Train autonomous positioning method and system in satellite signal limited environment

Abstract

The application provides a train autonomous positioning method and system in a satellite signal limited environment, and relates to the technical field of train positioning, wherein the method comprises the following steps: acquiring image data, laser point cloud data and inertial measurement data in the train running process; constructing an IMU pre-integration residual error factor by using the inertial measurement data respectively, constructing a laser radar inertial odometry residual error factor by using a local plane in combination with the laser point cloud data, and constructing a key point residual error factor by using key point positions pre-stored in an electronic map in combination with the image data; constructing a train optimal state equation in combination with the IMU pre-integration residual error factor, the odometry residual error factor and the key point residual error factor, and solving the train optimal state equation to obtain corresponding positioning data of the train; and the application can realize high-precision positioning of high-speed and long-distance railway train operation.

Description

Technical Field

[0001] This invention relates to the field of train positioning technology, and more specifically to a method and system for autonomous train positioning in environments with limited satellite signals. Background Technology

[0002] Currently, trains are the most important mechanical transportation tool in human history. To ensure the safety of train operation, the requirements for train operation and scheduling are becoming increasingly higher, and high-precision, long-duration train positioning is the foundation for safe train operation and information scheduling.

[0003] However, in areas with unstable satellite signals, the current common train positioning solution combines wheel speed odometers and track transponders. This solution suffers from low positioning accuracy and high construction and maintenance costs. With advancements in 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 satellite signal-constrained environments that can solve the above problems is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

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

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A method for autonomous train positioning in environments with limited satellite signals includes the following steps:

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

[0009] An error state equation is constructed, and the error state is predicted using the inertial measurement data. Point and surface residuals are constructed using the laser point cloud data, and the error state equation is updated. At the same time, the IMU pre-integration residual factor and the odometry residual factor are constructed using the laser point cloud data and the inertial measurement data.

[0010] The image data is processed to detect trackside signs and obtain the global position information of the feature points of the signs. The global position information of the feature points is used to construct a feature point residual factor in combination with the extracted sign feature points. The train position is then corrected and the cumulative error caused by continuous positioning of the odometer residual factor is eliminated.

[0011] By combining the IMU pre-integration residual factor and the odometer residual factor, the optimal state equation of the train is constructed, and the optimal equation of the train is solved to obtain the corresponding position information of the train.

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

[0013] A target detection model is constructed to detect trackside signs in the image data and obtain the corresponding target regions. The target detection model is the YOLOv5 model.

[0014] A scene text recognition model is constructed, the target area is input for identification information recognition, and pre-stored feature points are found from the electronic map based on the identification information.

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

[0016] The image data is inspected to obtain the target area and extract the sign feature points;

[0017] The local coordinates of the sign feature points are obtained using the laser point cloud data, and the global position coordinates of the feature points are obtained by combining the train position.

[0018] The target area is input into the scene text recognition model to identify the identification information and find the global location coordinates of pre-stored feature points from the electronic map based on the identification information;

[0019] By combining the sign feature points obtained from the laser point cloud with the global location coordinates of the feature points found from the electronic map based on the sign information, the feature point residuals are calculated and feature point residual factors are constructed.

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

[0021] An IMU motion model is constructed using the inertial measurement data;

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

[0023] Determine the IMU pre-integration residuals and apply Coriolis correction to the IMU pre-integration;

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

[0025] Preferably, the specific process for constructing the odometer residual factor includes:

[0026] Based on the laser point cloud data and inertial measurement data, the system dynamic equations under error conditions are constructed;

[0027] The laser point cloud data is preprocessed, and a residual from the preprocessed laser point cloud data to the plane is constructed using a local plane. The residual is then used to update the system dynamic equations.

[0028] An error Kalman filter is constructed, and the updated system dynamics equations are predicted using the error Kalman filter. The odometer residual factor is constructed based on the prediction results.

[0029] Preferably, the specific expression of the train's optimal state equation is as follows:

[0030]

[0031] In the formula, e prior e represents the prior factor marginalized by the Schur complement method. imu e represents the IMU pre-integration residual factor. lio e is the residual factor of the laser odometry. kps The residual between the corner coordinates of the trackside sign and its corresponding point coordinates on the electronic map is used to solve this optimization equation to obtain the optimal train pose.

[0032] This invention also provides a train autonomous positioning system in a satellite signal-constrained environment, comprising:

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

[0034] Interval positioning odometer module: Constructs an error state equation, predicts the error state using the inertial measurement data, updates the error state equation using the laser point cloud data, and simultaneously constructs the IMU pre-integration residual factor and the odometer residual factor;

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

[0036] Global positioning odometer module: Based on the identification information, query the global location information of the corresponding feature points through the electronic map, and then use the global location information of the feature points, combined with the feature points of the signboard, to construct the feature point residual factor; at the same time, combine the IMU pre-integration residual factor, the odometer residual factor and the feature point residual factor to construct and solve the train optimal state equation, correct the train position and eliminate the cumulative error caused by continuous positioning of the odometer.

[0037] Electronic map: Used to store track geographic information required for train positioning, as well as global location information of pre-marked trackside sign feature points.

[0038] As can be seen from the above technical solution, compared with the prior art, this invention discloses a method and system for autonomous train positioning in environments with limited satellite signals. Specifically, it designs an IMU pre-integration residual factor incorporating the Earth's rotation and an odometer residual factor incorporating degradation detection, tailored to the railway environment. Secondly, it introduces a layered visual detection method and designs a novel signage along the railway track, reducing reliance on expensive track transponders and eliminating the cumulative error caused by continuous odometer positioning. Finally, real-world railway train experiments demonstrate the effectiveness of the proposed method. The positioning accuracy of the proposed method is superior to existing solutions combining wheel odometers and transponders, providing a new approach for the development of intelligent railway trains. Attached Figure Description

[0039] 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 embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0040] Figure 1 A flowchart of an autonomous train positioning method in a satellite signal-constrained environment provided by the present invention;

[0041] Figure 2 This invention provides a system structure diagram of a train autonomous positioning method under satellite signal-constrained environments.

[0042] Figure 3 This is a schematic diagram illustrating the calculation of point-surface residuals provided in an embodiment of the present invention;

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

[0044] Figure 5 This is a schematic diagram of camera image and laser point cloud registration and trackside sign corner points provided in an embodiment of the present invention;

[0045] Figure 6 This is a schematic diagram of the PaddleOCR recognition technology architecture for trackside signs provided in an embodiment of the present invention. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] See Figure 1 As shown in the figure, this invention discloses a method for autonomous train positioning in a satellite signal-constrained environment, comprising the following steps:

[0048] The system acquires image data, laser point cloud data, and inertial measurement data along the railway line during train operation. The inertial measurement data includes the three-axis acceleration and three-axis angular velocity output by the inertial navigation system.

[0049] An error state equation is constructed, and the error state is predicted using the inertial measurement data. Point and surface residuals are constructed using the laser point cloud data to update the error state equation. At the same time, IMU pre-integration residual factor and odometry residual factor are constructed.

[0050] By combining the IMU pre-integration residual factor and the odometer residual factor, the optimal state equation of the train is constructed, and the optimal equation of the train is solved to obtain the corresponding position information of the train.

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

[0052] Construct an IMU motion model using inertial measurement data;

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

[0054] Determine the IMU pre-integration residuals and apply Coriolis correction to the IMU pre-integration;

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

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

[0057] Specifically, the measurement model expression for the IMU is as follows:

[0058]

[0059] in, These are the acceleration and angular velocity outputs measured by the IMU, respectively; a t w t These are the actual acceleration and angular velocity, respectively; b at b wt The zero-biased values ​​for acceleration and angular velocity are specifically modeled as random walks, with derivatives following a Gaussian distribution, as follows:

[0060]

[0061] n a n w The random errors in acceleration and angular velocity are modeled as Gaussian noise, as follows:

[0062]

[0063] g w This refers to gravitational acceleration in the world coordinate system. This is the transformation matrix from the world coordinate system to the local volume coordinate system.

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

[0065]

[0066] In the formula, t k and t k+1 Position in the world coordinate system at any given moment. t k and t k+1 Velocity in the world coordinate system at any given moment. t k and t k+1 Quaternion in the world coordinate system of time, 2[ω e ×]v t The term represents the Coriolis force caused by the Earth's rotation, where ω e g is the angular velocity of Earth's rotation. w This represents the gravitational acceleration in the world coordinate system.

[0067] The pre-integral estimate is calculated according to equation (1), and the specific expression is as follows:

[0068]

[0069] In the formula, Δp cor Δv cor The Coriolis correction for position and velocity is expressed as follows:

[0070]

[0071] The IMU pre-integration residual is defined as the difference between the estimated increments of position, velocity, and orientation and the measured increments, and is expressed as follows:

[0072]

[0073] In the formula, t k To t k+1 The pre-integral measurements of position, velocity, and direction at that moment are expressed as follows:

[0074]

[0075] in,

[0076]

[0077]

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

[0079] Based on the inertial measurement data, the system dynamics equations under the error state are constructed, and an error Kalman filter is constructed based on the equations to predict the error state.

[0080] The laser point cloud data is preprocessed, and a residual from the preprocessed laser point cloud data to the plane is constructed using a local plane. The residual is then used to update the error state vector.

[0081] Based on the filter output error state vector and the degradation detection factor, the odometer residual factor is constructed.

[0082] Specifically, the error Kalman filter maintains both the nominal state vector and the error state vector. The nominal state is the ideal state of the IMU output, containing only large signal quantity measurements and excluding random noise and disturbances. The equation of motion is:

[0083]

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

[0085]

[0086] Most of the symbols in the formula represent the error states of the corresponding symbols in formula (9), where δθ is the angular error.

[0087]

[0088] The corresponding equations of motion for the real state are:

[0089]

[0090] Abbreviated as:

[0091]

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

[0093] Specifically, the error Kalman filter iteration mainly consists of two steps: prediction and update.

[0094] In the prediction step, the nominal state vector is forward propagated through the measurement data from the integral IMU, while Gaussian estimation is performed on the random errors and disturbances accumulated over time, i.e., the error state vector.

[0095] The update process involves first preprocessing the received lidar point cloud data, then using the point-surface residuals constructed from the local plane as an error measurement to correct the error state vector and obtain a posterior Gaussian estimate of the error state.

[0096] Specifically, the preprocessing of the point cloud data involves combining the sampling points within the same scanning cycle with IMU data for motion compensation, which is divided into two steps: backpropagation and forward propagation.

[0097] First, by combining IMU data and using backpropagation, the sampling points within the same scan cycle are regressed from the processing time at the end of the scan to the actual sampling time of each point. The backpropagation formula is:

[0098]

[0099] in, These are estimates of the nominal state vectors at the actual sampling time and processing time of the sampling points, respectively. Δ t is the difference between the actual sampling time j and time k-1 of the sampling point; then, combined with the IMU data, the sampling points are 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 in formula (13), and Δt is the difference between the actual sampling time j and time k of the sampling point.

[0102] Figure 3This is a schematic diagram of the calculation of the point-surface residual, where For locally known points, Let be the normal vector of the small plane. Let these be the lidar sampling points. Based on the assumption that a single lidar sampling point and its five nearest known points lie within the same small plane, the residual measurement equation is constructed as follows:

[0103]

[0104] In the formula, This is the transformation matrix for transforming the lidar coordinate system to the IMU coordinate system. This refers to the random error in lidar measurements.

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

[0106]

[0107] In the formula, λ is the normalized degradation detection factor. This 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 equation (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 the two degrees of freedom at a 95% confidence level, E(x) is the expected value of the eigenvalues, and the degradation detection threshold is:

[0111]

[0112] In one specific embodiment, the process of constructing the keypoint residual factor includes:

[0113] The image data is subjected to trackside sign target detection to obtain the target area and extract the four corner points of the sign as key points;

[0114] The local coordinates of the corner points of the sign are obtained through the laser point cloud data, and the corner points of the sign are transformed from the local coordinate system to the global coordinate system by combining the previously calculated train position.

[0115] The target area is input into the scene text recognition model to identify the identification information and find the global position coordinates of the corner point of the pre-stored sign from the electronic map based on the identification information;

[0116] By combining the global position coordinates of the corner points and key points, the key point residuals are calculated to construct the key point residual factor.

[0117] Figure 4 The trackside sign is made of 1.2mm thick aluminum plate, with a size of 50cm×40cm. The number font is Arial, and the surface is covered with high-quality reflective film, which makes it reflective of auxiliary light sources in low light environment.

[0118] Specifically, the trackside sign target detection uses the YOLOv5 network model, which mainly consists of four parts: input, backbone, Neck, and output. The target region is the image region whose detection confidence exceeds a threshold, as outlined by the output of the YOLOv5 convolutional neural network.

[0119] Specifically, the local coordinates of the corner points of the trackside sign are extracted by registering the camera image with the laser point cloud, fitting the point cloud plane corresponding to the target area, and combining the view frustum to obtain the coordinates of the corner points of the trackside sign.

[0120] Figure 5 The diagram illustrates the registration of the camera image with the laser point cloud and the coordinates of the corner points of the trackside sign. The pixel coordinates of the corner points of the trackside sign within the target area, output through 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 registration process between the camera image and the laser point cloud is as follows: The laser point cloud at time i, combined with the offline pre-calibrated extrinsic parameter matrix between the lidar and the camera, is projected onto the camera coordinate system, as shown in the following equation:

[0122]

[0123] In the formula, These are the coordinates in the camera and lidar coordinate systems, respectively. This is the transformation matrix between the two coordinate systems. The displacement is between the two coordinate systems; combining the camera intrinsic parameter matrix and equation (19), the pixel coordinates corresponding to the laser point cloud can be obtained, as shown in the following equation:

[0124]

[0125] In the formula, f x f y The focal lengths on the x and y axes are respectively, cx c y The x and y axes are the optical center coordinates, respectively, and the other symbols are the same as in equation (19).

[0126] Specifically, the point cloud plane fitting process is as follows: Points projected from equations (19) and (20) and falling within the target region 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] Where, x i y i z i p in lidar point cloud i Point coordinates, v is the normal vector of the fitted plane.

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

[0130] Specifically, the scene text recognition model adopts the PaddleOCR model based on the PaddlePaddle deep learning framework. This model features an ultra-lightweight design and is easy to deploy on mobile and embedded systems. Figure 6 The process of recognizing 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 using the optical character recognition (OCR) method.

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

[0132]

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

[0134]

[0135] In the formula, e prior e represents the prior factor marginalized by the Schur complement method. imu e represents the IMU pre-integration residual factor. lio e is the residual factor of the laser odometry. kpsThe residual between the corner coordinates of the trackside sign and its corresponding point coordinates on the electronic map is used to solve this optimization equation to obtain the optimal train pose.

[0136] See Figure 2 As shown, this embodiment of the invention also provides a system utilizing any one of the above embodiments of a train autonomous positioning method under satellite signal-constrained environments, comprising:

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

[0138] Interval positioning odometer module: Constructs an error state equation, predicts the error state using the inertial measurement data, updates the error state equation using the laser point cloud data, and simultaneously constructs the IMU pre-integration residual factor and the odometer residual factor.

[0139] Layered visual detection module: performs trackside sign target detection on the image data, identifies the sign information of the detected target areas, and extracts the sign feature points by combining the laser point cloud data.

[0140] The global positioning odometer module: Based on the identification information, it queries the global location information of corresponding key points through an electronic map. Then, using the global location information of the key points and the feature points of the signposts, it constructs key point residual factors. Simultaneously, it constructs and solves the optimal state equation of the train by combining the IMU pre-integration residual factor, the odometer residual factor, and the key point residual factor, correcting the train position and eliminating the cumulative error caused by continuous odometer positioning.

[0141] Electronic map: Used to store track geographic information required for train positioning, as well as global location information of pre-marked trackside sign feature points.

[0142] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0143] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those 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 invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for autonomous train positioning in environments with limited satellite signals, characterized in that, Includes the following steps: Acquire image data, laser point cloud data, and inertial measurement data during train operation; An error state equation is constructed, and error state prediction is performed using the inertial measurement data. Point-area residuals are constructed using the laser point cloud data to update the error state equation. Simultaneously, IMU pre-integration residual factors and odometer residual factors are constructed using the laser point cloud data and inertial measurement data. The specific process of constructing the odometer residual factors includes: Based on the laser point cloud data and inertial measurement data, the system dynamic equations under error conditions are constructed; The laser point cloud data is preprocessed, and a residual from the preprocessed laser point cloud data to the plane is constructed using a local plane. The residual is then used to update the system dynamic equations. An error Kalman filter is constructed, and the updated system dynamics equations are predicted using the error Kalman filter. The odometer residual factor is constructed based on the prediction results. The error Kalman filter iteration mainly consists of two steps: prediction and update. In the prediction step, the nominal state vector is propagated forward using the integrated IMU measurement data. Simultaneously, Gaussian estimation is performed on the random errors and disturbances accumulated over time, i.e., the error state vector. The update step, upon receiving the lidar point cloud data, first performs preprocessing. Preprocessing involves combining the sampling points within the same scanning cycle with the IMU measurement data for motion compensation. This preprocessing is divided into two steps: backpropagation and forward propagation. Specifically: The backpropagation step involves combining IMU measurement data to backpropagate the sampling points within the same scan cycle from the processing time at the end of the scan to the actual sampling time of each point. The forward propagation step involves combining IMU measurement data again to propagate each sampling point to the processing time at the end of the scan through forward propagation. Then, the point-surface residuals constructed using the local plane are used as error measurements to correct the error state vector, thus obtaining the posterior Gaussian estimate of the error state. The image data is processed to detect trackside signs and obtain the global position information of the feature points of the signs. The global position information of the feature points is used to construct a feature point residual factor in combination with the extracted sign feature points. The train position is then corrected and the cumulative error caused by continuous positioning of the odometer residual factor is eliminated. By combining the IMU pre-integration residual factor and the odometer residual factor, the optimal state equation of the train is constructed, and the optimal 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-constrained environment according to claim 1, characterized in that, The specific process of obtaining global location information of feature points includes: A target detection model is constructed to detect trackside signs in the image data and obtain the corresponding target regions. The target detection model is the YOLOv5 model. A scene text recognition model is constructed, the target area is input for identification information recognition, and pre-stored feature points are found from the electronic map based on the identification information.

3. The method for autonomous train positioning in a satellite signal-constrained environment according to claim 2, characterized in that, The specific process of constructing feature point residual factors includes: The image data is inspected to obtain the target area and extract the sign feature points; The local coordinates of the feature points of the sign are obtained using the laser point cloud data, and the global position coordinates of the feature points are obtained. The target area is input into the scene text recognition model to identify the identification information and find the global location coordinates of pre-stored feature points from the electronic map based on the identification information; By combining the feature points of the sign and their global position coordinates, the feature point residuals are calculated and the feature point residual factors are constructed.

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

5. The method for autonomous train positioning in a satellite signal-constrained environment according to claim 1, characterized in that, The specific expression for the train's optimal state equation is as follows: In the formula, e prior e represents the prior factor marginalized by the Schur complement method. imu e represents the IMU pre-integration residual factor. lio e is the residual factor of the laser odometry. kps The residual between the corner coordinates of the trackside sign and its corresponding point coordinates on the electronic map is used to solve this optimization equation to obtain the optimal train pose.

6. A system utilizing the train autonomous positioning method under satellite signal-constrained environments as described in any one of claims 1-5, characterized in that, include: Sensor module: Collects image data, laser point cloud data, and inertial measurement data of the train in motion through cameras, lidar, and inertial navigation, respectively; Interval positioning odometer module: Constructs an error state equation, predicts the error state using the inertial measurement data, updates the error state equation using the laser point cloud data, and simultaneously constructs the IMU pre-integration residual factor and the odometer residual factor; Layered visual detection module: performs trackside sign target detection on the image data, identifies the sign information in the detected target area, and extracts sign feature points by combining the laser point cloud data; Global positioning odometer module: Based on the identification information, query the global location information of the corresponding feature points through the electronic map, and then use the global location information of the feature points, combined with the feature points of the signboard, to construct the feature point residual factor; at the same time, combine the IMU pre-integration residual factor, the odometer residual factor and the feature point residual factor to construct and solve the train optimal state equation, correct the train position and eliminate the cumulative error caused by continuous positioning of the odometer. Electronic map: Used to store track geographic information required for train positioning, as well as global location information of pre-marked trackside sign feature points.

Images