A high-speed rail continuous positioning system and method based on feature targeting

Abstract

This invention discloses a continuous positioning system and method for high-speed rail based on feature targeting, belonging to the field of rail transit positioning technology. The system completely abandons ground transponders and satellite navigation, using only the inherent geometric features of the track itself as the sole calibration benchmark. It achieves continuous sub-meter-level positioning throughout the entire journey through multi-sensor adaptive sensing, real-time feature matching, filtering fusion, and dynamic error compensation. The sensing module is adaptable to harsh environments such as rain, snow, fog, tunnels, icing, and oil contamination; the matching module includes temperature deformation correction and anchor point fallback mechanisms; the control module can identify and compensate for slippage and attitude deviation; and the closed-loop module has emergency positioning and automatic speed limiting safety strategies. This invention significantly reduces the construction and maintenance costs of railway lines, provides continuous and stable positioning, strong anti-interference capabilities, and sufficient safety redundancy, making it suitable for various high-speed rail and urban rail vehicles and possessing strong engineering practicality.

Description

Technical Field

[0001] This invention relates to the field of rail transit train positioning technology, specifically to a high-speed rail continuous positioning system and method based on feature targeting. It is applicable to high-speed EMU trains, intercity trains, and urban rail transit. It uses the inherent geometric features of the rails, sleepers, track slabs, and ballast bed as the sole basis for positioning calibration, and does not rely on ground transponders or satellite navigation signals. It achieves continuous sub-meter-level positioning in scenarios such as high-speed operation on the main line, tunnel passage, severe weather, station alignment, and operation on old lines. Background Technology

[0002] Existing high-speed rail positioning systems generally adopt a combination of "ground transponder point calibration + inertial navigation + odometer", which has inherent defects: Transponders are densely deployed, expensive, and difficult to maintain; failure can easily cause the area positioning to fail. Point-based positioning leads to the continuous accumulation of interval errors, and wheel slippage, wheel wear will significantly reduce positioning accuracy. Satellite navigation is unusable in tunnels, mountainous areas, and strong electromagnetic environments, and there is a lack of reliable autonomous positioning methods. Blizzards, ice, oil stains, and dust can obscure track features, making traditional visual matching prone to inaccuracies. Ballasted track bed is easily disturbed, and old lines have irregular geometry, making existing solutions poorly adaptable.

[0003] Currently, relevant visual / laser orbit positioning technologies still rely on transponders or satellite navigation calibration, and have not formed a complete closed-loop positioning system with the inherent geometric characteristics of the orbit itself as the sole calibration benchmark. Therefore, they cannot achieve continuous and reliable positioning without the need for external facilities throughout the entire process. Summary of the Invention

[0004] (a) Purpose of the invention To address the problems of existing high-speed rail positioning systems, such as reliance on ground transponders, discontinuous positioning, easy error divergence, and weak adaptability to harsh environments, this invention provides a feature-targeted continuous positioning system and method for high-speed rail. Using the inherent geometric features of the track itself as the sole calibration benchmark, and through multi-sensor adaptive sensing, real-time feature matching, dynamic error compensation, and multi-level safety redundancy, it achieves continuous, high-precision autonomous positioning throughout the entire journey without relying on ground transponders or satellite navigation signals, thereby improving system safety, economy, and environmental robustness.

[0005] (II) Technical Solution 1. System Composition A high-speed rail continuous positioning system based on feature targeting completely excludes ground transponder signal receiving modules and satellite navigation signal receiving modules, relying solely on the inherent geometric characteristics of the rails, sleepers, track slabs, and ballast bed as the sole basis for positioning calibration. This includes sequentially electrically connected components: Vehicle-mounted track bed feature perception module, vehicle-mounted feature targeting matching module, vehicle-mounted positioning control module, and vehicle-mounted closed-loop verification module; It is linked with the train control system, inertial measurement unit, odometer, wheel speed sensor, on-board temperature sensor, and vibration sensor.

[0006] (1) Onboard track bed feature perception module It consists of a linear array camera, a lidar, an orbital geometry sensor, a 940nm infrared illumination module, and an environmental adaptive compensation unit. Linear scan camera: resolution ≥12K, sampling frequency ≥200Hz, dynamic range ≥120dB, operating temperature -40℃~+60℃; LiDAR: ranging accuracy ≤0.05m, scanning frequency ≥50Hz, point cloud density ≥1000 points / ㎡; Track geometry sensor: Track gauge measurement accuracy ±0.1mm, track alignment / height resolution 0.01mm, built-in temperature drift compensation; Infrared illumination: 940nm with no red exposure, adjustable power in stages, suitable for tunnels / completely dark nighttime scenes; Electromagnetic interference immunity complies with EN 50121-3-2 standard.

[0007] Multi-modal data fusion rules: Normal sunny day conditions: Linear scan camera weight 0.5, LiDAR weight 0.5; Rain, snow, fog / water accumulation / dust storm: LiDAR weight 0.7, orbital geometry sensor weight 0.3, enable point cloud denoising and feature enhancement; Tunnel / Nighttime: LiDAR + Track Geometry Sensor + 940nm Infrared Illumination; Icing / oil cover: Prioritize using track geometry sequence (gauge, orientation, elevation) as matching features; Train passing / braking vibration: Enable short-time feature smoothing filter to remove abnormal point clouds.

[0008] The scope of feature extraction is strictly limited: Only the inherent geometric features of the orbital body are extracted, including: Ballastless track: the sequence of track slab joint spacing, the arrangement of fasteners, the outline of the track slab, and the continuous variation characteristics of the track gauge; Ballasted track: sleeper spacing sequence, rail inner profile, and track bed slope stability geometry; Exclude non-track features such as trackside markings, signals, overhead contact lines, tunnel walls, and ground coatings.

[0009] (2) Vehicle-mounted feature-targeted matching module It has a built-in offline high-precision track bed standard feature library with a storage capacity of ≤512MB. It is pre-deployed in sections according to the line and supports incremental updates through on-board storage media during operation windows. The feature library contains the geometric features-mileage mapping relationship of the entire track under the WGS-84 coordinate system.

[0010] Orbit deformation correction model: ΔL = α × ΔT × L in: ΔL: Offset of the orbital thermal expansion and contraction characteristics; α: Coefficient of linear expansion of rail, taken as 11.8e-6 / ℃; ΔT: Real-time temperature difference, collected by the vehicle-mounted temperature sensing unit; L: The reference length of the corresponding track section; The track settlement ΔS is pre-assigned using a feature library and does not rely on real-time onboard measurements.

[0011] Matching algorithm: Track geometry sequence matching + SIFT keypoint matching + ICP point cloud registration; The number of valid feature points in a regular section is ≥30; Tunnel / straight section: ≥40 characteristic points; Fixed calibration anchor points are set every 1 to 2 km along the entire line (rail sections corresponding to bridge bearings, fixed sleeper sets at tunnel entrances, and sections marked with special fasteners). When an anchor point fails or is obstructed, it automatically switches to segment geometric sequence weighted matching without interrupting the positioning process.

[0012] Matching delay ≤10ms, positioning accuracy: Mainline operation: ≤0.5m; Station alignment: ≤0.3m; Tunnel passage: ≤0.8m.

[0013] (3) Vehicle positioning control module Localization is achieved by fusing an extended Kalman filter (EKF) with a particle filter (PF). EKF process noise covariance Q=diag ([1e-7,1e-7,1e-7]); The observation noise covariance R = diag([1e-4, 1e-4]); Particle count ≥ 200, filter update cycle ≤ 20ms; The cumulative error calibration cycle for inertial / odometers is ≤50ms.

[0014] Idle / Slippage Recognition and Compensation: The system determines wheel spin / slippage by comparing sudden changes in wheel speed and acceleration with instantaneous relative ground velocity measured by lidar. Upon identification, the odometer integration is immediately cut off, and the mileage calculation is directly reset by using feature matching location.

[0015] Attitude compensation: The perception angle offset is corrected in real time based on the train's pitch and roll angles to eliminate the impact of slopes, curves, and swaying on the matching.

[0016] Location redundancy: If any sensor fails, the system automatically switches to the remaining sensor combination mode to maintain continuous positioning.

[0017] (4) Vehicle-mounted closed-loop verification module Grading accuracy threshold: Main line: Positioning error ≤ 0.5m; Station alignment: ≤0.3m; Tunnel: ≤0.8m.

[0018] Security trigger mechanism: If a single deviation exceeds twice the threshold, or if eight consecutive matching corrections fail to meet the standard, the emergency positioning mode will be activated. Emergency positioning: Inertial navigation dead reckoning + coarse matching of anchor point intervals, positioning error ≤1.5m, continuous stable operation ≥30min; In emergency mode, the system automatically outputs a downgrade signal to the train control system, triggering a speed limit of 300 km / h or below, prohibiting automatic driving from precisely aligning with the target, and prompting the driver to manually monitor the system.

[0019] 2. Methods and Steps A feature-targeted continuous positioning method for high-speed rail, which does not rely on ground transponders or satellite navigation signals, but only uses the inherent geometric features of the track itself as the sole calibration benchmark to achieve continuous positioning, includes: Initial reference calibration Before the train leaves the depot, initial calibration is completed by fixing the track slab seams or sleeper group anchor points at the ends of the station tracks. For stations without typical characteristics, a coarse positioning method using pre-set coordinates upon exiting the depot and inertial navigation is adopted, with a calibration accuracy of ≤0.1m.

[0020] Real-time track body feature acquisition During operation, the sensing mode automatically switches based on weather, lighting conditions, and tunnel scenarios. Only the inherent geometric features of the track slab, sleepers, rails, and ballast bed are collected. For areas with ice, oil stains, snow accumulation, and dust storms, a priority strategy for orbital geometry sequence is adopted to filter out interference.

[0021] Feature library correction and anchor point calibration Based on the measured track temperature ΔT on the vehicle, the feature matching benchmark is corrected in real time using the deformation correction model ΔL=α×ΔT×L; Every 1-2 km, a fixed calibration anchor point is identified and calibration error is forced. When the anchor point fails, continuous matching of the segment geometric sequence is automatically enabled as a fallback.

[0022] Feature-targeted precise matching The real-time acquired orbital geometry sequence and contour point cloud are registered with the standard feature library. Output mileage deviation, position deviation, and speed deviation, with a matching delay of ≤10ms.

[0023] Fusion positioning and error compensation EKF+PF fused output continuous position, Real-time compensation for inertial drift, wheel wear, slippage, and attitude deviation. Ensure the positioning does not diverge.

[0024] Closed-loop verification and security degradation Accuracy is verified at key points such as the main line, bridges, tunnels, and station entrances; If the deviation exceeds the limit, immediately perform a second rematch; Continuous abnormalities trigger emergency location mode and speed limit and driver alarm.

[0025] Full-process continuous positioning closed loop Continuously execute the process of "perception-matching-location-verification-correction". Achieve stable and continuous positioning without transponders or satellite navigation throughout the entire process.

[0026] (III) Beneficial Effects Completely independent of ground transponders and satellite navigation, significantly reducing the costs of line construction and full life-cycle operation and maintenance; It uses only the inherent geometric features of the orbit itself, has strong anti-interference, anti-occlusion, and anti-spoofing capabilities, and the features are stable and reliable; Continuous positioning throughout the entire process, real-time suppression of error divergence, and resolution of positioning drift caused by idling, slippage, and wheel wear; It is suitable for harsh environments such as tunnels, blizzards, ice, oil spills, sandstorms, and vibrations, and is compatible with both new and old lines, as well as ballastless and non-ballastless tracks. Multi-level redundancy and safety degradation mechanisms ensure a safety fallback mechanism in case of anchor point failure, and the emergency mode includes speed limit protection, meeting the safety integrity requirements of high-speed rail. Detailed Implementation

[0027] Example: High-speed rail positioning throughout its 350km / h journey without a transponder or satellite navigation. Route: Ballastless track passenger dedicated line, including a 12km long tunnel; Operating conditions: Transponder off, GNSS off; Weather: Light rain, track wet; Target: Main line error ≤ 0.5m, station alignment error ≤ 0.3m.

[0028] Initial calibration is completed at the track slab seam anchor points before leaving the warehouse, with an accuracy of 0.08m.

[0029] The system combines line-of-sight camera and laser sensing, with a matching latency of 8ms and a positioning error of 0.2–0.4m.

[0030] Upon entering the tunnel, the system automatically switches between laser, geometric sensor, and infrared supplementary lighting, with an error of ≤0.6m.

[0031] Every 2km, bridge anchor points are identified and mandatory calibration is performed; for sections without anchor points, sequence matching is used as a fallback.

[0032] The odometer is instantly reset when slight slippage is detected, and the positioning remains unchanged.

[0033] The alignment accuracy at the station is 0.18m, and the entire process is stable with no alarms.

Claims

1. A high-speed rail continuous positioning system based on feature targeting, characterized in that, Completely excluding ground transponder signal receiving modules and satellite navigation signal receiving modules, the positioning calibration is based solely on the inherent geometric characteristics of the rails, sleepers, track slabs, and track bed. This includes an onboard track bed feature sensing module, feature target matching module, positioning control module, and closed-loop verification module, which are linked with the train control system, inertial measurement unit, odometer, wheel speed sensor, onboard temperature sensing unit, and vibration sensor. The track bed feature perception module consists of a line array camera, a lidar, a track geometry sensor, a 940nm infrared supplementary lighting module, and an environmental adaptive compensation unit. It achieves multi-sensor fusion according to preset weights, extracts only the geometric features of the track body, and excludes non-track features such as trackside markings, signals, catenary, and tunnel walls. The feature-targeted matching module has a built-in offline track bed standard feature library and uses the track deformation correction model ΔL=α×ΔT×L for thermal expansion compensation. Fixed calibration anchor points are set every 1 to 2 km along the entire line. When the anchor points fail, the section geometric sequence matching is automatically switched. The matching delay is ≤10ms and the main line positioning error is ≤0.5m. The positioning control module adopts an extended Kalman filter + particle filter fusion, which identifies wheel slippage by comparing wheel speed with laser relative speed and instantly resets the odometer. It also has a train attitude deviation compensation function. The closed-loop verification module is set with graded accuracy thresholds. When an anomaly occurs, an emergency positioning mode is activated, which continues to work for ≥30 minutes and automatically outputs a train speed limit signal.

2. A continuous positioning method for high-speed trains based on feature targeting, characterized in that, Continuous positioning is achieved without relying on ground transponders or satellite navigation signals, using only the inherent geometric characteristics of the orbit itself as the sole calibration reference. The steps include: Before the train leaves the depot, the initial benchmark calibration is completed using the fixed track anchor points of the station tracks. For stations without typical characteristics, the preset coordinates + inertial coarse positioning are used, with a calibration accuracy of ≤0.1m. During operation, the sensing mode is automatically switched according to weather, lighting and tunnel scenarios, and only the inherent geometric features of track slabs, sleepers, rails and track bed are collected. In icing, oil pollution and snow accumulation scenarios, track geometry sequence is used for matching first. Based on the on-board measured temperature ΔT, the feature matching benchmark is corrected in real time through the deformation correction model ΔL=α×ΔT×L. The calibration is forced every 1 to 2 km through fixed anchor points. When the anchor points fail, the segment geometric sequence weighted matching is used as a fallback. The real-time features are matched with the standard feature library using orbital geometry sequence matching + SIFT + ICP registration, and the position deviation and mileage deviation are output with a matching delay of ≤10ms. By using filtered fusion positioning, inertial drift, wheel diameter error, idling slippage and attitude deviation are compensated in real time, and a continuous and stable position is output. Closed-loop verification is performed at key nodes. If the deviation exceeds the standard, a second correction is made. Continuous anomalies trigger emergency positioning and train speed limit and driver alarm. It operates in a continuous closed loop, achieving continuous high-precision vehicle-to-ground positioning without relying on external positioning facilities throughout the entire process.

3. The system according to claim 1, characterized in that, The multi-sensor fusion rules are as follows: in clear weather, the weights of the antenna array camera and the LiDAR are each 0.5; in rain, snow, fog, sand and dust scenes, the weight of the LiDAR is 0.7 and the weight of the orbital geometry sensor is 0.3; in tunnels / nighttime, a combination of 940nm infrared illumination, LiDAR, and orbital geometry sensor is used.

4. The system according to claim 1, characterized in that, In the track deformation correction model, α is the linear expansion coefficient of the rail (11.8e-6 / ℃), ΔT is the measured temperature difference of the on-board temperature sensing unit, L is the track reference length, and the settlement is pre-assigned using the feature library.

5. The system according to claim 1, characterized in that, The positioning control module has a particle count of ≥200, a filter update cycle of ≤20ms, and an inertial and odometer error calibration cycle of ≤50ms.

6. The system according to claim 1, characterized in that, The closed-loop verification thresholds are: mainline error ≤ 0.5m, station alignment ≤ 0.3m, tunnel ≤ 0.8m; emergency positioning error ≤ 1.5m, continuous operation ≥ 30min, and train speed limit of 300km / h or below in emergency mode.

7. The method according to claim 2, characterized in that, The inherent geometric features of the track body include: the joint spacing of ballastless track slabs, fastener arrangement, gauge sequence, and track profile; and the sleeper spacing sequence, rail profile, and ballast bed slope geometric profile of ballasted track.

8. The method according to claim 2, characterized in that, Wheel slippage is detected by combining sudden changes in wheel speed acceleration with the relative ground speed detected by lidar. Once identified, the odometer integration is instantly cut off and the location is directly reset based on feature matching.

9. The system or method according to claim 1 or 2, characterized in that, Suitable for high-speed trains, intercity trains, and urban rail transit, compatible with CRTS series ballastless and ballasted tracks, and supports continuous positioning at speeds of 350km / h and below.