Railway line inspection route making and self-adaptive adjusting method based on unmanned aerial vehicle

Abstract

The invention discloses an unmanned aerial vehicle-based railway line routing inspection route making and adaptive adjustment method, and relates to the technical field of unmanned aerial vehicle intelligent routing inspection. According to the method, feature marking is performed on the basis of a three-dimensional map to form an inspection reference model; in the inspection process, smooth switching of respective positioning modes of the strong signal area and the weak signal area is realized through a positioning transition section, and meanwhile, an IMU and an odometer are assisted to guarantee stable attitude; a precise global space frame is provided by means of the inspection reference model constructed by multi-source data, a strong and weak signal area differential positioning mode is combined with a positioning transition section switching mechanism, positioning drift in a weak signal environment is effectively avoided, the positioning precision along a railway in a complex terrain is improved, and the positioning accuracy of the railway in the complex terrain is improved. The unmanned aerial vehicle is ensured to be aligned with the inspection feature points at the preset angle all the time, accuracy and integrity of inspection data are guaranteed, and the defects of dependence on a single sensor and poor positioning stability in the prior art are overcome.

Description

Technical Field

[0001] This invention belongs to the field of unmanned aerial vehicle (UAV) intelligent inspection technology, specifically a method for formulating and adaptively adjusting railway inspection routes based on UAVs. Background Technology

[0002] With the development of railway transportation towards high-speed and networked operations, the demand for maintenance and inspection along railway lines is becoming increasingly urgent. Drones, with their advantages of high flexibility, high inspection efficiency, and low cost, have been widely applied in railway line inspection. Current drone railway inspection technologies typically begin by acquiring basic geographic data along the railway line to construct a simple two-dimensional or three-dimensional map, and then plan fixed inspection routes based on this map. In terms of positioning, most rely on a single GPS / RTK positioning technology, with some solutions supplementing it with visual sensors to assist positioning and attempt to improve positioning stability during the inspection process.

[0003] Existing technologies still have significant shortcomings in practical applications, making it difficult to meet the high-precision inspection requirements along railway lines with complex terrain. On the one hand, the map models constructed by existing technologies have low accuracy, providing only basic geographic outline references, resulting in a lack of reliable global benchmark support for positioning. On the other hand, the positioning method is poorly designed; relying solely on GPS / RTK positioning in areas with weak GPS signals, such as tunnel entrances, deep valleys, and dense forests, easily leads to positioning failures or significant drops in accuracy. These problems result in poor positioning stability of existing technologies during railway line inspections in complex terrain, leading to insufficient accuracy and completeness of inspection data, severely impacting inspection quality and the reliability of subsequent defect identification.

[0004] This invention provides a method for railway inspection route planning and adaptive adjustment based on unmanned aerial vehicles (UAVs) to solve the above-mentioned technical problems. Summary of the Invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes a method for railway inspection route planning and adaptive adjustment based on unmanned aerial vehicles (UAVs).

[0006] To achieve the above objectives, a first aspect of the present invention provides a method for railway line inspection route planning and adaptive adjustment based on unmanned aerial vehicles (UAVs), comprising: A baseline model for the inspection of the target railway is constructed based on the basic modeling data, and the basic inspection route of the UAV is planned according to the baseline model; the basic modeling data is obtained through database or survey. During the inspection process, the positioning mode is switched based on the real-time location of the UAV, enabling the UAV to inspect the target railway along the basic inspection route; the switching of positioning mode is used to improve positioning accuracy. The inspection baseline model is updated by collecting environmental data in real time using drones; the route is replanned based on the updated inspection baseline model, and the inspection work is completed according to the replanned basic inspection route.

[0007] In one possible implementation, an inspection benchmark model of the target railway is constructed based on the basic modeling data, including: Acquire basic modeling data for the target railway, and construct a 3D map along the target railway based on the basic modeling data; the basic modeling data includes GIS data, DEM topographic data, and oblique photogrammetry data; Feature marking is performed on a 3D map to generate an inspection baseline model; the feature marking includes feature points, feature areas and static obstacles, and the feature points include inspection feature points.

[0008] In one possible implementation, the positioning method based on the real-time location switching of the drone includes: The drone's real-time location is used to determine whether it is in a positioning transition phase; the positioning transition phase is set according to the inspection benchmark model. Yes, the positioning method is switched according to the relative position of the positioning transition segment and the weak signal area; the positioning method includes a visual positioning scheme for the weak signal area and an RTK positioning scheme for the strong signal area. No, keep the positioning method unchanged.

[0009] In one possible implementation, a positioning transition segment is set, including: Identify weak signal areas in the inspection baseline model; Based on the entry and exit threshold of the weak signal area, positioning transition sections are set on both sides of the weak signal area according to the preset route length; the preset route length is 5 meters.

[0010] In one possible implementation, a visual positioning scheme is used to inspect weak signal areas, including: The drone's position is determined using a feature point database from a visual sensor and an inspection baseline model; the feature point database includes feature points and their corresponding coordinate data. The local real-time map is calibrated based on the feature point database, and the UAV is adjusted using the calibrated local real-time map to ensure that the UAV completes the inspection of the feature points.

[0011] In one possible implementation, the feature points in the feature point database also include fixed feature points and their coordinate data; Fixed feature points are preferentially selected from the uniformly distributed basic structures on the target railway; the basic structures include contact wire poles, track edges, mileage markers, and tunnel entrances.

[0012] In one possible implementation, the real-time inspection route is corrected based on route deviation during the inspection process, including: The inspection deviation between the actual inspection route of the UAV and the basic inspection route is calculated in real time; the inspection deviation includes route deviation or relative deviation of feature points. If the inspection deviation does not exceed the preset inspection deviation threshold, only the attitude of the UAV is adjusted to collect inspection data; otherwise, the UAV is corrected based on the basic inspection route.

[0013] In one possible implementation, the inspection baseline model is updated using environmental data collected in real time by drones, including: The system collects environmental data within a preset range using drones, and identifies dynamic obstacles and easily missed areas based on this data. The easily missed areas are identified based on changes in terrain or lighting conditions. The inspection baseline model is updated based on dynamic obstacles and easily missed areas.

[0014] In one possible implementation, route replanning is performed based on the updated inspection baseline model, including: Based on the dynamic obstacles in the updated inspection benchmark model, a detour route is generated, and a blind spot filling route is generated for the inspection feature points in the easily missed areas. By smoothly connecting the bypass routes or blind spot filling routes with the basic inspection routes, the replanned basic inspection routes are obtained.

[0015] One possible implementation involves smoothly connecting the bypass route or gap-filling route with the basic inspection route, including: Choose the target route from either the detour route or the gap-filling route; if the detour route and the gap-filling route are executed sequentially, the latter one shall be used as the target route. A smooth regression path is planned for the target route, so that after the UAV completes the inspection of the target route, it can transition to the basic inspection route along the smooth regression path.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention first constructs a high-precision 3D map along the target railway line, marking feature points, feature areas, and static obstacles on the 3D map to form an inspection benchmark model containing a feature point database. During the inspection process, in strong signal areas, RTK positioning combined with the feature point database is used to calibrate deviations, while in weak signal areas, the UAV position is determined by matching the feature point database with a visual sensor. The fixed coordinates of the feature points are used to calibrate the local real-time map, and a positioning transition section is used to smoothly switch between the two positioning methods. At the same time, IMU and odometer are used to ensure attitude stability. This invention provides an accurate global spatial framework by constructing an inspection benchmark model with multi-source data. The combination of differentiated positioning modes for strong and weak signal areas and the positioning transition section switching mechanism effectively avoids positioning drift in weak signal environments, significantly improves the positioning accuracy along railway lines with complex terrain, ensures that the UAV always aligns with the inspection feature points at a preset angle, guarantees the accuracy and integrity of the inspection data, and overcomes the shortcomings of existing technologies that rely on a single sensor and have poor positioning stability.

[0017] 2. This invention collects environmental data within a preset range using a drone, identifies dynamic obstacles and easily missed areas caused by changes in terrain and lighting, and updates the inspection baseline model based on this data. Subsequently, based on the updated model, it generates detour routes for dynamic obstacles and fill-in-the-blank routes for inspection feature points within easily missed areas. A smooth regression path connects the detour routes or fill-in-the-blank routes with the basic inspection route, completing route replanning. This invention achieves dynamic updating of the inspection baseline model, enabling the model to adapt to environmental changes during the inspection process in real time, breaking the limitations of traditional fixed models. The generation of detour routes effectively avoids dynamic obstacles, ensuring the safety of drone inspections. The fill-in-the-blank routes, centered on inspection feature points, ensure comprehensive coverage of easily missed areas through multi-view design, avoiding missed inspections. The smooth connection between the replanned routes and the basic inspection routes ensures the rationality of the routes and maintains the continuity of inspection data collection, achieving a complete closed loop of global planning, local adaptation, and regression connection, overcoming the shortcomings of existing local obstacle avoidance methods such as incomplete coverage and discontinuous data. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the method steps for formulating and adaptively adjusting railway inspection routes in Embodiment 1 of the present invention; Figure 2This is a schematic diagram of the method steps for switching positioning modes in Embodiment 1 of the present invention; Figure 3 This is a schematic diagram of the route replanning method steps in Embodiment 2 of the present invention. Detailed Implementation

[0020] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. 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.

[0021] Please see Figures 1-2 The first aspect of the present invention provides a method for railway inspection route planning and adaptive adjustment based on unmanned aerial vehicles (UAVs), comprising: constructing an inspection benchmark model of the target railway based on basic modeling data; planning a basic inspection route of the UAV based on the inspection benchmark model; switching the positioning mode based on the real-time position of the UAV during the inspection process, so that the UAV inspects the target railway along the basic inspection route; updating the inspection benchmark model through environmental data collected by the UAV in real time; replanning the route based on the updated inspection benchmark model; and completing the inspection work according to the replanned basic inspection route.

[0022] In a preferred embodiment, constructing an inspection benchmark model of the target railway based on basic modeling data includes: acquiring basic modeling data of the target railway; constructing a three-dimensional map along the target railway based on the basic modeling data; and generating an inspection benchmark model by performing feature marking on the high-precision three-dimensional map.

[0023] Before inspecting the target railway line, a high-precision 3D map of the target railway is constructed based on GIS data, DEM topographic data, and lidar oblique photography data. Based on this high-precision 3D map, feature points or feature areas are marked to form an inspection baseline model.

[0024] Feature points mainly include inspection feature points, which are key locations that need to be inspected under the requirements of the inspection task of the target railway. Inspection feature points mainly include core parts that are prone to defects, such as catenary locators, bridge bearings, tunnel entrance lining joints, roadbed slope inflection points, and protective net interfaces. An inspection point database is constructed based on the inspection feature points and feature parameters to serve as a benchmark for subsequent positioning calibration and route planning. The feature parameters of inspection feature points mainly include unique ID, coordinate data, inspection priority, and suitable shooting angle.

[0025] Featured areas mainly refer to areas with weak signals, which can be identified through previous survey data or historical inspection records. When the GPS signal strength in an area is less than a preset signal strength threshold (such as -120dBm), it is marked as a weak signal area. For example, areas such as 50 meters outside the tunnel entrance, deep valleys, and dense forest areas.

[0026] In addition to marking feature points and feature areas on high-precision 3D maps, it is also necessary to mark the 3D outlines and safety boundaries of static obstacles along the target railway line. Static obstacles mainly include mountains, fixed towers, and permanent no-fly zones (such as substations, stations, etc.). Marking static obstacles is an important constraint for planning basic inspection routes and overall replanning.

[0027] After generating an inspection baseline model based on a high-precision 3D map, a basic inspection route for the UAV is planned according to the inspection baseline model. If the inspection baseline model is accurately constructed, controlling the UAV to inspect the target railway along the basic inspection route can cover all inspection feature points, while ensuring the shortest possible inspection path.

[0028] In one example, assume that the target railway is a mountain railway, which includes several bridges and tunnels. The terrain along the line is undulating, with many deep valleys and dense forests. The weak GPS environment occurs frequently, and the catenary and bridge supports are the high-risk areas for defects.

[0029] First, collect GIS data of the target railway, including track centerline, mileage markers, bridge / tunnel design drawings and coordinates; collect DEM topographic data of the target railway, with a resolution of 0.5 meters, covering a range of 50 meters on both sides of the line; collect LiDAR oblique photography data of the target railway, covering a range of 80 meters on both sides of the line, including blind spots such as the bottom of bridges and tunnel entrances.

[0030] By fusing three types of data using Context Capture software, the lidar point cloud data is first denoised, then aligned with DEM terrain data and GIS coordinates to construct a high-precision 3D map. Modeling accuracy control: ground point error ≤3cm, fine structure error such as bridges and catenary ≤2cm, overall map resolution 0.1m, clearly showing minute structures such as catenary locators and bridge supports.

[0031] Based on a high-precision 3D map, professional annotation tools, such as the CloudCompare+ custom annotation plugin, are used to mark feature points, feature regions, and static obstacles, forming a complete inspection benchmark model.

[0032] When marking feature points (mainly inspection feature points), in conjunction with the key inspection requirements of the target railway inspection task, such as contact wire wear, bridge bearing cracks, and potential landslide hazards on the roadbed slope, inspection feature points were marked point by point on a high-precision 3D map, totaling 286 points. Specifically, they are distributed as follows: 124 contact wire locators, evenly distributed along the line, one every 50 meters; 32 bridge bearings, 16 on each of the two bridges, corresponding to the bearings on the top of the piers; 16 tunnel lining joints, 8 at each tunnel entrance and exit, distributed in a ring along the tunnel entrance; 98 roadbed slope turning points, with denser markings in deep valley sections and areas with slopes >30°; and 16 protective netting interfaces, at the junction of forest areas and the roadbed. Each feature point was assigned a unique ID, and after entering coordinate data, inspection priority, and suitable shooting angles, an inspection point database was constructed. The feature point database was bound to the high-precision 3D map, allowing for the retrieval of corresponding parameters at any time, providing benchmark data for subsequent positioning calibration and route planning.

[0033] When marking characteristic areas, a preset GPS signal strength threshold of -120dBm is used; areas below this value are marked as weak signal areas. Combining prior field survey data (using a handheld GPS signal tester to detect signals segment by segment) and records from the last three historical inspections, weak signal areas are marked on a high-precision 3D map. For example, a 300-meter-long valley section with a signal strength of -128dBm to -130dBm is marked as a weak signal area. After marking the characteristic areas, a matching positioning method is established to facilitate subsequent matching.

[0034] When marking static obstacles, accurately mark the 3D outline and safety boundary of the static obstacles in the 3D map as a hard constraint for flight route planning. Specific marking content: Mountains: 3D outline of mountains on both sides of the route, with a safety boundary set at "≥8 meters from the edge of the mountain to the center line of the route" to avoid drone collisions; Fixed towers: 46 contact network towers and communication towers in total, mark the 3D outline of each tower, with a safety boundary ≥5 meters; Permanent no-fly zone: 1 substation (approximately 200㎡) and 1 small station (approximately 800㎡) within the section, mark the area outline, with a safety boundary ≥10 meters, prohibiting drones from entering this area.

[0035] After constructing the inspection baseline model, basic inspection routes are automatically generated using route planning software (such as the fusion improved A* algorithm). The planning principle requires covering all inspection feature points while avoiding all static obstacles and minimizing the path length.

[0036] In a preferred embodiment, the positioning method is switched based on the real-time location of the UAV, including: determining whether the UAV is in a positioning transition segment based on the UAV's real-time location; if yes, switching the positioning method based on the relative position of the positioning transition segment and the weak signal area; wherein the positioning method includes a visual positioning scheme for the weak signal area and an RTK positioning scheme for the strong signal area; if no, keeping the positioning method unchanged.

[0037] In a preferred embodiment, setting a positioning transition section includes: identifying weak signal areas in the inspection baseline model; setting positioning transition sections on both sides of the weak signal area based on the entry and exit critical points of the weak signal area and according to a preset route length; wherein the preset route length is 5 meters.

[0038] Based on the inspection benchmark model, a positioning transition section is set for weak signal areas. For example, a transition section with a fixed flight path length can be set based on the entry and exit point of the weak signal area. The fixed flight path length can be set to 5 meters. Using the nearest inspection feature point as the positioning benchmark, the positioning mode is switched in the positioning transition section to avoid flight path deviation and missed inspection feature points caused by sudden positioning changes.

[0039] It should be noted that the positioning transition section is set after the basic inspection route is planned. The pre-marked weak signal area is identified from the inspection benchmark model. Based on the entry and exit critical point of the weak signal area, two positioning transition sections are obtained in the opposite direction of the weak signal area according to the preset route length.

[0040] During the inspection process, the real-time location of the drone is acquired. Based on this location, it is determined whether the drone is in a positioning transition zone. If the drone is in a positioning transition zone, the positioning mode needs to be switched. Each weak signal area corresponds to a positioning transition zone at both ends, and this positioning transition zone is actually a strong signal area. Therefore, switching the positioning mode in this positioning transition zone will not affect the positioning accuracy in the weak signal area.

[0041] During the specific switching process, the switching method needs to be determined based on the relative position of the positioning transition section and the weak signal area. When the UAV passes through the positioning transition section and enters the weak signal area, it needs to switch to the visual positioning scheme corresponding to the weak signal area; when the UAV passes through the positioning transition section and enters the strong signal area, it needs to switch to the RTK positioning scheme corresponding to the strong signal area.

[0042] It should be noted that when a drone is in a positioning transition zone, it may not necessarily switch to the current positioning method; it may maintain the existing positioning method. For example, if the positioning transition zones set up in adjacent weak signal areas overlap, when the drone passes through the overlapping positioning transition zone, it will enter the next weak signal area. Since the drone has completed the inspection of the previous weak signal area using a visual positioning solution, it does not need to switch positioning methods in the overlapping positioning transition zone and can continue to use the visual positioning solution.

[0043] In one example, suppose a deep valley section of the target railway is a weak signal area. Mark the entry and exit points of the basic inspection route and this weak signal area. When the UAV performs its inspection task along the basic inspection route, if it flies past the entry point, it enters the weak signal area; if it passes the exit point, it flies out of the weak signal area.

[0044] Using the entry point as a reference, locate a route point of fixed length along the opposite direction of the inspection direction. The route segment between this route point and the entry point is the positioning transition segment. During this positioning transition segment, the UAV should switch to the positioning mode corresponding to the weak signal area. Similarly, using the exit point as a reference, locate a route point of fixed length along the inspection direction. The route segment between this route point and the exit point is the positioning transition segment. During this positioning transition segment, the UAV should switch to the positioning mode corresponding to the next area.

[0045] In a preferred embodiment, the inspection of weak signal areas based on a visual positioning scheme includes: determining the location of the UAV through a feature point database in a visual sensor and an inspection benchmark model; wherein the feature point database includes feature points and their corresponding coordinate data; calibrating a local real-time map based on the feature point database; and adjusting the UAV using the calibrated local real-time map to ensure that the UAV completes the inspection of the feature points.

[0046] In areas with strong signals (GPS signal strength ≥ -120dBm), RTK positioning is the core method, with IMU-assisted image stabilization. At the same time, the positioning deviation is calibrated using the inspection point database in the inspection benchmark model to ensure that the actual inspection route is consistent with the preset basic inspection route, laying the foundation for subsequent inspection and positioning in areas with weak signals.

[0047] In areas with weak signals (GPS signal strength < -120dBm), visual SLAM, IMU, and odometry are used for fusion positioning, with a focus on leveraging the inspection point database in the inspection benchmark model to enhance positioning accuracy. The specific workflow is as follows: The visual sensor prioritizes capturing inspection feature points on the target railway and matches them with the feature point database in the inspection benchmark model. The current position of the UAV is calculated based on the coordinate data of the matched inspection feature points. The SLAM algorithm, based on the high-precision 3D map in the inspection benchmark model, combines the coordinate data of the inspection feature points to calibrate the local real-time map, correcting the instantaneous deviation of visual positioning and ensuring that the UAV always maintains a precise inspection attitude around the key inspection points. The odometry supplements the position increment under high-speed movement, and the IMU ensures attitude stability, avoiding positioning drift that leads to missed key points.

[0048] The inspection process in weak signal areas can be summarized as follows: the drone's position is determined by visual sensors and an inspection point database; then, the local real-time map is calibrated using the fixed coordinates of the inspection feature points; finally, the drone is adjusted using the local real-time map to ensure that the drone completes the inspection operation around the inspection feature points.

[0049] The high-precision 3D map in the inspection benchmark model is a global framework used to provide the absolute coordinates of inspection feature points and the flight path benchmark. However, it cannot reflect environmental changes in real time during flight, and the real-time matching delay between the UAV and the high-precision 3D map is relatively high during high-speed movement. Therefore, a local real-time map is constructed using the SLAM algorithm.

[0050] It is worth noting that when the SLAM algorithm constructs a local real-time map in real time, if it relies solely on the UAV's own motion trajectory and visual features for estimation, the following problems exist: 1) Under high-speed motion, the inter-frame matching of visual sensors is prone to slight deviations, which, if accumulated over a long period of time, will cause the local real-time map to deviate from the pre-constructed high-precision 3D map, i.e., map drift occurs; 2) In areas with weak signals, the lighting is complex and the texture is simple, making visual feature extraction unstable, which will further aggravate the deviation of the local real-time map.

[0051] Due to the influence of local real-time map deviations, even if the current position of the UAV is calculated by matching inspection feature points, the subsequent attitude adjustment and path planning based on the local real-time map will also be biased, making it impossible to guarantee that the UAV will complete the inspection work of the inspection feature points.

[0052] Since the coordinate data of the inspection feature points is a globally fixed benchmark, calibrating the local real-time map with it can correct map drift, pulling the drifted local real-time map back into the pre-constructed high-precision 3D map framework. This ensures that the relative positions of the inspection feature points in the local real-time map are completely consistent with the inspection benchmark model, eliminating accumulated errors. It can also strengthen attitude constraints; the calibrated local real-time map provides the UAV with a more accurate spatial reference, not only determining its position but also locking its attitude (such as pitch, roll, and lateral angles), ensuring that the UAV always completes the inspection at the preset optimal angle, avoiding missed captures of detailed features of the inspection feature points due to attitude deviations.

[0053] In a preferred embodiment, the feature points in the feature point database also include fixed feature points and their coordinate data; the fixed feature points are preferentially selected from the uniformly distributed basic structures on the target railway; wherein, the basic structures include catenary poles, track edges, mileage markers and tunnel entrances.

[0054] As mentioned earlier, inspection feature points refer to the key locations that need to be inspected under the requirements of the inspection task of the target railway. Before the inspection, it is necessary to mark the inspection feature points in order to accurately complete the inspection task. Since the coordinate data of the inspection feature points is global coordinates, it can be used for UAV positioning or local real-time map calibration, thus avoiding the need to set other data for positioning or calibration.

[0055] However, the distribution of inspection feature points on the target railway is not uniform; some areas have a dense distribution of inspection feature points, while others have a relatively sparse distribution. In the actual environment, there may be obstacles that affect the inspection feature points. If enough inspection feature points cannot be effectively extracted, it will affect the accuracy of UAV positioning and local real-time maps, thus making it impossible to complete the inspection task of the target railway.

[0056] To address this issue, a fixed-point database is used. If the extracted inspection feature points are insufficient for drone positioning or local real-time map calibration, fixed feature points on the target railway are identified, and their coordinates are used for positioning or calibration. Fixed feature points refer to inherent, iconic structures on the target railway, such as overhead contact line poles, track edges, mileage markers, and tunnel entrances. A fixed-point database is constructed based on these feature points and their corresponding coordinate data. This database has a more uniform distribution than inspection feature points, resulting in a higher probability and accuracy of fixed-point extraction. Furthermore, fixed feature points play a supporting role; it's unnecessary to mark all fixed structures on the target railway as fixed feature points. Prioritizing uniformly distributed fixed structures ensures the implementation of auxiliary functions while reducing the data processing workload of building the fixed-point database.

[0057] In a preferred embodiment, the real-time inspection route is corrected based on the route deviation during the inspection process, including: calculating the inspection deviation between the actual inspection route of the UAV and the basic inspection route in real time; wherein the inspection deviation includes route deviation or relative deviation of feature points; when the inspection deviation does not exceed the preset inspection deviation threshold, the attitude of the UAV is adjusted only to collect inspection data; otherwise, the UAV is corrected based on the basic inspection route.

[0058] Different flight path deviation thresholds are set for different inspection targets along the target railway line, such as ±5cm for catenary inspection and ±10cm for roadbed inspection. Simultaneously, deviation verification is performed using inspection feature points in the inspection baseline model as core anchor points. If the flight path deviation does not exceed the preset threshold, minor attitude adjustments are made based on the IMU to maintain flight path stability and ensure the UAV always maintains an angle aligned with the inspection feature points. If the flight path deviation exceeds the preset threshold, priority is given to ensuring the corrected flight path covers the inspection feature points, guiding the UAV to smoothly return to the basic inspection flight path.

[0059] It should be noted that the flight path deviation should not only take into account the deviation of the real-time inspection flight path from the preset basic inspection flight path, but also the deviation between the UAV and the inspection feature points. If any deviation exceeds the preset flight path deviation threshold, the flight path needs to be corrected.

[0060] In one example, taking the overhead contact line as an example, when the UAV inspects the contact line locator, the linear deviation between the actual position and the basic inspection route is 3cm (threshold is 5cm), the three-dimensional relative distance deviation between the UAV and the basic contact line locator is 4cm (threshold is 5cm), and the shooting angle deviation is 2° (threshold is 5°). At this time, the route deviation and the relative deviation of the feature points do not exceed the preset threshold, so there is no need to correct the route. The attitude of the UAV can be adjusted by the IMU.

[0061] Taking a bridge section as an example, when a drone inspects a bridge bearing, the linear deviation between the actual position and the basic inspection route is 4cm (threshold is 5cm), the relative distance deviation from the bridge bearing is 7cm (threshold is 5cm), and the angle deviation is 6° (threshold is 5°). At this time, the route deviation does not exceed the preset threshold, but the relative deviation of the feature points (referring only to the inspected feature points, not including fixed feature points) exceeds the preset threshold, and route correction is required. A correction trajectory can be generated based on the fixed coordinates of the bridge bearing, and a smooth correction path can be planned through an interpolation algorithm to guide the drone to gradually return to the preset position with a relative distance deviation of 4cm from the bearing and a pitch angle of -15°.

[0062] Example

[0063] Please see Figure 3 In a preferred embodiment, updating the inspection benchmark model using environmental data collected in real time by a drone includes: collecting environmental data within a preset range using a drone, identifying dynamic obstacles and easily missed areas based on the environmental data; wherein, easily missed areas are identified based on terrain changes or light changes; and updating the inspection benchmark model based on dynamic obstacles and easily missed areas.

[0064] Based on the aforementioned inspection baseline model, the UAV inspection process requires real-time collection of environmental data for global replanning. During the UAV inspection, environmental data within a preset range is collected using its onboard LiDAR and visual sensors. This environmental data mainly includes dynamic obstacles and environmental change data. Dynamic obstacles include tree growth, construction machinery, and fallen vegetation. Environmental change data includes environmental changes such as the shadow area under bridges and the undulations of canyon terrain. Environmental change data is mainly used to identify areas prone to being missed due to environmental changes. Inspection feature points in these areas cannot guarantee obtaining qualified inspection data by following the original inspection baseline route or shooting angle due to environmental changes or terrain undulations. The inspection baseline model is updated based on the real-time captured dynamic obstacle and environmental change data. Deviation areas before and after the update can also be marked, providing data support for UAV detours.

[0065] In one example, taking a dense forest area as an example, the drone uses lidar and visual sensors to collect environmental data within a 50-meter radius to identify dynamic obstacles such as tree growth, construction machinery, and fallen vegetation. If construction machinery is identified as a dynamic obstacle but is not shown in the inspection baseline model, the inspection baseline model is updated based on the 3D outline of the construction machinery and the safety boundary.

[0066] The inspection baseline model can be updated at the ground control center. The UAV will upload the dynamic obstacles it identifies to the ground control center and compare them with the data of that area in the inspection baseline model. If they are different, the inspection baseline model will be updated and the updated model data will be fed back to the UAV.

[0067] In a preferred embodiment, route replanning based on the updated inspection baseline model includes: generating a detour route based on dynamic obstacles in the updated inspection baseline model, generating a blind spot filling route for inspection feature points in easily missed areas, and smoothly connecting the detour route or blind spot filling route with the basic inspection route to obtain the replanned basic inspection route.

[0068] In a preferred embodiment, the bypass route or blind spot filling route is smoothly connected to the basic inspection route, including: selecting a target route from the bypass route or blind spot filling route; wherein, if the bypass route and the blind spot filling route are executed sequentially, the one executed later is used as the target route; and planning a smooth return path for the target route so that after the UAV completes the inspection of the target route, it transitions to the basic inspection route along the smooth return path.

[0069] Flight path replanning aims to achieve comprehensive coverage of inspection feature points. When obstacles affect the basic inspection flight path, the drone cannot perform inspections according to the preset basic inspection flight path, resulting in the inability to collect inspection data for some inspection feature points; when obstacles affect inspection feature points, even if the drone inspects along the basic inspection flight path, some inspection feature points will still be missed.

[0070] Based on the updated inspection benchmark model and combined with railway inspection safety constraints (such as minimum distance from railway track ≥ 5 meters, no-fly zone boundary ≥ 10 meters, etc.), the global optimal route replanning is achieved.

[0071] Global optimal route replanning includes the following steps: 1. Generate detour routes: When an obstacle is detected, based on the static and dynamic obstacles and the distribution of inspection feature points in the inspection baseline model, the influence range of the obstacle is quickly calculated, and multiple candidate detour routes are generated. Based on the principle of "priority of inspection feature point coverage > shortest path > smooth connection", the scheme that can completely cover the inspection feature points within the influence range of the obstacle and does not deviate from the core inspection area is selected first. At the same time, the detour trajectory and shooting angle are adjusted for the inspection feature points to ensure that the inspection of key points is not affected by the detour and to avoid the omission of inspection feature points due to local detour.

[0072] 2. Generate blind spot filling routes: For easily missed inspection areas such as the bottom of bridges and shadowed areas in canyons, combined with real-time perceived light and terrain data, the system prioritizes inspection feature points in these areas as blind spot filling points and automatically generates spiral, diving, or circular blind spot filling routes. The route parameters (such as altitude, speed, and shooting angle) are optimized based on the coordinate data and structural dimensions of the inspection feature points to ensure that each inspection feature point can obtain inspection data from at least two different perspectives. At the same time, the system avoids structures such as bridges and mountains to avoid collision risks and achieves dual coverage of inspection feature points and easily missed inspection areas.

[0073] 3. Route re-planning and connection: After the detour and supplementary inspection are completed, a smooth re-planning path is planned based on the pre-built basic inspection route. By adjusting the flight attitude and flight speed, the re-planned route and the basic inspection route are connected to ensure the continuity of data collection. At the same time, the re-planned trajectory and obstacle information are fed back to the ground station to update the inspection route record.

[0074] It is worth noting that the bypass flight path completes the feature inspection while avoiding the influence of obstacles. In other words, the bypass flight path can collect inspection data for all inspection feature points that are not affected by obstacles. If the inspection feature points are affected by obstacles, blind spot detection needs to be performed on the affected inspection feature points, that is, a blind spot detection path needs to be generated to ensure that inspection data from multiple perspectives can be obtained for areas that are easily missed, thus ensuring the quality of inspection.

[0075] It should be understood that the descriptions of technical features, technical solutions, beneficial effects, or similar language in this application do not imply that all features and advantages can be achieved in any single embodiment. Rather, it is understood that the description of a feature or beneficial effect means that a specific technical feature, technical solution, or beneficial effect is included in at least one embodiment. Therefore, the descriptions of technical features, technical solutions, or beneficial effects in this specification do not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions, and beneficial effects described in this embodiment can be combined in any suitable manner. Those skilled in the art will understand that embodiments can be implemented without one or more specific technical features, technical solutions, or beneficial effects of a particular embodiment. In other embodiments, additional technical features and beneficial effects may be identified in specific embodiments that do not embody all embodiments.

[0076] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any other combination thereof. When implemented using a software program, it can be implemented entirely or partially in the form of a computer program product. This computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device containing one or more servers, data centers, etc., that can be integrated with the medium. The available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state disks (SSDs)).

[0077] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A method for railway line inspection route planning and adaptive adjustment based on unmanned aerial vehicles (UAVs), characterized in that, include: A benchmark model for the inspection of the target railway is constructed based on the basic modeling data, and a basic inspection route for the UAV is planned according to the benchmark model; wherein, the basic modeling data is obtained through database or survey. During the inspection process, the positioning mode is switched based on the real-time location of the UAV, enabling the UAV to inspect the target railway along the basic inspection route; wherein, the switching of the positioning mode is used to improve positioning accuracy. The inspection baseline model is updated by collecting environmental data in real time using drones; the route is replanned based on the updated inspection baseline model, and the inspection work is completed according to the replanned basic inspection route.

2. The method for railway line inspection route planning and adaptive adjustment based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, Based on the basic modeling data, a benchmark model for the inspection of the target railway is constructed, including: Acquire basic modeling data of the target railway, and construct a three-dimensional map along the target railway based on the basic modeling data; wherein, the basic modeling data includes GIS data, DEM terrain data and oblique photogrammetry data; Feature markings are performed on the three-dimensional map to generate an inspection baseline model; the feature markings include feature points, feature areas and static obstacles, and the feature points include inspection feature points.

3. The method for railway line inspection route planning and adaptive adjustment based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, Real-time location switching positioning methods based on drones include: The drone's real-time location is used to determine whether it is in a positioning transition phase; the positioning transition phase is set according to the inspection benchmark model. Yes, the positioning method is switched according to the relative position of the positioning transition segment and the weak signal area; the positioning method includes a visual positioning scheme for the weak signal area and an RTK positioning scheme for the strong signal area. No, keep the positioning method unchanged.

4. The method for railway line inspection route planning and adaptive adjustment based on unmanned aerial vehicles (UAVs) according to claim 3, characterized in that, Setting the positioning transition segment includes: Identify weak signal regions in the inspection benchmark model; Based on the entry and exit critical points of the weak signal area, positioning transition sections are set on both sides of the weak signal area according to the preset route length; wherein, the preset route length is 5 meters.

5. The method for railway line inspection route planning and adaptive adjustment based on unmanned aerial vehicles (UAVs) according to claim 3, characterized in that, Inspection of weak signal areas based on visual positioning schemes includes: The location of the UAV is determined by a visual sensor and a feature point database in the inspection benchmark model; wherein, the feature point database includes feature points and their corresponding coordinate data; The local real-time map is calibrated based on the feature point database, and the UAV is adjusted using the calibrated local real-time map to ensure that the UAV completes the inspection of the feature points.

6. The method for railway line inspection route planning and adaptive adjustment based on unmanned aerial vehicles (UAVs) according to claim 5, characterized in that, The feature point database also includes fixed feature points and their coordinate data; The fixed feature points are preferentially selected from the uniformly distributed basic structures on the target railway; wherein, the basic structures include contact wire poles, track edges, mileage markers and tunnel entrances.

7. The method for railway line inspection route planning and adaptive adjustment based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, During the inspection process, the real-time inspection route is corrected based on route deviations, including: The inspection deviation between the actual inspection route of the UAV and the basic inspection route is calculated in real time; wherein, the inspection deviation includes route deviation or relative deviation of feature points; If the inspection deviation does not exceed the preset inspection deviation threshold, the attitude of the UAV is adjusted only to collect inspection data; otherwise, the UAV is corrected based on the basic inspection route.

8. The method for railway line inspection route planning and adaptive adjustment based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The inspection baseline model is updated using environmental data collected in real time by drones, including: The drone collects environmental data within a preset range, and identifies dynamic obstacles and easily missed areas based on the environmental data; wherein, easily missed areas are identified based on changes in terrain or lighting. The inspection baseline model is updated based on the dynamic obstacles and the areas prone to being missed.

9. The method for railway line inspection route planning and adaptive adjustment based on unmanned aerial vehicles (UAVs) according to claim 8, characterized in that, Route replanning based on the updated inspection baseline model includes: Based on the dynamic obstacles in the updated inspection benchmark model, a detour route is generated, and a blind spot filling route is generated for the inspection feature points in the easily missed inspection area. The bypass route or the blind spot filling route is smoothly connected with the basic inspection route to obtain the replanned basic inspection route.

10. The method for railway line inspection route planning and adaptive adjustment based on unmanned aerial vehicles (UAVs) according to claim 9, characterized in that, Smoothly connecting the bypass route or the gap-filling route with the basic inspection route includes: The target route is selected from the detour route or the gap-filling route; wherein, if the detour route and the gap-filling route are executed sequentially, the latter route is selected as the target route. A smooth regression path is planned for the target route so that after the UAV completes the inspection of the target route, it transitions to the basic inspection route along the smooth regression path.

Images