High-precision railway map generation method and system based on single-pass trajectory correction

Abstract

The application provides a high-precision railway map generation method and system based on one-way trajectory correction, relates to the technical field of heavy-load railway transportation, and comprises the following steps: simultaneously collecting a continuous image sequence and spatial attitude data of a current train in a driving process; determining pose change data of the current train based on visual feature points in the continuous image sequence; performing synchronous positioning and map construction based on the pose change data and the spatial attitude data to generate an initial electronic map; and correcting the mileage and trajectory of the current train in the initial electronic map based on mileage information and image positions of preset markers in the continuous image sequence to generate an electronic map corresponding to the driving process of the current train. The method and system provided by the application improve the generation efficiency and accuracy of the railway electronic map.

Description

Technical Field

[0001] This application relates to the field of heavy-haul railway transportation technology, and in particular to a method and system for generating high-precision railway maps based on single-journey trajectory correction. Background Technology

[0002] Heavy-haul rail transport plays a vital role globally. The safe operation, efficient planning, and meticulous maintenance of railway systems increasingly rely on accurate electronic maps. Therefore, the development of high-precision railway electronic maps presents significant technological demands and development potential.

[0003] However, traditional railway surveying methods exhibit significant limitations when dealing with long distances and complex terrain. For example, mobile surveying vehicles and UAV aerial surveying, while providing a certain level of accuracy, require substantial time and resources, especially on long-distance railways in complex terrain. Furthermore, these methods have limitations in data fusion and processing, particularly in the fusion of multi-source heterogeneous data. Integrating data from different sensors struggles to achieve a unified representation, thus affecting map accuracy and system robustness.

[0004] Therefore, improving the efficiency and accuracy of railway electronic map generation has become a pressing technical problem for the industry. Summary of the Invention

[0005] This application provides a method and system for generating high-precision railway maps based on single-journey trajectory correction, which addresses the technical problem of how to improve the generation efficiency and accuracy of railway electronic maps.

[0006] This application provides a method for generating high-precision railway maps based on single-journey trajectory correction, including:

[0007] Simultaneously, continuous image sequences and spatial attitude data of the train during its current operation are collected;

[0008] The pose change data of the current train is determined based on the visual feature points in the continuous image sequence;

[0009] Based on the pose change data and the spatial attitude data, synchronous positioning and map construction are performed to generate an initial electronic map;

[0010] Based on the mileage information and image position corresponding to the preset markers in the continuous image sequence, the mileage and trajectory of the current train in the initial electronic map are corrected to generate the electronic map corresponding to the current train during its operation.

[0011] In some embodiments, after simultaneously acquiring continuous image sequences and spatial attitude data of the current train during its operation, the method further includes:

[0012] The continuous image sequence is preprocessed; the preprocessing includes at least one of distortion correction, denoising, and image enhancement.

[0013] Based on the timestamps of the left and right eye images in the continuous image sequence, the binocular images in the continuous image sequence are filtered;

[0014] Based on the timestamps of the stereo images in the continuous image sequence, the spatial pose data is synchronized and calibrated.

[0015] In some embodiments, the step of synchronously locating and mapping based on the pose change data and the spatial pose data to generate an initial electronic map includes:

[0016] The state vector is determined based on the actual measurement values ​​of each sensor; the sensors include a binocular camera that acquires the continuous image sequence and an inertial measurement unit that acquires the spatial attitude data;

[0017] Based on the actual measured values ​​and expected calculated values ​​of each sensor at each time point, and the covariance matrix of the actual measured values, the state estimation cost function is determined.

[0018] The state estimation cost function is optimized to obtain the optimal estimate of the state vector of the current train during its operation.

[0019] Based on the optimal estimate of the state vector, synchronous positioning and map construction are performed to generate an initial electronic map.

[0020] In some embodiments, the step of correcting the mileage and trajectory of the current train in the initial electronic map based on the mileage information and image position corresponding to preset markers in the continuous image sequence, and generating an electronic map corresponding to the current train during its operation, includes:

[0021] The continuous image sequence is subjected to marker recognition to determine the target type of each preset marker in the continuous image sequence;

[0022] Based on the target type of each preset marker, determine the mileage information corresponding to each preset marker;

[0023] The mileage information corresponding to each preset marker is compared with the measured mileage of the current train, and the mileage of the current train in the initial electronic map is corrected based on the mileage deviation between the mileage information and the measured mileage.

[0024] In some embodiments, the step of correcting the mileage and trajectory of the current train in the initial electronic map based on the mileage information and image position corresponding to preset markers in the continuous image sequence, and generating an electronic map corresponding to the current train during its operation, includes:

[0025] The continuous image sequence is subjected to marker recognition to determine the image position of each preset marker in the continuous image sequence;

[0026] Based on the image position and spatial coordinate position of each preset marker, and the parameters of the binocular camera that acquires the continuous image sequence, the actual pose information of the binocular camera is determined.

[0027] Based on the actual pose information of the binocular camera, the trajectory of the current train in the initial electronic map is corrected.

[0028] In some embodiments, the marker identification of the continuous image sequence includes:

[0029] The continuous image sequence is input into the marker recognition model to obtain the target type, image location and confidence level of each preset marker output by the marker recognition model;

[0030] The marker recognition model is trained based on sample images corresponding to each preset marker.

[0031] In some embodiments, after correcting the mileage and trajectory of the current train in the initial electronic map based on the mileage information and image position corresponding to preset markers in the continuous image sequence, and generating the electronic map corresponding to the current train during its operation, the method further includes:

[0032] Determine the latitude and longitude coordinates of the starting point of the current train on the electronic map;

[0033] Based on the latitude and longitude coordinates of the starting point and the relative displacement data of each trajectory point of the current train, the latitude and longitude coordinates of each trajectory point are determined.

[0034] Based on the latitude and longitude coordinates, trajectory point number, road segment name, road segment type, and road segment length of each trajectory point, a key point table, a road segment table, and a road segment details table corresponding to the electronic map are generated.

[0035] This application provides a high-precision railway map generation system based on single-journey trajectory correction, including:

[0036] The data acquisition module is used to simultaneously acquire continuous image sequences and spatial attitude data of the train during its current operation.

[0037] The feature extraction module is used to determine the pose change data of the current train based on visual feature points in the continuous image sequence;

[0038] The synchronous mapping module is used to perform synchronous positioning and map construction based on the pose change data and the spatial attitude data to generate an initial electronic map.

[0039] The map correction module is used to correct the mileage and trajectory of the current train in the initial electronic map based on the mileage information and image position corresponding to preset markers in the continuous image sequence, and generate an electronic map corresponding to the current train during its operation.

[0040] This application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the high-precision railway map generation method based on single-journey trajectory correction.

[0041] This application provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the high-precision railway map generation method based on single-journey trajectory correction.

[0042] The high-precision railway map generation method and system based on single-journey trajectory correction provided in this application simultaneously collects continuous image sequences and spatial attitude data of the current train during its journey; determines the pose change data of the current train based on visual feature points in the continuous image sequence; performs synchronous positioning and map construction based on the pose change data and spatial attitude data to generate an initial electronic map; and corrects the mileage and trajectory of the current train in the initial electronic map based on the mileage information and image position corresponding to preset markers in the continuous image sequence to generate the corresponding electronic map of the current train during its journey. By combining continuous image sequences collected by a binocular camera with spatial attitude data collected by an inertial measurement unit, high-precision positioning and map construction in a railway environment are achieved. This enables more accurate handling of positioning and mapping problems under high-speed train movement, improving the stability and accuracy of electronic map construction. Furthermore, by using the mileage information and image position of preset markers in the railway environment to correct the mileage and trajectory of the current train in the initial electronic map, the accumulated error during long-distance travel is effectively reduced. This method is particularly suitable for single-journey long-distance railway scenarios where traditional closed-loop detection is not feasible, improving the generation efficiency and accuracy of railway electronic maps. Attached Figure Description

[0043] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0044] To more clearly illustrate the technical solutions in this application 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 some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is a flowchart illustrating the high-precision railway map generation method based on single-journey trajectory correction provided in this application.

[0046] Figure 2 This is one of the structural schematic diagrams of the high-precision railway map generation system based on single-journey trajectory correction provided in this application.

[0047] Figure 3 This is an architecture diagram of the high-precision railway map generation system based on single-journey trajectory correction provided in this application.

[0048] Figure 4 This is the second schematic diagram of the high-precision railway map generation system based on single-journey trajectory correction provided in this application.

[0049] Figure 5 This is a flowchart of the method for generating a high-precision railway map based on single-journey trajectory correction provided in this application.

[0050] Figure 6 This is a schematic diagram of the structure of the electronic device provided in this application. Detailed Implementation

[0051] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0052] It should be noted that the terms "first," "second," etc., used in this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that comprises a series of steps, units, or modules is not necessarily limited to those explicitly listed, but may include other steps, units, or modules not explicitly listed or inherent to such processes, methods, products, or devices.

[0053] When constructing high-precision maps, the relevant technologies can be divided into two categories: traditional methods such as aerial surveying using mobile surveying vehicles or drones, and non-traditional methods such as remote sensing imagery and trajectory data production. These technologies involve multiple key technologies such as multi-source heterogeneous data fusion, Simultaneous Localization and Mapping (SLAM), large-scale artificial intelligence models, and knowledge graphs.

[0054] While mobile surveying vehicles and drone aerial surveying can provide a certain level of accuracy, they require significant time and resources, especially on long-distance railways with complex terrain. Furthermore, these methods have limitations in data fusion and processing, particularly when fusing multi-source heterogeneous data. Integrating data from different sensors struggles to achieve a unified representation, thus impacting map accuracy and system robustness.

[0055] While SLAM technology has made progress in autonomous driving and robot navigation, its application in railway map building remains challenging. For example, visual SLAM performs poorly in railway tunnel environments with insufficient lighting or missing features, and while laser SLAM can generate high-resolution point cloud maps, it is costly and slow. Furthermore, existing SLAM technologies often suffer from inter-frame registration problems when dealing with high-speed moving scenes, affecting the system's stability and accuracy.

[0056] Meanwhile, while large-scale AI models can significantly improve environmental perception capabilities, efficiently integrating and processing massive amounts of data in the railway environment, and optimizing models for complex and ever-changing railway scenarios, remain major challenges. Furthermore, although knowledge graphs possess rich semantic information and decision support potential, effectively utilizing this knowledge and its dynamic updates and maintenance in railway map construction is also a technological challenge.

[0057] In order to address the shortcomings of related technologies, Figure 1 This is a flowchart illustrating the high-precision railway map generation method based on single-journey trajectory correction provided in this application, as shown below. Figure 1 As shown, the method includes steps 110, 120, 130 and 140.

[0058] Step 110: Simultaneously acquire continuous image sequences and spatial attitude data of the train during its current movement.

[0059] Specifically, the method provided in this application is implemented by a high-precision railway map generation system. This system can be implemented in software, such as a high-precision railway map generation program running on a computer or server; or it can be implemented in hardware, such as a mobile terminal, computer, or server that executes the high-precision railway map generation method.

[0060] The method provided in this application is used to install a binocular camera and an inertial measurement unit (IMU) on a current train, with the IMU positioned at the head of the train. The current train is one that builds a railway electronic map in real time during its journey.

[0061] A continuous image sequence is a series of images captured in real time by a stereo camera while the train is in motion, resulting in images that are consecutive in time. It can be understood that the images acquired by the stereo camera include left-eye and right-eye images.

[0062] Spatial attitude data consists of the three-axis velocities and angular velocities collected by the inertial measurement unit (IMU) during train operation. The IMU comprises three single-axis accelerometers and three single-axis gyroscopes. The accelerometers detect the acceleration signals of the object along the three independent axes of the carrier coordinate system, while the gyroscopes detect the angular velocity signals of the carrier relative to the navigation coordinate system. By measuring the object's angular velocity and acceleration in three-dimensional space, the object's attitude can be calculated.

[0063] During the current train's operation, continuous image sequences and spatial attitude data are collected simultaneously.

[0064] Step 120: Determine the pose change data of the current train based on visual feature points in the continuous image sequence.

[0065] Specifically, the ORB (Oriented FAST and Rotated BRIEF) algorithm is an algorithm that combines FAST corner detection and BRIEF feature descriptors. The ORB algorithm can be used to extract visual feature points from continuous image sequences. By matching these extracted visual feature points and comparing their positional changes, the camera's pose change data in space can be inferred. Pose includes both position and orientation. Since the camera is fixed at the front of the train, the camera's pose change data is actually also the train's pose change data.

[0066] Step 130: Perform synchronous positioning and map construction based on pose change data and spatial attitude data to generate an initial electronic map.

[0067] Specifically, spatial attitude data is data collected by the inertial measurement unit. The camera's motion trajectory, including its position and attitude (orientation), can be obtained through integration.

[0068] By optimizing the algorithm, pose change data (visual information) and spatial attitude data (inertial information) can be effectively integrated. Through the SLAM algorithm in related technologies, the accuracy and robustness of localization and mapping can be improved, thereby outputting an initial electronic map.

[0069] Step 140: Based on the mileage information and image position corresponding to the preset markers in the continuous image sequence, correct the mileage and trajectory of the current train in the initial electronic map, and generate the electronic map corresponding to the current train during its journey.

[0070] Specifically, considering that trains typically travel long distances in one direction and do not have a closed-loop path, closed-loop detection in related technologies cannot be effectively implemented.

[0071] The method provided in this application can identify preset markers in a continuous image sequence. Preset markers may include road signs such as kilometer markers, half-kilometer markers, hundred-meter markers, tunnel entrances, and tunnel exits. These road signs are typically used to represent fixed distances in a railway environment, have corresponding mileage information, and can serve as reference points to assist in the positioning process. Furthermore, the image positions of the preset markers can be obtained through continuous image sequence identification. By comparing the image positions with the physical dimensions of the preset markers, the camera pose can be calculated, thus obtaining the train pose.

[0072] The mileage of the current train in the initial electronic map can be corrected by using the mileage information corresponding to the preset markers, and the trajectory of the current train in the initial electronic map can be corrected by using the image position corresponding to the preset markers. This effectively eliminates the cumulative error generated during long-distance train travel and ultimately generates the electronic map corresponding to the current train during its journey.

[0073] The high-precision railway map generation method based on single-journey trajectory correction provided in this application simultaneously collects continuous image sequences and spatial attitude data of the current train during its journey; determines the pose change data of the current train based on visual feature points in the continuous image sequence; performs synchronous positioning and map construction based on the pose change data and spatial attitude data to generate an initial electronic map; and corrects the mileage and trajectory of the current train in the initial electronic map based on the mileage information and image position of preset markers in the continuous image sequence to generate the corresponding electronic map of the current train during its journey. Because it combines continuous image sequences collected by a binocular camera with spatial attitude data collected by an inertial measurement unit, it achieves high-precision positioning and map construction in a railway environment. This method can more accurately handle positioning and mapping problems under high-speed train movement, improving the stability and accuracy of electronic map construction. By using the mileage information and image position of preset markers in the railway environment to correct the mileage and trajectory of the current train in the initial electronic map, it effectively reduces the cumulative error during long-distance travel. It is particularly suitable for single-journey long-distance railway scenarios where traditional closed-loop detection is not feasible, thus improving the generation efficiency and accuracy of railway electronic maps.

[0074] It should be noted that each implementation method of this application can be freely combined, rearranged, or executed individually, and does not need to rely on or depend on a fixed execution order.

[0075] In some embodiments, after simultaneously acquiring continuous image sequences and spatial attitude data of the current train during its operation, the method further includes:

[0076] Preprocessing a continuous image sequence; the preprocessing includes at least one of distortion correction, denoising, and image enhancement.

[0077] Based on the timestamps of the left and right images in a continuous image sequence, the binocular images in the continuous image sequence are filtered;

[0078] Spatial pose data is synchronized and calibrated based on the timestamps of stereo images in a continuous image sequence.

[0079] Specifically, the continuous image sequence acquired by the binocular camera can be preprocessed, for example, by using preset distortion parameters to correct the image, and then to perform noise reduction and image enhancement.

[0080] The images in a continuous image sequence include a pair of left and right eye images acquired together. The timestamps of the paired left and right eye images are compared. If the difference between the two timestamps exceeds a preset difference, it indicates that the two images were not acquired synchronously and can be discarded. If the difference between the two timestamps is less than or equal to the preset difference, it indicates that the two images were acquired synchronously and can be retained.

[0081] For the spatial attitude data acquired by the inertial measurement unit, the timestamps of the spatial attitude data can be compared with the timestamps of the stereo images in the continuous image sequence for synchronization and calibration, so that the stereo images and spatial attitude data can be aligned on the time axis. At the same time, the acceleration and angular velocity in the spatial attitude data are initialized with zero bias using the random error in the configuration file, and the processed data is available for subsequent use.

[0082] The high-precision railway map generation method based on single-journey trajectory correction provided in this application ensures the accuracy and consistency of the data and improves the accuracy of railway electronic maps by preprocessing and correcting images captured by binocular cameras and synchronizing and calibrating spatial attitude data.

[0083] In some embodiments, synchronous localization and map building are performed based on pose change data and spatial attitude data to generate an initial electronic map, including:

[0084] The state vector is determined based on the actual measurement values ​​of each sensor; the sensors include a binocular camera that acquires continuous image sequences and an inertial measurement unit that acquires spatial attitude data;

[0085] Based on the actual and expected calculated values ​​of each sensor at each time point, as well as the covariance matrix of the actual measured values, the state estimation cost function is determined.

[0086] The state estimation cost function is optimized to obtain the optimal estimate of the state vector of the train during its current operation.

[0087] Synchronous positioning and map building are performed based on the optimal estimate of the state vector to generate an initial electronic map.

[0088] Specifically, when a train travels through a tunnel, the number of extracted feature points drops sharply due to deteriorating lighting conditions, potentially leading to localization failure. In this environment where visual information is limited, the key to optimizing the SLAM algorithm lies in increasing the system's reliance on non-visual sensor data and enhancing the algorithm's efficiency in utilizing existing features.

[0089] This application provides a SLAM optimization method for environments with limited visual information. This method supports combining visual data and IMU data, treating each sensor's measurement as a factor. Factors sharing the same state variable are aggregated into a state vector. Through state optimization, accurate estimation of the train's state is achieved. The method includes:

[0090] Step 1: System State Definition

[0091] The state vector is determined based on the actual measurement values ​​of each sensor; the sensors include a binocular camera that acquires continuous image sequences and an inertial measurement unit that acquires spatial attitude data.

[0092] During the system (train) positioning process, the system aims to estimate its precise position in three-dimensional space. ) and posture ( In addition, it is necessary to estimate the depth of visual landmarks observed by the camera. ), that is, the straight-line distance from the camera to each detected feature point, and the motion variables generated by the IMU: velocity ( ), acceleration bias and gyroscope bias .

[0093] State vector It typically includes dynamic parameters such as position, attitude, and velocity, used to describe the instantaneous state of the system in the environment:

[0094]

[0095] For visual systems, indexes This represents a specific step in the time series, each step corresponding to a set of data collected from sensors, which is used to calculate and update the train's position and attitude. State vector This represents the depth value of each feature point in the initial frame (the initial image in a continuous image sequence). For inertial measurement units (inertial sensors), the state vector... This includes the velocity vector. and two sensor bias parameters and .

[0096] Step 2: Definition of State Estimation Cost Function and Weight Adjustment

[0097] The state estimation problem is constructed as a maximum likelihood estimation (MLE) problem within this framework, aiming to solve it by optimizing the joint probability distribution of the pose over a time period. In this process, it is assumed that all measurements are independent and that the measurement uncertainties follow a Gaussian distribution. Based on these assumptions, the problem can be further transformed into a nonlinear least squares problem, namely, bundle adjustment (BA).

[0098] The state estimation cost function can be determined based on the actual and expected calculated values ​​of each sensor at each time point, as well as the covariance matrix of the actual measurements, as follows:

[0099]

[0100] in, Represents the state vector The optimal estimate is aimed at minimizing the error of the entire system. It is a collection of measurements from cameras, IMUs, and other sensors; It is the observation vector, which contains the actual measurement data (actual measurement value) obtained from the sensor. It is a time index used to indicate a specific point in time in a state or measurement; It is a set of measured values A specific measurement value; Indicates in At any given moment, the system acquires observation data from each sensor; Representing observation data The covariance matrix quantifies the uncertainty of the observed data; It is a sensor model. express At any given moment, the expected values ​​of each sensor are calculated from the system's current attitude and position (expected calculated values).

[0101] When a train enters a tunnel, ambient light decreases rapidly, leading to a sharp reduction in the number of extractable feature points. This reduces the reliability of visual information. Therefore, it's necessary to appropriately increase the weight of IMU information in the SLAM process to reduce localization errors caused by insufficient visual information. The states of the train entering and exiting a tunnel can be determined through environmental feature detection. In the cost function, the measurements from each sensor appear as residuals, and their importance is determined by the corresponding covariance matrix. To enhance the influence of IMU data and reduce the influence of camera data under specific conditions, these covariance matrices can be adjusted. Specifically, reducing the covariance matrix of IMU measurements effectively reduces the uncertainty weight of IMU data, thus giving it greater influence in the optimization process. Conversely, increasing the covariance matrix of camera measurements increases the uncertainty weight of camera data, correspondingly reducing its relative importance in state estimation.

[0102] In mathematical terms, the cost function can be expressed as:

[0103]

[0104] in, This represents the matrix transpose operation; and These represent the measurement sets of the camera and IMU sensors, respectively. and This is the covariance matrix corresponding to the sensor measurements. By adjusting these covariance matrices according to the ratio of extracted feature points to the preset number of feature points to be extracted, the influence of different sensor data on the optimization process can be effectively controlled.

[0105] Step 3: Definition of Sensor Factors

[0106] The camera factor represents the stereo camera. Corner features are detected in each camera frame, and these features are tracked across frames (single images) using a KLT (Kanade-Lucas-Tomasi) tracker. Based on feature associations, a camera factor is constructed for each feature in each frame.

[0107]

[0108]

[0109] in, Indicates in Features in the image at time 1 The observed pixel coordinates; Indicates based on system state ,exist Features in the image at time 1 The predicted pixel coordinates; It is in time The camera rotation matrix (i.e., a specific starting frame); and It is in time The camera rotation matrix (i.e., for a subsequent frame). These matrices describe the rotation from the world coordinate system to the camera coordinate system. It is in time The camera position; and It is in time The camera position; express The distance from the camera to the current feature point at any given time; It is a camera projection function used to map three-dimensional points in the camera coordinate system to two-dimensional points in the image plane; and They represent time respectively and The transformation matrix of the camera's pose (rotation and translation). It is the external transformation matrix from the system center to the camera center, which is pre-calibrated offline; , Is Features in the image at time step The initial observed pixel x and y coordinates, Is The horizontal and vertical coordinates of the same feature pixels observed at any given time.

[0110] The IMU factor uses a known IMU pre-integration algorithm, assuming that the additive noise of the acceleration and gyroscope measurements is Gaussian white noise.

[0111]

[0112] in, Indicates time arrive The IMU measurement vectors between them include the relative position, velocity, and direction changes of the integral. Indicates from state The obtained IMU prediction vector. , , These represent the relative position, velocity, and rotation measurements integrated from the IMU, respectively. express The rotation matrix of the system at time t. and They are respectively and The position of the time system in three-dimensional space. It represents the acceleration due to gravity. and They represent and The speed of the time system. express and The time interval between them. and They represent and Deviation in acceleration readings at any given time. and They represent and Deviation in the gyroscope reading at any given time.

[0113] Step 4: Optimize the solution

[0114] The cost function is solved using the Gauss-Newton method. The cost function is relative to the state. initial guess value Linearization yields the following result:

[0115]

[0116] in, It is a state vector An increment is given, and the optimization process attempts to find this increment that minimizes the total sum of squared residuals. It is in time and measurement The residual vector below represents the predicted state. and actual observation The error between them. It is a residual Regarding the status The Jacobian matrix is ​​the partial derivative of the residual with respect to the state vector. It is a measurement In time The covariance matrix is ​​used to weight the residuals during the minimization process to ensure that observation uncertainties are properly accounted for.

[0117] After linear approximation, the cost function is obtained. The closed solution is obtained as follows:

[0118]

[0119] in, Representing the Hessian matrix, this matrix is ​​the sum of all Jacobian matrices. The weighted sum of the second-order partial derivatives with respect to the prediction error. This vector is a cumulative term that includes the weighted residuals of all time steps and measurements.

[0120] Then, the solution obtained Used to update the current state ,pass (in The update is performed using an addition operation on a rotated manifold. This process iterates multiple times until convergence.

[0121] Step 5: Marginalization

[0122] To control computational complexity, historical measurements are converted into prior terms through marginalization, preserving past information without loss. The marginalization operation is performed using Shure complements, converting information about the marginalized state into prior terms.

[0123]

[0124] , , , These symbols represent information matrices. The different parts, of which: It is a sub-block of the information matrix related to the state variables that are about to be marginalized. and ( The transposes of the variables represent the cross-information matrix portions between the marginalized state variables and the retained state variables, respectively. It is a sub-block of the information matrix associated with the retained state variables. This represents the increment of the retained state variable, which is the variable that needs to be solved during the optimization process. , Represents information vector Different parts, among which It is the part related to the state variables that are about to be marginalized, and It is the part related to preserving state variables. It is a new information matrix obtained through Schul complement, used to update the optimization problem of the preserved state. It is the updated information vector used to solve optimization problems while preserving the state.

[0125] In this way, a new prior term about the remaining state will be obtained. This includes information about the marginalized state without any loss of information. The system maintains a fixed number of spatial camera frames. When a new keyframe (the filtered image) arrives, visual and inertial factors associated with the first frame are marginalized.

[0126] After obtaining prior information about the current state using Bayes' theorem, the posterior is calculated as the product of the likelihood and the prior: Next, state estimation becomes a maximum a posteriori (MAP) problem. The state is preserved within a sliding window from time [time value missing]. At the time The state. At any given moment. The previous state is marginalized and transformed into a priori term. Therefore, the MAP problem is written as:

[0127]

[0128] +( )

[0129] Finally, the final state can be obtained by using a nonlinear optimizer, which yields the optimal estimate of the train's state vector during its current operation. Based on this optimal estimate, synchronous positioning and map construction are performed to generate an initial electronic map.

[0130] The high-precision railway map generation method based on single-journey trajectory correction provided in this application achieves accurate estimation of train status by fusing binocular video and IMU data into an optimized framework. This process involves state definition, cost function establishment, sensor factor construction, optimization process design, and edge-shifting processing, ensuring the system's efficiency and flexibility. Through appropriate edge-shifting and optimization strategies, this application can effectively retain support for different sensor configurations. When the train is in a low-light environment, utilizing inertial measurement unit (IMU) data can significantly improve the robustness of the SLAM system. The acceleration and angular velocity information provided by the IMU can be used to estimate the short-term motion state, helping to bridge the gaps in visual information. By performing precise pre-integration processing on the IMU data and tightly fusing it with visual data, the continuity and accuracy of positioning can be maintained even when visual features are scarce.

[0131] In some embodiments, based on the mileage information and image position corresponding to preset markers in a continuous image sequence, the mileage and trajectory of the current train in the initial electronic map are corrected to generate an electronic map corresponding to the current train during its operation, including:

[0132] Perform marker recognition on a continuous image sequence to determine the target type of each preset marker in the continuous image sequence;

[0133] Based on the target type of each preset marker, determine the mileage information corresponding to each preset marker;

[0134] The mileage information corresponding to each preset marker is compared with the measured mileage of the current train, and the mileage of the current train in the initial electronic map is corrected based on the mileage deviation between the mileage information and the measured mileage.

[0135] Specifically, trains accumulate errors during long-distance travel, including mileage errors and trajectory errors. Precise mileage correction can be achieved using pre-set markers on the railway, such as kilometer markers, half-kilometer markers, and 100-meter markers (each kilometer marker represents a fixed point along the line, with intervals typically 1 kilometer; the same applies to half-kilometer markers and 100-meter markers) and the train's own odometer.

[0136] When a train passes a kilometer marker and is successfully identified, the system compares the distance measured by the odometer with the actual fixed distance (usually 1 kilometer). If a discrepancy exists—that is, the distance measured by the odometer does not match the actual distance between the kilometer marker and the actual distance—the system will adjust the train's real-time position based on the discrepancy value, thereby correcting the mileage error. The mileage correction steps include:

[0137] Step 1: Marker Identification

[0138] Integrated neural network models (such as the YOLO model) can be used to identify markers in continuous image sequences, determining the target type of each preset marker in the sequence. Target types can include kilometer markers, half-kilometer markers, 100-meter markers, etc.

[0139] Step 2: Obtaining the actual location

[0140] Each preset marker can be considered a fixed point. For each identified fixed point, the system can obtain its corresponding mileage information, which represents the actual distance along the route. For example, a kilometer marker represents a distance of 1 kilometer. Let the first... The actual positions of the fixed points are .

[0141] Step 3: Mileage Data Comparison

[0142] The train's odometer provides the current measured mileage, which is the distance. The system compares the distance measured by the odometer with the actual distance to the identified fixed point. Let's assume the train passes the [number missing]... At a fixed point, the distance measured by the odometer is .

[0143] Step 4: Mileage Correction

[0144] Suppose the train passes through the first At a fixed point, the distance measured by the odometer is The actual distance to that fixed point is (For example, 1 kilometer). Mileage deviation It can be calculated using the following formula: .

[0145] If there is a deviation Therefore, the real-time position of the train needs to be corrected. Let the position of the train before adjustment be... The adjusted position is The correction can be performed using the following formula: .

[0146] in, Based on bias The calculated correction vector is obtained by multiplying the normalized deviation vector by a correction factor, i.e.: × Correction factor. The correction factor can be set according to the actual situation.

[0147] Utilizing new real-time locations The system can optimize historical trajectories. Let the position sequence in the historical trajectory be... , , ..., The system can determine the deviation at each position. Adjustments will be made.

[0148] A weighted average algorithm is used to optimize historical trajectories and reduce the impact of accumulated errors. For example, the mathematical expression for optimizing historical trajectories using a weighted average method can be:

[0149]

[0150] in, This is the optimized position. It is the deviation after being averaged over each key position.

[0151] The high-precision railway map generation method based on single-journey trajectory correction provided in this application accurately corrects the train position by recognizing the features of preset landmarks at fixed distances in real time and combining them with odometer data, effectively eliminating the cumulative errors generated during long-distance travel.

[0152] In some embodiments, based on the mileage information and image position corresponding to preset markers in a continuous image sequence, the mileage and trajectory of the current train in the initial electronic map are corrected to generate an electronic map corresponding to the current train during its operation, including:

[0153] Marker identification is performed on a continuous image sequence to determine the image position of each preset marker in the continuous image sequence;

[0154] Based on the image position and spatial coordinate position of each preset marker, as well as the parameters of the stereo camera that acquires a continuous image sequence, the actual pose information of the stereo camera is determined.

[0155] Based on the actual pose information of the binocular camera, the trajectory of the current train in the initial electronic map is corrected.

[0156] Specifically, trajectory errors can also be corrected by identifying preset markers. Similar to odometry correction, when identifying each preset marker, for each identified fixed point, the system obtains its position information in the image and its corresponding category label. Let the image coordinates of the identified fixed point be... ,in This indicates the serial number of the identified fixed point.

[0157] Once a preset marker is identified, the system will obtain the position of the preset marker in the image and calculate the camera pose based on the physical size and known position of the preset marker.

[0158] Assuming the spatial coordinates (world coordinates) of the preset marker are: The camera's intrinsic parameter matrix is The projection of the logo in the image is Based on the relationship between camera projection and world coordinates, the camera pose can be calculated using the following formula:

[0159]

[0160]

[0161] in, and These represent the camera's rotation matrix and translation vector, respectively. It is the camera's intrinsic parameter matrix. It is the mark at the first Projection in a frame image.

[0162] Based on the identified marker pose, the system can correct the actual pose of the camera. Let... and The pose before correction. and The corrected pose. Pose correction can be performed using the following formula:

[0163]

[0164] in, and It is the corrected rotation and translation. The rotation and translation are obtained from the recognition of the markers. This is a possible deviation.

[0165] After pose correction, the system needs to update the trajectory. Let... For keyframes (filtered images) To the keyframe (the filtered image) Relative transformation, This is the corrected relative transformation.

[0166] Trajectory updates can be performed using the following formula:

[0167]

[0168] here, yes inverse transform, It is a correction transformation obtained from mark recognition.

[0169] Through the steps described above, the system can correct its trajectory and pose by recognizing specific environmental landmarks, without relying on traditional closed-loop detection. This method effectively reduces accumulated errors and improves the system's positioning accuracy and robustness.

[0170] The high-precision railway map generation method based on single-journey trajectory correction provided in this application corrects the trajectory of the current train in the initial electronic map according to the actual pose information of the binocular camera, effectively eliminating the cumulative error generated during long-distance travel.

[0171] In some embodiments, marker identification of a continuous image sequence includes:

[0172] Input a continuous image sequence into the marker recognition model to obtain the target type, image location, and confidence level of each preset marker output by the marker recognition model;

[0173] The marker recognition model is trained based on sample images corresponding to each preset marker.

[0174] Specifically, mileage correction and trajectory correction mainly rely on identifying pre-set markers along the railway line, such as kilometer markers, half-kilometer markers, 100-meter markers, tunnel entrances, and other fixed structures. These structures, due to their uniqueness and equidistant nature, serve as highly reliable positioning reference points.

[0175] A landmark recognition model can be constructed using a neural network model. This model identifies the target type, image location, and confidence level of each preset landmark. The neural network model can be a YOLO (You Only Look Once) model. Preset landmark target types include kilometer markers, half-kilometer markers, 100-meter markers, and tunnel entrance markers, etc. Specific steps include:

[0176] Step 1: Data Collection and Labeling

[0177] Collect sample images corresponding to various preset markers. For example, collect a large amount of image or video data of railway line markers, especially images containing kilometer markers, half-kilometer markers, 100-meter markers, and tunnel entrances. Then, manually label these data, indicating the type and location of each feature, to prepare a training set for training the YOLO model.

[0178] To ensure the model's generalization ability, the collected image and video data should cover different road segments. Therefore, map creation should be carried out during periods of good weather and high visibility. Furthermore, to improve recognition accuracy, high-resolution images should be used and taken from multiple angles and positions, including bird's-eye views, side views, and frontal views.

[0179] In the data preprocessing stage, rigorous data cleaning is performed to remove blurry, overexposed, and poor-quality data. Simultaneously, to facilitate processing, images are cropped to remove irrelevant backgrounds and scaled to a uniform size. Furthermore, data augmentation techniques such as rotation and flipping are used to further enhance the model's adaptability to different situations.

[0180] The identification tags are divided into four categories: 0 km marker, 1 half km marker, 200 m marker, and 3 tunnel entrance marker.

[0181] After labeling, the complete dataset is divided into training, validation, and test sets in a ratio of 8:1:1 for use in subsequent model training and evaluation.

[0182] By following the steps above, a high-quality dataset of railway landscapes can be constructed, laying a solid foundation for training the YOLO model.

[0183] Step 2: Train the YOLO object detection model

[0184] A YOLO object detection model was trained using a labeled dataset, aiming to enable the model to identify and locate railway landmarks in images. As an end-to-end deep learning model, YOLO can predict the location and category of objects in an image with a single forward pass, giving it significant advantages in speed and accuracy, making it particularly suitable for applications requiring real-time processing.

[0185] Step 3: Real-time identification of preset markers

[0186] The trained YOLO model is integrated into the train's SLAM system. During train operation, cameras capture the surrounding environment in real time, generating a continuous sequence of images. These image data, after preprocessing (including image resizing and pixel value normalization), are used as input to the YOLO model, enabling real-time identification of distinctive railway features.

[0187] For each frame of the image, the YOLO model calculates the prediction result through forward propagation. The output of the YOLO model can be represented as a three-dimensional tensor of S×S×(5+C), where S is the grid size, C is the number of classes, and 5 represents the four coordinates and confidence score of each bounding box. Assuming the input image has dimensions W×H, where W is the width and H is the height, the YOLO model divides it into S grids, each grid responsible for predicting targets whose center point falls within that grid. For each grid, the model predicts B bounding boxes and the corresponding confidence score and class probability for each bounding box.

[0188] For each predicted bounding box, the YOLO model outputs its coordinates (x, y, w, h) in the image, where x and y are the image center coordinates of the predefined marker, w is the image width of the predefined marker, and h is the image height of the predefined marker. These coordinates are normalized relative to the grid size. These coordinates can be converted to their actual coordinates in the image using the following formula: True coordinates = Normalized coordinates × (x / S, H / S). The model also outputs a confidence score for each bounding box. and category probability The confidence score indicates the model's confidence that the predicted bounding box contains the target, while the class probability indicates the model's confidence that the target belongs to a specific class.

[0189] For each detected target, the SLAM system will receive the following information:

[0190] Target type: Based on the highest category probability Defined target category.

[0191] Image position: The position of the target in the image determined by the transformed bounding box coordinates (x, y, w, h).

[0192] Confidence score: The confidence score of the model for detecting this target. It is used to assess the reliability of the test results.

[0193] The SLAM system integrates feature information returned by the YOLO model into the train localization and map building process. It uses identified railway landmarks to help trains determine their location within the railway network and update map information along the railway lines.

[0194] The high-precision railway map generation method based on single-journey trajectory correction provided in this application uses deep learning models (such as YOLO) to identify specific landmark features in the railway environment (such as kilometer markers, tunnel entrances, etc.), providing rich reference information for subsequent positioning and mapping.

[0195] In some embodiments, after correcting the mileage and trajectory of the current train in the initial electronic map based on the mileage information and image position corresponding to preset markers in a continuous image sequence, and generating the electronic map corresponding to the current train during its operation, the method further includes:

[0196] Determine the latitude and longitude coordinates of the current train's starting point on the electronic map;

[0197] Based on the latitude and longitude coordinates of the starting point and the relative displacement data of each trajectory point of the current train, the latitude and longitude coordinates of each trajectory point are determined.

[0198] Based on the latitude and longitude coordinates, trajectory point number, road segment name, road segment type, and road segment length of each trajectory point, a key point table, road segment table, and road segment details table corresponding to the electronic map are generated.

[0199] Specifically, the generated electronic map can be post-processed to further refine and optimize the map information to output the final target format. The system's raw output data mainly includes information such as timestamps, three-axis relative displacement, and system pose. However, in order to construct a detailed railway map, it is also necessary to integrate key geographic information such as latitude and longitude, kilometer markers, and node types.

[0200] Precise starting and ending point coordinates can be obtained through configuration files. To obtain the most accurate coordinate data, RTK (Real-Time Kinematic) technology can be used for point marking. Once the starting point is determined, the system will combine the relative displacement data in the x and y directions and calculate the latitude and longitude coordinates of subsequent trajectory points using an integral method, ensuring the continuity and accuracy of map location information. Furthermore, the system can identify and mark different types of nodes, such as starting points, ending points, kilometer markers, half-kilometer markers, 100-meter markers, tunnel entrances, and tunnel exits. In this way, the map optimization and generation module can transform the raw trajectory data into a detailed and accurate railway map, which will be output in a target format for use. The target format can include a key point table, a section table, and a section detail table, as shown in Tables 1 to 3.

[0201] Table 1 Key Points Table of Electronic Map

[0202]

[0203] Table 2 Electronic Map Road Segment Table

[0204]

[0205] Table 3. Electronic Map Road Segment Details

[0206]

[0207] The high-precision railway map generation method based on single-journey trajectory correction provided in this application integrates key geographic information such as latitude and longitude, kilometer markers, and node types into the generated electronic map, thereby improving the usability and accuracy of the railway electronic map.

[0208] The system provided in the embodiments of this application is described below. The system described below can be referred to in correspondence with the method described above.

[0209] Figure 2 This is one of the structural schematic diagrams of the high-precision railway map generation system based on single-journey trajectory correction provided in this application, such as... Figure 2 As shown, the system includes:

[0210] The data acquisition module 210 is used to simultaneously acquire continuous image sequences and spatial attitude data of the train during its current operation.

[0211] The feature extraction module 220 is used to determine the pose change data of the current train based on visual feature points in a continuous image sequence;

[0212] The synchronous mapping module 230 is used to perform synchronous positioning and map construction based on pose change data and spatial attitude data, and generate an initial electronic map.

[0213] The map correction module 240 is used to correct the mileage and trajectory of the current train in the initial electronic map based on the mileage information and image position corresponding to preset markers in the continuous image sequence, and generate the corresponding electronic map of the current train during its operation.

[0214] The high-precision railway map generation system based on single-journey trajectory correction provided in this application embodiment simultaneously collects continuous image sequences and spatial attitude data of the current train during its journey; determines the pose change data of the current train based on visual feature points in the continuous image sequence; performs synchronous positioning and map construction based on the pose change data and spatial attitude data to generate an initial electronic map; and corrects the mileage and trajectory of the current train in the initial electronic map based on the mileage information and image position corresponding to preset markers in the continuous image sequence to generate an electronic map corresponding to the current train during its journey. Because it combines continuous image sequences collected by a binocular camera with spatial attitude data collected by an inertial measurement unit, it achieves high-precision positioning and map construction in a railway environment. This system can more accurately handle positioning and mapping problems under high-speed train movement, improving the stability and accuracy of electronic map construction. By using the mileage information and image position of preset markers in the railway environment to correct the mileage and trajectory of the current train in the initial electronic map, it effectively reduces the cumulative error during long-distance travel. It is particularly suitable for single-journey long-distance railway scenarios where traditional closed-loop detection is not feasible, improving the generation efficiency and accuracy of railway electronic maps.

[0215] Figure 3 This is an architecture diagram of the high-precision railway map generation system based on single-journey trajectory correction provided in this application, such as... Figure 3 As shown, the system uses multiple sensors to collect environmental information. It acquires forward-sequential images of the train using a binocular camera, while simultaneously using the IMU (Integrated Measurement Unit) on the camera to collect the train's three-axis angular velocity and linear acceleration. For the image information, after preprocessing, the system extracts feature points from the left and right cameras, continuously tracks the train, and removes unsuitable points. For the IMU data, it aligns with gravity, obtains the initial IMU attitude, and continuously updates the pose using median integration. After completing the tight coupling initialization of the binocular camera and IMU, it constructs a least-squares problem using prior information, the residuals of the left and right cameras, the IMU residuals, and the current environmental factors to solve for the pose. When a kilometer marker is detected, the system optimizes the historical pose, and the optimized result is post-processed to form a complete map.

[0216] Figure 4This is the second schematic diagram of the high-precision railway map generation system based on single-journey trajectory correction provided in this application, as shown below. Figure 4 As shown, the system can be divided into 6 modules, and the specific modules and their functions are as follows:

[0217] (1) Data acquisition module, installed at the head of the train, includes a binocular camera and an IMU. During train operation, the binocular camera continuously acquires binocular video, and the IMU acquires the three-axis speed and angular velocity of the train in real time, and transmits these data to the central processing unit.

[0218] (2) The data preprocessing module processes the binocular video data and IMU data separately. For binocular images, the system uses preset distortion parameters for image correction, followed by denoising and image enhancement. Simultaneously, it compares the timestamps of the left and right eye images; if the difference exceeds a threshold, the image is discarded. For IMU data, it needs to be compared with the timestamps of the video data for synchronization and calibration. At the same time, it uses random errors from the configuration file to initialize acceleration and angular velocity with zero bias. The processed data is then used for subsequent applications.

[0219] (3) The feature extraction and analysis module is mainly divided into two parts: target environment feature detection and specific image information enhancement. For target environment feature detection, the system uses a YOLO pre-trained model to identify key landmarks with semantic information along the railway line, such as kilometer markers, half-kilometer markers, 100-meter markers, and tunnel entrances. Through accurate detection of these features, the system can obtain precise geographical location information, providing an important basis for subsequent map generation. For specific image information enhancement, the system employs a series of image processing techniques to improve the quality of image data, thereby ensuring stable operation even in visually limited environments. This includes image contrast adjustment, sharpening, and noise reduction to enhance useful information and suppress noise. Through these enhancement measures, the system can improve the stability of its self-localization and mapping. The processed feature information will be sent to the next module for further data analysis and map construction.

[0220] (4) The SLAM core processing module performs system localization and mapping by fusing binocular video data and IMU data. Visual features extracted from the binocular video data are matched, and changes in the positions of these feature points are compared to infer the camera's pose in space. Simultaneously, acceleration and angular velocity information provided by the IMU data are used to calculate the camera's trajectory through integration. These trajectories can be used to estimate the camera's pose, including its position and orientation. Through algorithm optimization, visual and inertial information are effectively integrated to improve the accuracy and robustness of localization and mapping, thus outputting preliminary mapping results.

[0221] (5) The trajectory correction and error elimination module is an alternative to traditional SLAM loop closure detection. Considering that trains typically travel long distances in one direction without path closure, traditional loop closure detection is ineffective. This system utilizes fixed-distance features in the railway environment, such as kilometer markers, half-kilometer markers, and 100-meter markers, as reference points to assist the positioning process. By matching these fixed positional features, the odometer error can be corrected, and historical pose can be corrected. This method overcomes the limitations of traditional SLAM loop closure detection in long-distance one-way travel, improving the accuracy and reliability of the generated trajectory.

[0222] (6) The map optimization and generation module further improves and optimizes the information of the generated map by post-processing it in order to output the final target format.

[0223] Figure 5 This is a flowchart of the high-precision railway map generation system based on single-journey trajectory correction provided in this application, such as... Figure 5 As shown, this system integrates modules for binocular camera and IMU data acquisition, preprocessing, feature extraction, SLAM processing, trajectory correction, and map generation and optimization, achieving fully automated processing of railway electronic maps. Through real-time data transmission from the data acquisition module and precise processing by the preprocessing module, the system effectively acquires key information during train operation. The feature extraction and analysis module improves the robustness of positioning and environmental awareness through binocular vision and environmental feature detection. The SLAM core processing module combines visual and inertial information for positioning and mapping, the trajectory correction module uses fixed-distance features to correct positioning errors, and finally, map optimization using a global BA algorithm improves map accuracy and stability. These components together constitute a complete system, significantly improving the efficiency and accuracy of railway electronic map generation. The system's advantage lies in its full utilization of multi-sensor data, effectively addressing the challenges of long-distance single-journey train travel and providing reliable support for safe railway operation.

[0224] Figure 6 This is a schematic diagram of the structure of the electronic device provided in this application, such as... Figure 6 As shown, the electronic device may include a processor 610, a communications interface 620, a memory 630, and a communications bus 640, wherein the processor 610, the communications interface 620, and the memory 630 communicate with each other via the communications bus 640. The processor 610 can call logical commands stored in the memory 630 to execute the methods described in the above embodiments, for example:

[0225] Simultaneously, continuous image sequences and spatial attitude data of the current train during its operation are collected; the pose change data of the current train is determined based on visual feature points in the continuous image sequence; synchronous positioning and map construction are performed based on the pose change data and spatial attitude data to generate an initial electronic map; based on the mileage information and image position corresponding to preset markers in the continuous image sequence, the mileage and trajectory of the current train in the initial electronic map are corrected to generate the corresponding electronic map of the current train during its operation.

[0226] Furthermore, the logical commands in the aforementioned memory can be implemented as software functional units and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several commands to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0227] The processor in the electronic device provided in this application embodiment can call logical instructions in the memory to implement the above method. Its specific implementation method is the same as the aforementioned method implementation method and can achieve the same beneficial effects, which will not be repeated here.

[0228] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the methods provided in the above embodiments.

[0229] The specific implementation method is the same as the aforementioned method implementation method and can achieve the same beneficial effects, so it will not be repeated here.

[0230] This application provides a computer program product, including a computer program that, when executed by a processor, implements the method described above.

[0231] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0232] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0233] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for generating high-precision railway maps based on single-journey trajectory correction, characterized in that, include: Simultaneously, continuous image sequences and spatial attitude data of the train during its current operation are collected; The pose change data of the current train is determined based on the visual feature points in the continuous image sequence; Based on the pose change data and the spatial attitude data, synchronous positioning and map construction are performed to generate an initial electronic map; Based on the mileage information and image position corresponding to the preset markers in the continuous image sequence, the mileage and trajectory of the current train in the initial electronic map are corrected to generate the electronic map corresponding to the current train during its journey. The step of correcting the mileage and trajectory of the current train in the initial electronic map based on the mileage information and image position corresponding to preset markers in the continuous image sequence, and generating an electronic map corresponding to the current train during its operation, includes: The continuous image sequence is subjected to marker recognition to determine the image position of each preset marker in the continuous image sequence; Based on the image position and spatial coordinate position of each preset marker, and the parameters of the binocular camera that acquires the continuous image sequence, the actual pose information of the binocular camera is determined. Based on the actual pose information of the binocular camera, the trajectory of the current train in the initial electronic map is corrected; The preset markers include at least one of the following: kilometer markers, half-kilometer markers, 100-meter markers, tunnel entrances, and tunnel exits.

2. The high-precision railway map generation method based on single-journey trajectory correction according to claim 1, characterized in that, After simultaneously acquiring continuous image sequences and spatial attitude data of the train during its current operation, the method further includes: The continuous image sequence is preprocessed; the preprocessing includes at least one of distortion correction, denoising, and image enhancement. Based on the timestamps of the left and right eye images in the continuous image sequence, the binocular images in the continuous image sequence are filtered; Based on the timestamps of the stereo images in the continuous image sequence, the spatial pose data is synchronized and calibrated.

3. The high-precision railway map generation method based on single-journey trajectory correction according to claim 1, characterized in that, The process of synchronously locating and constructing a map based on the pose change data and the spatial pose data to generate an initial electronic map includes: The state vector is determined based on the actual measurement values ​​of each sensor; the sensors include a binocular camera that acquires the continuous image sequence and an inertial measurement unit that acquires the spatial attitude data; Based on the actual measured values ​​and expected calculated values ​​of each sensor at each time point, and the covariance matrix of the actual measured values, the state estimation cost function is determined. The state estimation cost function is optimized to obtain the optimal estimate of the state vector of the current train during its operation. Based on the optimal estimate of the state vector, synchronous positioning and map construction are performed to generate an initial electronic map.

4. The high-precision railway map generation method based on single-journey trajectory correction according to claim 1, characterized in that, The step of correcting the mileage and trajectory of the current train in the initial electronic map based on the mileage information and image position corresponding to preset markers in the continuous image sequence, and generating an electronic map corresponding to the current train during its operation, includes: The continuous image sequence is subjected to marker recognition to determine the target type of each preset marker in the continuous image sequence; Based on the target type of each preset marker, determine the mileage information corresponding to each preset marker; The mileage information corresponding to each preset marker is compared with the measured mileage of the current train, and the mileage of the current train in the initial electronic map is corrected based on the mileage deviation between the mileage information and the measured mileage.

5. The high-precision railway map generation method based on single-journey trajectory correction according to claim 1 or 4, characterized in that, The step of identifying markers in the continuous image sequence includes: The continuous image sequence is input into the marker recognition model to obtain the target type, image location and confidence level of each preset marker output by the marker recognition model; The marker recognition model is trained based on sample images corresponding to each preset marker.

6. The high-precision railway map generation method based on single-journey trajectory correction according to claim 1, characterized in that, After correcting the mileage and trajectory of the current train in the initial electronic map based on the mileage information and image position corresponding to preset markers in the continuous image sequence, and generating the electronic map corresponding to the current train during its operation, the method further includes: Determine the latitude and longitude coordinates of the starting point of the current train on the electronic map; Based on the latitude and longitude coordinates of the starting point and the relative displacement data of each trajectory point of the current train, the latitude and longitude coordinates of each trajectory point are determined. Based on the latitude and longitude coordinates, trajectory point number, road segment name, road segment type, and road segment length of each trajectory point, a key point table, a road segment table, and a road segment details table corresponding to the electronic map are generated.

7. A high-precision railway map generation system based on single-journey trajectory correction, characterized in that, include: The data acquisition module is used to simultaneously acquire continuous image sequences and spatial attitude data of the train during its current operation. The feature extraction module is used to determine the pose change data of the current train based on visual feature points in the continuous image sequence; The synchronous mapping module is used to perform synchronous positioning and map construction based on the pose change data and the spatial attitude data to generate an initial electronic map. The map correction module is used to correct the mileage and trajectory of the current train in the initial electronic map based on the mileage information and image position corresponding to the preset markers in the continuous image sequence, and generate the electronic map corresponding to the current train during its operation. The step of correcting the mileage and trajectory of the current train in the initial electronic map based on the mileage information and image position corresponding to preset markers in the continuous image sequence, and generating an electronic map corresponding to the current train during its operation, includes: The continuous image sequence is subjected to marker recognition to determine the image position of each preset marker in the continuous image sequence; Based on the image position and spatial coordinate position of each preset marker, and the parameters of the binocular camera that acquires the continuous image sequence, the actual pose information of the binocular camera is determined. Based on the actual pose information of the binocular camera, the trajectory of the current train in the initial electronic map is corrected; The preset markers include at least one of the following: kilometer markers, half-kilometer markers, 100-meter markers, tunnel entrances, and tunnel exits.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the high-precision railway map generation method based on single-journey trajectory correction as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the high-precision railway map generation method based on single-journey trajectory correction as described in any one of claims 1 to 6.

Images