Train autonomous positioning method and system in satellite signal limited environment

Abstract

The invention provides a train autonomous positioning method and system in a satellite signal limited environment, and relates to the technical field of train positioning, and the method comprises the following steps: obtaining image data, laser point cloud data and inertial measurement data in a train driving process; constructing an IMU pre-integration residual error factor by using the inertial measurement data, constructing a laser radar inertial odometer residual error factor by using a local plane construction point-surface residual error in combination with the laser point cloud data, and constructing a key point residual error factor by using a key point position pre-stored in an electronic map in combination with the image data; combining the IMU pre-integration residual factor, the odometer residual factor and the key point residual factor to construct a train optimal state equation, and solving the train optimal state equation to obtain positioning data corresponding to the train; according to the invention, high-precision positioning of high-speed and long-distance railway train operation can be realized.

Description

Technical Field

[0001] The present invention relates to the technical field of train positioning, and more particularly to a method and system for autonomous train positioning in an environment with limited satellite signals. Background Art

[0002] At present, trains are the most important mechanical means of transportation in human history. In order to ensure the safety of train operation, the requirements for train operation and scheduling are becoming increasingly higher. High-precision and long-duration train positioning is the basis for safe train operation and information scheduling.

[0003] However, in areas where satellite signals are unstable, a train positioning solution that combines wheel speedometers and track transponders is currently widely used. However, this solution suffers from low positioning accuracy and high construction and maintenance costs. With the advancement of artificial intelligence technology, the railway transportation sector urgently needs to develop intelligent train positioning solutions.

[0004] Therefore, how to provide a method for autonomous train positioning in a satellite signal-limited environment that can solve the above-mentioned problems is an issue that those skilled in the art urgently need to solve. Summary of the Invention

[0005] In view of this, the present invention provides a method and system for autonomous train positioning in an environment with limited satellite signals, which can achieve high-precision positioning of high-speed and long-distance railway trains.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] A method for autonomous train positioning in a satellite signal-limited environment comprises the following steps:

[0008] Acquire image data, laser point cloud data, and inertial measurement data during train travel;

[0009] Constructing an error state equation, using the inertial measurement data to predict the error state, using the laser point cloud data to construct the point surface residual, updating the error state equation, and simultaneously using the laser point cloud data and the inertial measurement data to construct an IMU pre-integration residual factor and an odometry residual factor;

[0010] Processing the image data, detecting trackside signs and obtaining global position information of feature points of the signs, using the global position information of the feature points in combination with the extracted feature points of the signs to construct feature point residual factors, correcting the train position and eliminating cumulative errors generated by continuous positioning of the odometer residual factors;

[0011] The IMU pre-integration residual factor and the odometer residual factor are combined to construct the optimal state equation of the train, and the optimal state equation of the train is solved to obtain the corresponding position information of the train.

[0012] Preferably, the specific process of obtaining the global position information of the feature points includes:

[0013] Constructing a target detection model to perform trackside sign target detection on the image data to obtain a corresponding target area, wherein the target detection model is a YOLOv5 model;

[0014] Build a scene text recognition model, input the target area for identification information recognition, and search for pre-stored feature points from the electronic map based on the identification information.

[0015] Preferably, the specific process of constructing the feature point residual factor includes:

[0016] Detecting the image data to obtain a target area and extracting feature points of the sign;

[0017] Using the laser point cloud data to obtain the local coordinates of the characteristic points of the sign, and combining the train position to obtain the global position coordinates of the characteristic points;

[0018] Inputting the target area into a scene text recognition model, identifying identification information and searching for pre-stored feature point global position coordinates from an electronic map according to the identification information;

[0019] The feature points of the sign acquired from the laser point cloud are combined with the global position coordinates of the feature points found from the electronic map according to the identification information to calculate the feature point residuals and construct feature point residual factors.

[0020] Preferably, the specific process of constructing the IMU pre-integration residual factor includes:

[0021] constructing an IMU motion model using the inertial measurement data;

[0022] Taking into account the influence of the earth's rotation, the IMU motion model is processed to obtain an IMU pre-integration with earth rotation compensation;

[0023] Determine the IMU pre-integration residual and perform Coriolis correction on the IMU pre-integration;

[0024] The final IMU pre-integration residual factor is determined by combining the IMU pre-integration residual and the correction result.

[0025] Preferably, the specific process of constructing the odometry residual factor includes:

[0026] Constructing a system dynamics equation under an error state based on the laser point cloud data and the inertial measurement data;

[0027] Preprocessing the laser point cloud data, constructing a residual from the preprocessed laser point cloud data to a plane using a local plane, and updating the system dynamics equation using the residual;

[0028] An error Kalman filter is constructed, and the updated system dynamics equation is predicted using the error Kalman filter, and an odometer residual factor is constructed according to the prediction result.

[0029] Preferably, the specific expression of the train optimal state equation is:

[0030]

[0031] Where, e prior represents the prior factor marginalized by Schur's complement method, e imu Represents the IMU pre-integration residual factor, e lio is the residual factor of the laser odometry, e kps It represents the residual between the coordinates of the corner points of the trackside sign and the coordinates of its corresponding point in the electronic map. Solving this optimization equation can obtain the optimal train posture.

[0032] The present invention also provides a train autonomous positioning system in a satellite signal-limited environment, comprising:

[0033] Sensor module: collects image data, laser point cloud data, and inertial measurement data of the train through cameras, lidar, and inertial navigation respectively;

[0034] Interval positioning odometry module: constructs an error state equation, uses the inertial measurement data to predict the error state, uses the laser point cloud data to update the error state equation, and constructs the IMU pre-integration residual factor and the odometry residual factor;

[0035] Layered visual inspection module: performs trackside sign target detection on the image data, identifies the sign information of the detected target area, and extracts the sign feature points in combination with the laser point cloud data;

[0036] Global positioning odometer module: Based on the identification information, the global position information of the corresponding feature point is queried through the electronic map. Then, the feature point global position information is used in combination with the feature point of the sign to construct the feature point residual factor. At the same time, the train optimal state equation is constructed and solved by combining the IMU pre-integration residual factor, the odometer residual factor and the feature point residual factor to correct the train position and eliminate the accumulated error caused by the continuous positioning of the odometer.

[0037] Electronic map: used to store the track geographic information required for train positioning and the global location information of pre-calibrated trackside sign feature points.

[0038] It can be seen from the above technical solutions that, compared with the existing technology, the present invention discloses a method and system for autonomous positioning of trains in a satellite signal-constrained environment, and designs an IMU pre-integrated residual factor that incorporates the earth's rotation and an odometer residual factor that adds degradation detection for the railway environment. Secondly, a layered visual detection method is introduced, and a new type of signboard next to the railway track is designed, which reduces the dependence on expensive track transponders and eliminates the cumulative error caused by the continuous positioning of the odometer. Finally, real railway train experiments have demonstrated the effectiveness of the proposed method. The positioning accuracy of the proposed method is better than the existing solution that combines wheel odometers and transponders, providing a new path for the development of intelligent railway trains. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0040] Figure 1 This is an overall flow chart of a method for autonomous train positioning in a satellite signal-restricted environment provided by the present invention;

[0041] Figure 2 This is a system structure diagram of a train autonomous positioning method in a satellite signal-restricted environment provided by the present invention;

[0042] Figure 3 A schematic diagram of point-surface residual calculation provided by an embodiment of the present invention;

[0043] Figure 4 A schematic diagram of a trackside sign provided in an embodiment of the present invention.

[0044] Figure 5 Schematic diagram of camera image and laser point cloud registration and trackside sign corner points provided by an embodiment of the present invention;

[0045] Figure 6 Schematic diagram of the PaddleOCR technology architecture for trackside sign recognition provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0046] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0047] See also Figure 1 As shown, an embodiment of the present invention discloses a method for autonomous train positioning in an environment with limited satellite signals, comprising the following steps:

[0048] Acquire image data, laser point cloud data, and inertial measurement data along the railway line during train travel, where the inertial measurement data includes the three-axis acceleration and three-axis angular velocity output by the inertial navigation system;

[0049] Constructing an error state equation, using the inertial measurement data to predict the error state, using the laser point cloud data to construct the point surface residual, updating the error state equation, and constructing the IMU pre-integration residual factor and the odometry residual factor at the same time;

[0050] The IMU pre-integration residual factor and the odometer residual factor are combined to construct the optimal state equation of the train, and the optimal equation of the train is solved to obtain the corresponding position information of the train.

[0051] In a specific embodiment, the specific process of constructing the IMU pre-integration residual factor includes:

[0052] Use inertial measurement data to build an IMU motion model;

[0053] Considering the influence of the earth's rotation, the IMU motion model is processed to obtain the IMU pre-integration with earth rotation compensation;

[0054] Determine the IMU pre-integration residual and perform Coriolis correction on the IMU pre-integration;

[0055] The final IMU pre-integration factor is determined by combining the IMU pre-integration residual and the correction result.

[0056] Specifically, considering that high-speed trains can reach speeds exceeding 400 km / h, the Earth's rotation at such speeds will affect IMU integration. Therefore, the IMU pre-integration residual factor provided by the embodiments of the present invention retains the Coriolis acceleration caused by the Earth's rotation to improve integration accuracy.

[0057] Specifically, the measurement model expression of IMU is:

[0058]

[0059] in, are the acceleration and angular velocity outputs measured by IMU respectively; a t 、w t are the real acceleration and angular velocity respectively; b at 、b wt The zero bias of acceleration and angular velocity are specifically modeled as random walks, and the derivatives satisfy the Gaussian distribution, as follows:

[0060]

[0061] n a 、n w are the random errors of acceleration and angular velocity, respectively, modeled as Gaussian noise, as follows:

[0062]

[0063] g w is the gravitational acceleration in the world coordinate system; is the transformation matrix from the world coordinate system to the local body coordinate system.

[0064] Based on the IMU measurement model, the specific expression of the motion model with earth rotation compensation in the local body coordinate system is derived as follows:

[0065]

[0066] Where, t k and t k+1 The position in the world coordinate system at that moment, t k and t k+1 The speed in the world coordinate system at that moment, t k and t k+1 Quaternion in the world coordinate system at the moment, 2[ω e ×]v t The term is the Coriolis force caused by the rotation of the Earth, where ω e is the Earth's rotational angular velocity, g w is the gravitational acceleration in the world coordinate system.

[0067] The pre-integration estimate is calculated according to formula (1). The specific expression is:

[0068]

[0069] Where Δp cor , Δv cor is the Coriolis correction term for position and velocity, and its specific expression is:

[0070]

[0071] The IMU pre-integration residual is defined as the difference between the estimated increment and the measured increment of position, velocity, and orientation. The specific expression is:

[0072]

[0073] Where, t k to t k+1 The pre-integrated measurement values ​​of position, velocity, and direction at a given moment are expressed as follows:

[0074]

[0075] in,

[0076]

[0077]

[0078] In a specific embodiment, the specific process of constructing the odometry residual factor includes:

[0079] constructing a system dynamics equation under an error state according to the inertial measurement data, constructing an error Kalman filter according to the equation and performing error state prediction;

[0080] Preprocessing the laser point cloud data, constructing a residual from the preprocessed laser point cloud data to a plane using a local plane, and updating the error state vector using the residual;

[0081] An odometer residual factor is constructed based on the filter output error state vector combined with a degradation detection factor.

[0082] Specifically, the error Kalman filter maintains the nominal state vector and the error state vector at the same time, where the nominal state is the ideal state output by the IMU containing only large signal measurements, excluding random noise and disturbances, and the motion equation is:

[0083]

[0084] Most of the symbols in the formula are the same as those in the IMU measurement model of formulas (1) to (3). are the derivatives of the accelerometer and gyroscope bias respectively. The error state is the small signal part of the IMU output that meets the linear Gaussian filter, and the motion equation is:

[0085]

[0086] Most of the symbols in the formula are the error states of the corresponding symbols in formula (9), where δθ is the angle error,

[0087]

[0088] The corresponding true state equation of motion is:

[0089]

[0090] In short:

[0091]

[0092] Among them, x t is the system state vector, u is the IMU measurement data without random noise, and w is the derivative of the IMU bias error.

[0093] Specifically, the error Kalman filter iteration is mainly divided into two steps: prediction and update:

[0094] In the prediction step, the nominal state vector is forward propagated by integrating the IMU measurement data, and a Gaussian estimate is performed on the random errors and disturbances accumulated over time, i.e., the error state vector.

[0095] The update step first preprocesses the lidar point cloud data after receiving it, and then uses the local plane to construct the point-surface residual as the error measurement to correct the error state vector and obtain the posterior Gaussian estimate of the error state.

[0096] Specifically, the preprocessing of the point cloud data is to combine the sampling points in the same scanning cycle with the IMU data for motion compensation, which is divided into two steps: back propagation and forward propagation:

[0097] First, the sampling points in the same scanning cycle are backpropagated from the processing time at the end of the scan to the actual sampling time of each point in combination with the IMU data. The backpropagation formula is:

[0098]

[0099] in, are the estimates of the nominal state vector at the actual sampling moment and processing moment of the sampling point, Δ t is the difference between the actual sampling time j and the time k-1 of the sampling point; then, combined with the IMU data, each sampling point is uniformly propagated to the processing time at the end of the scan through forward propagation. The forward propagation formula is:

[0100]

[0101] Most of the symbols in the formula are the same as those in formula (13), and Δt is the difference between the actual sampling time j and the time k of the sampling point

[0102] Figure 3is a schematic diagram of the point-surface residual calculation, where is a local known point, is the normal vector of the facet, is the laser radar sampling point. Based on the assumption that a single laser radar sampling point and the five nearest known points are located in the same small plane, the residual measurement equation is constructed:

[0103]

[0104] Where, is the transformation matrix from the lidar coordinate system to the IMU coordinate system, is the random error of the lidar measurement.

[0105] The odometry residual factor is composed of an error vector and a normalized degradation factor, as shown in the following formula:

[0106]

[0107] Where λ is the normalized degradation detection factor, is the error state vector.

[0108] Specifically, the normalized degradation detection factor is a non-heuristic adaptive degradation detection factor. The chi-square test is performed using the eigenvalue normalization of the information matrix in formula (15) as the threshold for degradation detection. The chi-square test formula for rejecting the null hypothesis is:

[0109]

[0110] Where 0.103 represents the chi-square value of two degrees of freedom at a 95% confidence level, E(x) is the expected value of the eigenvalue, and the degradation detection threshold is:

[0111]

[0112] In a specific embodiment, the process of constructing the key point residual factor includes:

[0113] Performing trackside sign target detection on the image data to obtain a target area and extracting four corner points of the sign as key points;

[0114] Obtain the local coordinates of the corner points of the signboard using the laser point cloud data, and convert the corner points of the signboard from the local coordinate system to the global coordinate system in combination with the previously calculated train position;

[0115] Inputting the target area into a scene text recognition model, identifying identification information and searching for pre-stored global position coordinates of the corner points of the sign from an electronic map according to the identification information;

[0116] The global position coordinates of the corner points and key points are combined to calculate the key point residuals to construct the key point residual factors.

[0117] Figure 4 The trackside sign is made of 1.2mm thick aluminum plate with a size of 50cm×40cm. The digital font is Arial and the surface is covered with high-quality reflective film to provide a better reflection effect on the auxiliary light source in a weak light environment.

[0118] Specifically, the trackside sign target detection adopts the YOLOv5 network model, which is mainly composed of four parts: input end, backbone network, Neck and output end; the target area is the image area framed by the output end of the YOLOv5 convolutional neural network and the detection confidence exceeds the threshold.

[0119] Specifically, the local coordinates of the corner points of the trackside sign are extracted, the camera image is aligned with the laser point cloud, the point cloud plane corresponding to the target area is fitted, and the coordinates of the corner points of the trackside sign are obtained in combination with the viewing cone.

[0120] Figure 5 The camera image and laser point cloud registration and the coordinates of the trackside sign corners are shown, wherein the pixel coordinates of the trackside sign corners contained in the target area output by visual target detection are denoted as (u a ,v a )、(u b ,v b )、(u c ,v c )、(u d ,v d ).

[0121] Specifically, the camera image and laser point cloud registration process is as follows: the laser point cloud at time i is combined with the offline pre-calibrated external parameter matrix between the laser radar and the camera to be projected onto the camera coordinate system, as shown in the following formula:

[0122]

[0123] Where, are the coordinates of the camera and lidar coordinate systems respectively, is the transformation matrix between the two coordinate systems, is the displacement between the two coordinate systems; combined with the camera intrinsic parameter matrix and formula (19), the pixel coordinates corresponding to the laser point cloud can be obtained as shown below:

[0124]

[0125] Where, f x 、f y are the focal lengths of the x and y axes, respectively, cx 、c y are the coordinates of the optical center of the x-axis and y-axis respectively, and the other symbols are the same as those in formula (19).

[0126] Specifically, the point cloud plane fitting process is as follows: the points that fall within the target area after projection according to equations (19) and (20) are fitted using the RANSAC algorithm, a least squares problem is constructed based on the plane equation, and the plane equation is solved using QR decomposition, as shown in the following equation:

[0127]

[0128] Among them, x i 、y i 、z i is the p in the lidar point cloud i Point coordinates, v is the normal vector of the fitted plane.

[0129] Specifically, the local coordinates of the corner points of the trackside signboard can be obtained by combining the fitting plane equation with equations (19) and (20) to obtain the local coordinates of the corner points in the laser radar fitting plane point cloud set, and then combined with the previous train position to obtain the global coordinates of the corner points.

[0130] Specifically, the scene text recognition model uses the PaddleOCR model based on the PaddlePaddle deep learning framework, which adopts an ultra-lightweight design and is easy to deploy on mobile and embedded systems. Figure 6 The recognition process of the identification information of the trackside sign is described. After obtaining the coordinates of the target area, the scene text model recognizes the digital information in the area through the optical character recognition (OCR) method.

[0131] Specifically, the key point residual factor is the difference between the pre-stored key point global coordinates extracted from the electronic map through the digital information identified by the trackside sign and the global coordinates of the corner points extracted from the laser radar fitted plane point cloud set of the target area, as follows:

[0132]

[0133] In a specific embodiment, the specific expression of the train optimal state equation is:

[0134]

[0135] Where, e prior represents the prior factor marginalized by Schur's complement method, e imu Represents the IMU pre-integration residual factor, e lio is the residual factor of the laser odometry, e kpsIt represents the residual between the coordinates of the corner points of the trackside sign and the coordinates of its corresponding point in the electronic map. Solving this optimization equation can obtain the optimal train posture.

[0136] See also Figure 2 As shown, an embodiment of the present invention further provides a system for utilizing the method for autonomous train positioning in a satellite signal-restricted environment as described in any one of the above embodiments, comprising:

[0137] Sensor module: collects image data, point cloud data and inertial measurement data of the train through cameras, lidar and inertial navigation respectively.

[0138] Interval positioning odometry module: constructs an error state equation, uses the inertial measurement data to predict the error state, uses the laser point cloud data to update the error state equation, and constructs the IMU pre-integration residual factor and the odometry residual factor.

[0139] Layered visual inspection module: performs trackside sign target detection on the image data, identifies the sign information of the detected target area, and extracts the sign feature points in combination with the laser point cloud data.

[0140] The global positioning odometry module queries the global position information of the corresponding key points on the electronic map based on the identification information. This global position information is then used in conjunction with the signboard feature points to construct key point residual factors. Simultaneously, the train's optimal state equation is constructed and solved by combining the IMU pre-integration residual factors, the odometry residual factors, and the key point residual factors. This corrects the train's position and eliminates the accumulated errors generated by the odometry's continuous positioning.

[0141] Electronic map: used to store the track geographic information required for train positioning and the global location information of pre-calibrated trackside sign feature points.

[0142] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0143] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for autonomous train positioning in a satellite signal-limited environment, characterized in that: The following steps are involved: Acquire image data, laser point cloud data, and inertial measurement data during train travel; Constructing an error state equation, using the inertial measurement data to predict the error state, using the laser point cloud data to construct the point surface residual, updating the error state equation, and simultaneously using the laser point cloud data and the inertial measurement data to construct an IMU pre-integration residual factor and an odometry residual factor; Processing the image data, detecting trackside signs and obtaining global position information of feature points of the signs, using the global position information of the feature points in combination with the extracted feature points of the signs to construct feature point residual factors, correcting the train position and eliminating cumulative errors generated by continuous positioning of the odometer residual factors; The IMU pre-integration residual factor and the odometer residual factor are combined to construct the optimal state equation of the train, and the optimal state equation of the train is solved to obtain the corresponding position information of the train.

2. The method for autonomous train positioning in a satellite signal-limited environment according to claim 1, characterized in that: The specific process of obtaining the global position information of feature points includes: Constructing a target detection model to perform trackside sign target detection on the image data to obtain a corresponding target area, wherein the target detection model is a YOLOv5 model; Build a scene text recognition model, input the target area for identification information recognition, and search for pre-stored feature points from the electronic map based on the identification information.

3. The method for autonomous train positioning in a satellite signal-limited environment according to claim 2, characterized in that: The specific process of constructing the feature point residual factor includes: Detecting the image data to obtain a target area and extracting feature points of the sign; Using the laser point cloud data to obtain the local coordinates of the feature points of the sign, and obtain the global position coordinates of the feature points; Inputting the target area into a scene text recognition model, identifying identification information and searching for pre-stored feature point global position coordinates from an electronic map according to the identification information; The feature point residuals are calculated and feature point residual factors are constructed based on the feature points of the sign and the global position coordinates of the feature points.

4. The method for autonomous train positioning in a satellite signal-limited environment according to claim 1, characterized in that: The specific process of constructing the IMU pre-integration residual factor includes: constructing an IMU motion model using the inertial measurement data; Taking into account the influence of the earth's rotation, the IMU motion model is processed to obtain an IMU pre-integration with earth rotation compensation; Determine the IMU pre-integration residual and perform Coriolis correction on the IMU pre-integration; The final IMU pre-integration residual factor is determined by combining the IMU pre-integration residual and the correction result.

5. The method for autonomous train positioning in a satellite signal-limited environment according to claim 1, characterized in that: The specific process of constructing the odometry residual factor includes: Constructing a system dynamics equation under an error state based on the laser point cloud data and the inertial measurement data; Preprocessing the laser point cloud data, constructing a residual from the preprocessed laser point cloud data to a plane using a local plane, and updating the system dynamics equation using the residual; An error Kalman filter is constructed, and the updated system dynamics equation is predicted using the error Kalman filter, and an odometer residual factor is constructed according to the prediction result.

6. The method for autonomous train positioning in a satellite signal-limited environment according to claim 1, characterized in that: The specific expression of the train optimal state equation is: Where, e prior represents the prior factor marginalized by Schur's complement method, e imu Represents the IMU pre-integration residual factor, e lio is the residual factor of the laser odometry, e kps It represents the residual between the coordinates of the corner points of the trackside sign and the coordinates of its corresponding point in the electronic map. Solving this optimization equation can obtain the optimal train posture.

7. A system utilizing the method for autonomous train positioning in a satellite signal-restricted environment according to any one of claims 1 to 6, characterized in that: include: Sensor module: collects image data, laser point cloud data, and inertial measurement data of the train through cameras, lidar, and inertial navigation respectively; Interval positioning odometry module: constructs an error state equation, uses the inertial measurement data to predict the error state, uses the laser point cloud data to update the error state equation, and constructs the IMU pre-integration residual factor and the odometry residual factor; Layered visual inspection module: performs trackside sign target detection on the image data, identifies the sign information of the detected target area, and extracts the sign feature points in combination with the laser point cloud data; Global positioning odometer module: Based on the identification information, the global position information of the corresponding feature point is queried through the electronic map. Then, the feature point global position information is used in combination with the feature point of the sign to construct the feature point residual factor. At the same time, the train optimal state equation is constructed and solved by combining the IMU pre-integration residual factor, the odometer residual factor and the feature point residual factor to correct the train position and eliminate the accumulated error caused by the continuous positioning of the odometer. Electronic map: used to store the track geographic information required for train positioning and the global location information of pre-calibrated trackside sign feature points.

Images