. Through mechanisms such as relay node generation of simulated equipment locations, dynamic fusion of positioning results, and image comparison and fine-tuning, it solves the data splicing discontinuity problem caused by positioning deviations during multi-equipment collaborative operations, thereby improving the spatial continuity and integrity of inspection data. Furthermore, by employing task priority evaluation and a rapid succession strategy, it optimizes the scheduling efficiency of interrupted tasks and reduces resource waste caused by task switching.

[0005] To achieve the aforementioned objectives, the rail transit multi-equipment inspection platform provided by this invention includes a central server, multiple relay nodes, and multiple sets of inspection equipment. The central server is equipped with a task planning module to generate inspection tasks and receives image data from the inspection equipment for stitching and anomaly analysis. The relay nodes are deployed at key locations along the track line to record the initial position information of the inspection equipment as it passes, generate simulated equipment positions, and perform task scheduling and status synchronization. Each inspection equipment is equipped with a positioning sensor, an image acquisition module, and a communication module to execute inspection tasks and support continuation operations after task interruption. When a relay node detects a task interruption of an inspection equipment, it generates a simulated equipment position based on the task termination information of the previous equipment and sends this simulated position to the succeeding inspection equipment. The succeeding inspection equipment calculates a fused positioning result based on its own positioning data and the simulated equipment position, acquires an image frame at that position, compares it with the task termination image of the previous equipment, and completes position fine-tuning. The central server stitches together the inspection image data acquired by multiple equipment according to the finely calibrated position information to form a continuous inspection map.

[0006] Furthermore, the relay node includes a simulated location generation subunit, an image reference cache subunit, and a confidence evaluation subunit. The simulated location generation subunit generates a target simulated location. Its generation process is based on the task completion location, trajectory direction vector, and average running speed of the previous device, combined with a trajectory prediction model to generate multiple candidate locations. The optimal candidate location is then selected as the simulated device location through image similarity filtering in the image reference cache subunit. The image reference cache subunit stores image frames acquired at task termination, serving as a benchmark for subsequent fine-tuning image comparison. The confidence evaluation subunit calculates the confidence index of the sensor data and constructs a sensor confidence vector, which is used for weight allocation in subsequent fusion positioning results.

[0007] Furthermore, the location information collected by the inspection equipment comes from various sensor devices, including inertial sensors for recording the equipment's movement trajectory, laser rangefinders for sensing the shape of the surrounding space, and camera components for identifying environmental feature images. The platform determines the equipment's current location using a weighted calculation method based on the reliability of the location information measured by each type of sensor. Specifically, after receiving the simulated equipment location from the relay node, the inspection equipment calculates the weight coefficients of its own location data and the simulated equipment location based on its own positioning results and sensor confidence levels, ultimately obtaining a fused positioning result.

[0008] Furthermore, the platform dynamically adjusts the confidence weights of the corresponding sensors based on the offset error of the image comparison and corrects the subsequent fusion positioning results to improve the spatial alignment accuracy between devices. Specifically, when the inspection device acquires an image at the location of the fusion positioning result and compares it with the reference image, if the image matching degree is lower than a preset threshold, the weight coefficients of the relevant sensors are automatically reduced, and the fusion positioning result is recalculated.

[0009] Furthermore, the image comparison module employs a feature point matching method based on SIFT or SURF algorithms to extract key points between the current image and the reference image and calculate displacement vectors, thereby correcting the position of the inspection equipment to achieve fine-tuning. Specifically, the image comparison module first extracts feature points from the two images, establishes a correspondence through descriptor matching, and then calculates displacement vectors and applies them to correct the position of the inspection equipment.

[0010] Furthermore, if the image matching degree is lower than a set threshold, the platform controls the inspection device to make a slight movement within a preset range near the current position, and re-acquires image frames for repeated comparison until a match is successful or the area is marked as a stitching failure area. Specifically, the inspection device moves according to rules within a slight movement range centered on the fusion positioning result. After each movement, the image is re-acquired and compared. If multiple consecutive comparisons fail, the location is marked as a stitching failure area, and the relevant status is returned to the central server.

[0011] Furthermore, during the image stitching process, if the overlapping area of ​​the images is lower than a set ratio or the comparison error exceeds the tolerance threshold, the central server marks the image segment as an unstitchable area and displays an anomaly warning in the visualized inspection map. Specifically, the central server receives each image segment and its corresponding location marker, performs stitching judgment based on the image overlap index and spatial continuity constraints, and if the conditions are not met, the area is marked with a specific color in the visualization interface, indicating the existence of an inspection data breakpoint.

[0012] Furthermore, when scheduling task succession, the platform calculates a comprehensive priority based on the distance between the current location and the interruption location of the inspection equipment, its remaining power, and the current task load, prioritizing the scheduling of the equipment with the highest comprehensive priority to take over the task. Specifically, the central server calculates the priority score of each candidate device according to the following formula.

[0013] ,

[0014] in, The distance from the device to the interruption point is represented by E, the remaining power is represented by L, the current task load is represented by α, β, and γ, which are adjustable weight parameters.

[0015] Furthermore, after a detection task is interrupted, the relay node automatically identifies the segment to which the task belongs and the detection content, and sends the task instructions, reference images, and simulated device locations to the selected successor device within a short time, achieving rapid task recovery and seamless connection. Specifically, after the relay node obtains the device with the highest score, it immediately sends the successor task instructions, the previous device reference image, the simulated device location, and fine-tuning parameters and thresholds to ensure that the successor device can quickly enter the designated location and complete the task succession.

[0016] Furthermore, in other embodiments, the relay node may also integrate an IMU calibration module and an environment recognition unit to dynamically correct the position estimation model; the image comparison method may also employ a feature matching network based on deep learning to improve robustness in complex environments. Specifically, the IMU calibration module dynamically adjusts the parameters of the position estimation model by real-time correction of the drift error of the inertial measurement unit and combining the perception results of external conditions such as light and humidity from the environment recognition unit, thereby improving positioning accuracy.

[0017] Furthermore, the central server dynamically adjusts the receiving and processing priorities of inspection data from relay nodes based on the current system load. When the central server is under high load, it prioritizes receiving and processing high-priority task data, while delaying the processing of low-priority task data. Specifically, the central server monitors system resource utilization and dynamically adjusts the priority sorting rules of the data queue to ensure that high-priority task data can be processed in a timely manner.

[0018] Furthermore, the relay node is configured with a task reassignment strategy. When a relay node detects that an inspection device has gone offline or has not responded for a preset waiting time, it defines the inspection device as an offline device and triggers the task reassignment strategy. Specifically, the task reassignment strategy includes defining the currently executing task of the offline device as a task to be reassigned, evaluating the priority of the task, and reassigning the unexecuted task and the task to be reassigned based on the task priority. If the corresponding relay node cannot complete the reassignment, it generates a task rollback signal to the central server and redefines the task scope. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the system architecture of the rail transit multi-equipment inspection platform of the present invention;

[0020] Figure 2 This is a schematic diagram illustrating the process by which relay nodes generate simulated device locations;

[0021] Figure 3 This is a flowchart of the inspection equipment's position fine-tuning based on image comparison;

[0022] Figure 4This is a schematic diagram of the task succession scheduling priority scoring calculation logic;

[0023] Figure 5 This is a flowchart of the central server stitching together inspection images and marking abnormal areas.

[0024] The attached figures are labeled as follows:

[0025] 1. Central server; 2. Relay node; 3. Inspection equipment; 4. Task planning module; 5. Simulated position generation subunit; 6. Image reference cache subunit; 7. Confidence assessment subunit; 8. Image comparison module; 9. Task interruption detection module. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] It should be noted that when a component is described as "fixed to" another component, it can be directly on the other component or may have a component in between. When a component is considered "connected to" another component, it can be directly connected to the other component or may have a component in between. When a component is considered "set on" another component, it can be directly set on the other component or may have a component in between. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0028] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0029] The rail transit multi-equipment inspection platform of the present invention achieves seamless splicing and spatiotemporal continuity of cross-equipment inspection data through the collaborative work of multiple modules. Its specific implementation is as follows. (Combined with...) Figures 1 to 5 The accompanying reference numerals are explained in detail.

[0030] The specific implementation method of the multi-equipment inspection platform for rail transit of the present invention is combined with the appendix. Figure 1 To be continued Figure 5 Detailed explanation provided. See attached document. Figure 1As shown, the platform includes a central server 1, relay nodes 2, and inspection equipment 3. The central server 1 and relay nodes 2 are connected via a wireless communication network. Relay nodes 2 are deployed at key locations along the track to record the initial position information of the inspection equipment 3 as it passes and generate simulated equipment positions. Inspection equipment 3 is equipped with positioning sensors, an image acquisition module, and a communication module, and interacts with relay nodes 2 and the central server 1 via the wireless communication module. The central server 1 integrates a task planning module 4 to generate inspection tasks and receive image data from the inspection equipment 3 for stitching and anomaly analysis. Relay nodes 2 include a simulated position generation subunit 5, an image reference cache subunit 6, and a confidence assessment subunit 7, used to generate simulated equipment positions, store image frames acquired when the task terminates, and calculate confidence indices for sensor data. Inspection equipment 3 completes inspection tasks through its positioning sensors, image acquisition module, and communication module and supports resuming operations after task interruption.

[0031] As attached Figure 2 As shown, when relay node 2 detects an interruption in the task of inspection device 3, the task interruption detection module 9 first obtains the task termination position information of the previous device. Then, the simulated position generation subunit 5 generates multiple candidate positions based on the task termination position, track direction vector, and average running speed, combined with a trajectory prediction model. These candidate points are then filtered using image similarity in the image reference cache subunit 6 to select the optimal candidate point as the simulated device position. For example, in a practical application scenario, assuming that inspection device 3 experiences a task interruption due to insufficient power at a certain point on the track, relay node 2 will generate three candidate points A, B, and C using the trajectory prediction model based on the position information before the interruption, the track direction vector, and the device's running speed. Next, the image reference cache subunit 6 retrieves the image frame acquired at the time of task termination and compares it with the environmental images corresponding to candidate points A, B, and C, respectively, ultimately selecting the candidate point with the highest image similarity as the simulated device position. This process ensures the accuracy of the simulated device position, providing a reliable foundation for the subsequent resuming of inspection device 3.

[0032] The trajectory prediction model employs a recursive prediction algorithm based on extended Kalman filtering. Input parameters include the task completion coordinates (x0, y0) of the previous inspection device, the trajectory direction vector θ, the average operating speed v, and the historical operating acceleration variance σ of the device. 2 The model predicts the equation:

[0033] x(t) = x0 + v × t × cosθ + 0.5 × a × t 2 ×cosθ, y(t)=y0+v×t×sinθ+0.5×a×t 2×sinθ generates candidate points for future locations at time t, where t ranges from 0.1 to 1 second and the step size is 0.1 seconds. Here, a is an acceleration estimate based on historical data.

[0034] As attached Figure 3 As shown, after receiving the simulated device location from relay node 2, the continuous inspection device 3 calculates the fused positioning result by combining its own positioning data. Specifically, the positioning sensors of inspection device 3 collect visual inertial odometry data, LiDAR scanning results, and visual SLAM output. The confidence evaluation subunit 7 of relay node 2 constructs a sensor confidence vector based on the confidence of each positioning data point and assigns weight coefficients. For example, assuming the confidence of the visual inertial odometry data is 0.8, the confidence of the LiDAR scanning results is 0.7, and the confidence of the visual SLAM output is 0.9, the fused positioning result is calculated using a weighted average method. Inspection device 3 acquires image frames at the location of the fused positioning result and compares them with the image of the previous device task termination through image comparison module 8 to complete position fine-tuning. Image comparison module 8 uses a feature point matching method based on SIFT or SURF algorithms to extract key points between the current image and the reference image and establish a correspondence through descriptor matching, thereby calculating the displacement vector and applying it to the position correction of inspection device 3. If the image matching degree is lower than a preset threshold, the platform controls the inspection device 3 to make a slight movement near its current position and re-acquire image frames for repeated comparison until a match is successful or it is marked as a stitching failure area. For example, in a certain real-world scenario, the inspection device 3 acquires an image at the location of the fused positioning result and compares it with a reference image, finding that the matching degree is only 60%, lower than the preset threshold of 80%. At this time, the inspection device 3 moves within a slight movement range centered on the fused positioning result according to preset rules, re-acquiring and comparing images after each movement until the matching degree reaches the threshold or after three consecutive failed comparisons, it is marked as a stitching failure area.

[0035] As attached Figure 4 As shown, when relay node 2 detects a task interruption, it needs to quickly schedule the next inspection device 3. The central server 1 calculates a comprehensive priority score S based on the distance d between the current location and the interruption location of inspection device 3, the remaining power E, and the current task load L. Specifically, the scoring formula is:

[0036] ,

[0037] Where, d max E represents the maximum distance. max L represents the maximum battery capacity. maxThis represents the maximum task load, where α, β, and γ are adjustable weight parameters. For example, in a real-world scenario, assuming the first inspection device 3 is 50 meters from the interruption point, has 80% remaining battery power, and a current task load of 30%, while the maximum distance is 100 meters, the maximum battery power is 100%, and the maximum task load is 50%, with weight parameters α=0.4, β=0.3, and γ=0.3 respectively, then the priority score for the first inspection device 3 is:

[0038] S=0.4×(1-50 / 100)+0.3×(80 / 100)+0.3×(1-30 / 50)=0.62.

[0039] Based on the scoring results, the central server 1 selects the device with the highest overall priority to take over the task, and sends the task instructions, reference images, and simulated device locations to the selected device to ensure rapid task recovery and seamless transition.

[0040] As attached Figure 5 As shown, the central server 1 processes the received inspection image data during the image stitching process. Specifically, the central server 1 receives each image segment and its corresponding location marker, and makes stitching judgments based on image overlap indicators and spatial continuity constraints. For example, in a real-world scenario, assuming the first inspection device 3 and the second inspection device 3 each acquire two image segments, the central server 1 first calculates the overlap ratio of the two image segments. If the overlap ratio is lower than the set 30% threshold or the comparison error exceeds the tolerance threshold, the image segment is marked as an unstitchable area, and an anomaly is indicated in the visualized inspection map. For example, if the overlap ratio between the image acquired by the first inspection device 3 and the image acquired by the second inspection device 3 is only 20%, which is lower than the set 30% threshold, the central server 1 will mark the area in red in the visualization interface, indicating the existence of an inspection data breakpoint. In addition, the central server 1 also dynamically adjusts the receiving and processing priorities of inspection data from relay node 2 according to the current system load. For example, when the central server 1 is under high load, it prioritizes receiving and processing high-priority task data and delays processing low-priority task data. Central server 1 monitors system resource utilization and dynamically adjusts the priority sorting rules of the data queue to ensure that high-priority task data can be processed in a timely manner.

[0041] As attached Figure 1 and attached Figure 2As shown, after detecting a task interruption, relay node 2 automatically identifies the segment and content of the task, and sends the task instructions, reference image, and simulated device location to the selected successor device within a short time. For example, in a real-world scenario, assuming the first inspection device 3 experiences a task interruption due to a malfunction, relay node 2, after obtaining the second inspection device 3 with the highest score, immediately sends the successor task instructions, the previous device reference image, the simulated device location, and fine-tuning parameters and thresholds to ensure the successor device can quickly enter the designated location and complete the task succession. Furthermore, relay node 2 is also configured with a task reassignment strategy. When relay node 2 detects that inspection device 3 has gone offline or has not responded for more than a preset waiting time, it defines the inspection device as an offline device and triggers the task reassignment strategy. For example, assuming the first inspection device 3 goes offline during task execution and has not responded for more than a preset 5-minute waiting time, relay node 2 defines the inspection device as an offline device, defines the currently executing task as a pending task, evaluates the task priority, and reassigns the unexecuted task and the pending task based on the task priority. If the corresponding relay node 2 cannot complete the reallocation, a task rollback signal is generated to the central server 1 and the task scope is redefined.

[0042] In other embodiments, relay node 2 can also integrate an IMU calibration module and an environment recognition unit to dynamically correct the position estimation model. For example, the IMU calibration module can dynamically adjust the parameters of the position estimation model by real-time correction of the drift error of the inertial measurement unit and combining the perception results of external conditions such as lighting and humidity by the environment recognition unit, thereby improving positioning accuracy. Image comparison can also employ deep learning-based feature matching networks, such as SuperGlue and LoFTR, to improve robustness in complex environments. For example, assuming that in a tunnel environment, the traditional SIFT algorithm has a low image matching degree due to poor lighting conditions, the platform can switch to a deep learning-based feature matching network to extract more robust feature points through a trained model, thereby improving the success rate of image comparison. To better enable those skilled in the art to fully understand and implement this invention, the specific implementation principle of this invention is further explained below with reference to a specific application scenario.

[0043] In the actual operation of the multi-equipment inspection platform for rail transit, it is first necessary to clarify the collaborative workflow between the central server 1, relay node 2, and inspection equipment 3. (See attached...) Figure 1 As shown, the central server 1 is connected to the relay node 2 via a wireless communication network. The relay node 2 is deployed at key locations on the track line to record the initial position information of the inspection equipment 3 as it passes by and to generate a simulated equipment position. The inspection equipment 3 completes the inspection task through its positioning sensor, image acquisition module, and communication module, and supports resuming operations after task interruption.

[0044] When relay node 2 detects an interruption in the task of inspection device 3, task interruption detection module 9 obtains the task termination position information of the previous device. Subsequently, the simulated position generation subunit 5 generates multiple candidate positions based on the task termination position, trajectory direction vector, and average running speed, combined with a trajectory prediction model. These candidate points are then filtered using image similarity in image reference cache subunit 6 to select the optimal candidate point as the simulated device position.

[0045] Image similarity is calculated using the Structural Similarity Index (SSIM), and the formula is as follows:

[0046] SSIM(x,y)=(2μxμy+C1)(2σxy+C2) / [(μx 2 +μy 2 +C1)(σx 2 +σy 2 +C2)], where μ is the mean, σ is the variance, and σxy is the covariance.

[0047] C1=(K1L) 2 C2=(K2L) 2 Where K1=0.01, K2=0.03, L=255; during the screening, candidate points with SSIM values ​​greater than 0.85 are selected. If there are multiple points that meet the conditions, the position corresponding to the maximum SSIM value is selected as the position of the simulation device.

[0048] For example, in a real-world scenario, suppose inspection device 3 experiences a task interruption due to insufficient power. Relay node 2, based on its pre-interruption position information, track direction vector, and device speed, uses a trajectory prediction model to generate three candidate points A, B, and C. Next, image reference buffer subunit 6 retrieves the image frames acquired at the time of task termination and compares them with the environmental images corresponding to candidate points A, B, and C, respectively. Finally, the candidate point with the highest image similarity is selected as the simulated device location. This process ensures the accuracy of the simulated device location, providing a reliable foundation for the subsequent resuming of inspection device 3.

[0049] After receiving the simulated device location from relay node 2, the follow-up inspection device 3 calculates the fused positioning result by combining its own positioning data. Specifically, the positioning sensors of inspection device 3 collect visual inertial odometry data, lidar scan results, and visual SLAM output. The confidence assessment subunit 7 of relay node 2 constructs a sensor confidence vector based on the confidence of each positioning data point and assigns weight coefficients. For example, assuming the confidence of the visual inertial odometry data is 0.8, the confidence of the lidar scan results is 0.7, and the confidence of the visual SLAM output is 0.9, the fused positioning result is calculated using a weighted average method. The weights in the weighted calculation are assigned based on the sensor confidence vector, which is calculated by the confidence assessment subunit using the reciprocal of the sensor error variance: wi = 1 / σ i 2 / Σ(1 / σ j 2 ), where σ i The standard deviation of the position measurement for the i-th type of sensor is σ = 0.1 m by default for inertial sensors, σ = 0.05 m by default for laser ranging, and σ = 0.2 m by default for cameras. When the displacement vector error of image matching is > 0.1 m, the camera weight is reduced to 50% of the original weight, and the fused position is recalculated.

[0050] The inspection device 3 acquires image frames at the location of the fused positioning result and compares them with the image from the previous device's task termination via the image comparison module 8 to complete position fine-tuning. The image comparison module 8 uses a feature point matching method based on SIFT or SURF algorithms to extract key points between the current image and the reference image and establish a correspondence through descriptor matching. It then calculates the displacement vector and applies it to the position correction of the inspection device 3. If the image matching degree is lower than a preset threshold, the platform controls the inspection device 3 to make a slight movement near the current position and re-acquire image frames for repeated comparison until a successful match is achieved or the area is marked as a stitching failure.

[0051] When relay node 2 detects a task interruption, it needs to quickly schedule the next inspection device 3. The central server 1 calculates a comprehensive priority score S based on the distance d between the current location and the interruption location of inspection device 3, the remaining power E, and the current task load L. Specifically, the scoring formula is:

[0052] ,

[0053] Where, d max E represents the maximum distance. max L represents the maximum battery capacity. maxThe maximum task load is represented by α, β, and γ, which are adjustable weight parameters. The adjustable weight parameters α, β, and γ range from 0 to 1, and satisfy α + β + γ = 1. Specifically, the default value for α (distance weight) is 0.4, suitable for short-distance inspection scenarios; the default value for β (battery weight) is 0.3, ensuring device endurance; and the default value for γ (load weight) is 0.3, prioritizing idle devices. These values ​​can be dynamically adjusted according to the urgency of the task. For example, α can be increased to 0.5 for urgent tasks, and β can be increased to 0.4 for non-urgent tasks. For instance, in a real-world scenario, assuming the first inspection device 3 is 50 meters from the interruption point, has 80% remaining battery power, and a current task load of 30%, while the maximum distance is 100 meters, the maximum battery power is 100%, and the maximum task load is 50%, with weight parameters α = 0.4, β = 0.3, and γ = 0.3 respectively, then the priority score for the first inspection device 3 is:

[0054] S=0.4×(1-50 / 100)+0.3×(80 / 100)+0.3×(1-30 / 50)=0.62.

[0055] Based on the scoring results, the central server 1 selects the device with the highest overall priority to take over the task, and sends the task instructions, reference images, and simulated device locations to the selected device to ensure rapid task recovery and seamless transition.

[0056] Central server 1 processes the received inspection image data during the image stitching process. Specifically, central server 1 receives each image segment and its corresponding location marker, and makes stitching judgments based on image overlap indicators and spatial continuity constraints. For example, in a real-world scenario, assuming the first inspection device 3 and the second inspection device 3 each acquire two image segments, central server 1 first calculates the overlap ratio of the two image segments. If the overlap ratio is lower than a set 30% threshold or the comparison error exceeds the tolerance threshold, the image segment is marked as an unstitchable area, and an anomaly is indicated in the visualized inspection map.

[0057] The tolerance threshold is defined as a pixel offset ≤ 3 pixels, corresponding to an actual physical distance ≤ 5 cm. The comparison error is calculated using the average Euclidean distance of the feature point matching, as shown in the formula:

[0058] ,

[0059] (xi',yi') are the coordinates of the feature point in the current image, and (xi,yi) are the coordinates of the corresponding point in the reference image. When d > 3 pixels, it is determined that the tolerance threshold has been exceeded.

[0060] For example, if the overlap between the images captured by the first inspection device 3 and the second inspection device 3 is only 20%, which is lower than the set threshold of 30%, the central server 1 will mark this area in red on the visualization interface, indicating the existence of an inspection data breakpoint. Furthermore, the central server 1 dynamically adjusts the receiving and processing priorities of inspection data from relay node 2 based on the current system load. For example, when the central server 1 is under high load, it prioritizes receiving and processing high-priority task data, delaying the processing of low-priority task data. By monitoring system resource utilization, the central server 1 dynamically adjusts the priority sorting rules of the data queue to ensure that high-priority task data is processed in a timely manner.

[0061] After detecting a task interruption, relay node 2 automatically identifies the segment and content of the task, and quickly sends the task instructions, reference image, and simulated device location to the selected successor device. For example, in a real-world scenario, if the first inspection device 3 malfunctions and the task is interrupted, relay node 2, after obtaining the second inspection device 3 with the highest score, immediately sends the successor task instructions, the previous device reference image, the simulated device location, and fine-tuning parameters and thresholds to ensure the successor device can quickly enter the designated location and complete the task succession. Furthermore, relay node 2 is configured with a task redistribution strategy. When relay node 2 detects that inspection device 3 is offline or has not responded for more than a preset waiting time, it defines the inspection device as an offline device and triggers the task redistribution strategy. For example, if the first inspection device 3 goes offline during task execution and has not responded for more than a preset 5-minute waiting time, relay node 2 defines the inspection device as an offline device, defines the currently executing task as a pending task, evaluates the task priority, and redistributes the unexecuted tasks and pending tasks based on the task priority. If the corresponding relay node 2 cannot complete the reallocation, a task rollback signal is generated to the central server 1 and the task scope is redefined.

[0062] In other implementations, relay node 2 can also integrate an IMU calibration module and an environment recognition unit to dynamically correct the position estimation model. For example, the IMU calibration module can dynamically adjust the parameters of the position estimation model by real-time correction of the drift error of the inertial measurement unit and combining the perception results of external conditions such as lighting and humidity by the environment recognition unit, thereby improving positioning accuracy. Image comparison can also employ deep learning-based feature matching networks, such as SuperGlue and LoFTR, to improve robustness in complex environments. For example, assuming that in a tunnel environment, the traditional SIFT algorithm has a low image matching rate due to poor lighting conditions, the platform can switch to a deep learning-based feature matching network to extract more robust feature points through a trained model, thereby improving the success rate of image comparison.

[0063] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention. Contents not described in detail in the specification are all prior art known to those skilled in the art, and the model parameters of each electrical appliance are not specifically limited; conventional equipment can be used. Electrical control components not mentioned in this technical solution are prior art and are therefore not shown in the figures and will not be described further here.

Claims

1. A multi-equipment inspection platform for rail transit, characterized in that, include: The central server (1) is equipped with a task planning module (4) to generate inspection tasks and receive image data from the inspection equipment (3) for splicing and anomaly analysis. Multiple relay nodes (2) are deployed at key locations on the track line to record the initial position information when the inspection equipment (3) passes by, generate simulated equipment positions, and perform task scheduling and state synchronization. The relay node (2) includes a simulated position generation subunit (5) which generates multiple position candidate points based on the task end position, track direction vector and average running speed of the previous inspection equipment, combined with the trajectory prediction model, and selects the optimal candidate point as the simulated equipment position through the image similarity in the image reference cache subunit (6). Multiple inspection devices (3), each equipped with a positioning sensor, an image acquisition module and a communication module, are used to perform inspection tasks and support resume operations after task interruption. When the relay node detects that the inspection equipment task is interrupted, it generates a simulated equipment location based on the task termination information of the previous inspection equipment and sends it to the next inspection equipment. The continuous inspection equipment calculates the fusion positioning result based on its own positioning data and the position of the simulated equipment, and collects image frames at the position of the fusion positioning result, compares them with the termination image of the previous equipment task, and the continuous inspection equipment performs position fine adjustment based on the comparison result until the image frame coincides with the termination image; the positioning information collected by the continuous inspection equipment (3) comes from multiple sensors, and the calculation steps of the fusion positioning result include the continuous inspection equipment constructing a sensor confidence vector based on the confidence of the positioning data of each sensor after receiving the simulated equipment position sent by the relay node (2), and assigning weight coefficients, and calculating the fusion positioning result according to the weighted average method; The central server stitches together the inspection image data collected by the previous inspection equipment and the subsequent inspection equipment based on the finely adjusted and calibrated location information to form a continuous inspection map.

2. The multi-equipment inspection platform for rail transit according to claim 1, characterized in that, The relay node (2) includes: The image reference buffer subunit (6) is used to store the image frames acquired when the task is terminated, as a reference for subsequent fine-tuning image comparison; The confidence assessment subunit (7) is used to calculate the confidence index of sensor data and construct the sensor confidence vector for subsequent weight allocation of the fusion positioning results.

3. The multi-equipment inspection platform for rail transit according to claim 1, characterized in that, The location information collected by the inspection equipment (3) comes from a variety of sensors, including an inertial sensor for recording the movement trajectory of the equipment, a laser ranging device for sensing the shape of the surrounding space, and a camera component for identifying environmental feature images. The platform determines the current position of the equipment by weighted calculation based on the reliability of the location information measured by various sensors.

4. The multi-equipment inspection platform for rail transit according to claim 1, characterized in that, The inspection device (3) acquires image frames at the location of the fusion positioning result and compares them with the image of the previous device task termination through the image comparison module (8) to complete the position fine adjustment. The image comparison module (8) uses the feature point matching method to extract the key points between the current image and the reference image and calculate the displacement vector to correct the position of the inspection device (3).

5. The multi-equipment inspection platform for rail transit according to claim 4, characterized in that, In the image comparison module (8), if the image comparison matching degree is lower than the set threshold, the platform controls the inspection device (3) to move within a preset range near the current position and re-acquire image frames for repeated comparison until the matching is successful or it is marked as a stitching failure area.

6. The multi-equipment inspection platform for rail transit according to claim 1, characterized in that, During the image stitching process, if the overlapping area of ​​the images collected by the previous inspection device and the subsequent inspection device is lower than the set ratio or the comparison error exceeds the tolerance threshold, the central server (1) will mark the image collected by the subsequent inspection device as an unstitchable area and make an abnormal prompt in the visual inspection map.

7. The multi-equipment inspection platform for rail transit according to claim 1, characterized in that, When the platform performs task succession scheduling, it calculates the comprehensive priority based on the distance between the current location and the interruption location of the inspection equipment (3), the remaining power, and the current task load, and prioritizes scheduling the equipment with the highest comprehensive priority to take over the task.

8. The multi-equipment inspection platform for rail transit according to claim 1, characterized in that, The relay node (2) is configured with a task reassignment strategy. When the relay node (2) detects that the inspection device (3) is offline or has not responded for more than a preset waiting time, it defines the inspection device as an offline device and triggers the task reassignment strategy. The task reassignment strategy includes defining the task currently being executed by the offline device as a task to be assigned, evaluating the priority of the task, and reassigning the unexecuted task and the task to be assigned based on the task priority.

9. The multi-equipment inspection platform for rail transit according to claim 6, characterized in that, After the detection task is interrupted, the relay node (2) automatically identifies the segment to which the task belongs and the detection content, and sends the task instructions, reference images and simulation device locations to the selected successor device in a short time.

Images