Ballast track bed surface ballast space pose estimation method based on machine vision

Abstract

The invention belongs to the technical field of rail transit, and particularly discloses a ballasted track bed surface ballast space pose estimation method based on machine vision. Comprising the steps of ArUco code deployment on the surface of a railway ballast, three-dimensional scanning to construct a point cloud model, local coordinate system establishment based on a mark position, camera parameter calibration, first frame image coordinate system association relation solving and railway ballast pose space estimation. The reasonability of logic and the robustness of an algorithm are tested in a laboratory, the accuracy of the method is guaranteed through multi-mark error analysis, meanwhile, railway ballast group motion tracking is achieved through multiple combination marks, a high-precision kinematics data basis is provided for intelligent monitoring of the state of a ballast bed, and the reliability of the method is improved. The follow-up research on the association relationship between the railway ballast motion characteristics and the ballast bed state is facilitated.

Description

Technical Field

[0001] The present invention relates to the field of rail transportation technology, and in particular to a method for estimating the spatial posture of ballast on a ballasted track bed surface based on machine vision. Background Art

[0002] Railway tracks, as critical infrastructure, are primarily categorized into ballastless and ballasted track structures. While ballastless track offers high stability and uniform stiffness, making it widely used on high-speed railways, it suffers from the drawback of being difficult to repair. Ballasted track, on the other hand, remains indispensable for special sections and high-speed railways with speeds below 250 km / h, thanks to its superior elasticity, good drainage, and easy maintenance.

[0003] The core problem of ballasted track performance degradation lies in the difficulty of assessing the condition of the subgrade: the dynamic behavior of the subgrade is significantly more complex than that of the supergrade. With the increasing speed and heavier loads of trains, the micromechanical behavior (fragmentation and deformation) and kinematic response (displacement, rotation, and rearrangement) of ballast particles under cyclic loading can lead to the accumulation of macroscopic defects. Therefore, accurately quantifying the spatial position evolution of ballast is of great engineering value for track condition prediction, studying subgrade instability mechanisms, and analyzing force chain distribution and strength.

[0004] Current research on the motion of ballast in ballasted trackbeds relies primarily on two methods: discrete element simulation and intelligent sensing experiments. Discrete element simulation is widely used to study the mechanical properties of granular materials such as railway ballast, forming a relatively mature methodology. Based on the collaborative simulation method of discrete element method and multi-body dynamics, Zhang Zhihai et al. constructed a refined coupling model of three-sleeper tamping device, track panel and ballasted trackbed, and analyzed the motion characteristics and energy evolution law of ballast during tamping operation; Zhang Jie et al. used discrete element method to study the mechanical influence mechanism of ballast embedding on roadbed deformation from multiple angles; Xu Peng et al. simulated different axle load conditions through discrete element simulation, focusing on analyzing the force distribution law of trackbed bearing capacity; Chen Cheng et al. used three-dimensional scanning technology to reconstruct the real geometric shape of ballast particles, combined with the discrete element method to simulate the single-sleeper ballast box test, and explored the real force boundary conditions; Liu Ganzhong established a three-sleeper trackbed model on a bridge based on discrete element simulation technology, and found that when the sleeper vibrates, the ballast produces significant flow in the bottom and ballast shoulder areas of the sleeper, and the flow velocity near the sleeper is relatively high; Peng Hui et al. used discrete element software to construct a simplified two-dimensional trackbed model at the mid-span of the main span of the bridge, revealing the ballast flow characteristics and trackbed thickness evolution law under the coupling of temperature cycle and train load from macroscopic and microscopic dimensions.

[0005] Current research focuses on a refined description of the true motion mechanisms of ballast particles. By integrating emerging technologies such as intelligent particle sensors and machine vision with traditional trackbed experiments, the correlation between ballast motion and trackbed condition is revealed. Xiao Yuanjie et al. quantified the mapping relationship between ballast deformation behavior and dynamic stress amplitude based on large-scale triaxial experiments. Using SmartRock sensors in plate-type vibration compaction tests to monitor differences in particle rotation at different spatial locations, they analyzed the evolution of the ballast acceleration energy spectrum throughout the compaction process. Wang Meng et al. conducted large-scale monotonic loading triaxial compression experiments, introduced a soiling condition variable, and used intelligent sensor motion data to construct a coupled model of soiling rate, macroscopic shear strength, and microscopic particle rotation. Wang Meng et al. simulated the trackbed compaction process through indoor rotational compaction experiments, establishing a quantitative correlation between motion characteristics and compaction degree based on the relative ballast rotation angle and angle ratio. Fu et al. used intelligent sensors to identify the service status of full-scale trackbeds and proposed a trackbed condition recognition method that integrates the multidimensional motion characteristics of particles.

[0006] In current research on roadbed structural defects, methods for identifying gravel roadbed conditions based on electromagnetic signatures have been widely used. Khakiev et al. constructed a quantitative index for roadbed moisture by integrating radar reflection signals from roadbed layers, and verified its accuracy through field excavation. Wang Shilei et al. analyzed the time-frequency characteristics of radar electromagnetic signals, extracted multi-parameter curves evolving along the line, and determined that three indicators were significantly correlated with roadbed service condition. Silvast et al. proposed constructing an index representing roadbed contamination rate using spectral domain integration. In addition to electromagnetic identification technology, emerging detection methods continue to expand their application areas. Liang et al. used infrared thermal imaging to compare the internal thermodynamic properties of clean and dirty roadbeds to identify optimal detection indicators. Bian et al. applied color marking to ballast particles to quantitatively analyze their motion behavior under different train speeds and axle loads. Kumara et al. developed an image-based ballast assessment technology that accurately quantifies sand contamination and simultaneously generates particle grading curves.

[0007] Existing research on ballast motion characteristics primarily relies on intelligent particle sensors, but these sensors suffer from the triple limitations of time-domain error accumulation, contact mechanics simulation distortion, and long-term energy constraints. While traditional machine vision marking methods can partially address these shortcomings, existing research is limited by dimensional deviations in circular markers, the lack of vertical information in two-dimensional projection coordinates (capturing only Z-axis rotation), the ability to record only two-dimensional planar motion data, data interruptions due to occlusions, and insufficient sensitivity for detecting small rotations. These limitations make them difficult to meet the requirements for analyzing ballast motion in six degrees of freedom (6DOF). Summary of the Invention

[0008] In order to solve the defects of the existing technology such as accumulated displacement errors, missing rotation dimensions and insufficient anti-occlusion ability, the present invention introduces a 2D-3D correspondence method in the field of pose estimation and proposes a method for estimating the spatial pose of ballast on the surface of a ballasted track bed based on machine vision. The method aims to achieve the operational efficiency of rapid detection of multiple targets in a single frame, the breakthrough accuracy of synchronous analysis of sub-millimeter displacement and 0.5° rotation, the information completeness of the complete output of six-degree-of-freedom pose parameters, and the system scalability of synchronous tracking of multiple particle motions, providing a high-precision data basis for the study of ballast motion characteristics and ultimately filling the technical gap in the precise detection of ballast motion in all dimensions.

[0009] To achieve the above object, the present invention provides the following technical solution: a method for estimating the spatial pose of ballast on a ballasted track bed surface based on machine vision, comprising the following steps: S1. Paste ArUco codes on the ballast surface: Select the flat surface of the ballast particles and paste three 7×7 ArUco codes with IDs of 0, 1, and 2 respectively; S2. 3D scanning of ballast profile: Use a 3D scanning device to scan the marked ballast particles at multiple angles to obtain a point cloud model of the ballast profile; S3, establishing the local coordinate system of the ballast particles based on the three ArUco code positions; S4, perform camera calibration; S5. Establish associations between the three ArUco code coordinate systems and the local coordinate system through the first frame image; S6. Realize the spatial pose estimation of ballast.

[0010] Preferably, the ArUco code in step S1: The ArUco code is a marker with unique information encoding. Its structure consists of an outer black border and an inner binary information grid. ArUco codes with different IDs contain unique information, providing a position and direction reference for machine vision detection. Identification and detection are implemented through the ArucoDetection algorithm in the Python environment, and an independent coordinate system is established for each marker: an independent coordinate system is established with the upper left corner of the marker as the origin, the edge of the marker as the X / Y axis, and the perpendicular marker plane as the Z axis.

[0011] Preferably, step S3 specifically includes: a) Identify the three ArUco code locations in the point cloud processing software; b) Select the upper left corner of each ArUco code (i.e. the origin of its coordinate system) as the reference point (ID 0→A, ID 1→B, ID2→C); c) Construct a local coordinate system: take A as the origin, vector AB→ AB is the X-axis, plane ABC (the plane where points A, B, and C are located) is the XY plane, and the axis perpendicular to this plane is the Z-axis; d) Calculate the point cloud centroid Coordinates in the local coordinate system.

[0012] Preferably, the centroid calculation formula is as follows: ; Where, are the three-dimensional space coordinates of A, B, and C; ; Where, is the unit vector of the coordinate axis, and each vector is obtained through the coordinates of points A, B, and C. The direction of each coordinate axis is obtained according to this formula; ; Since the mass of ballast particles is evenly distributed, the centroid and center of mass coincide in space, and the center of mass coordinates can be solved using the centroid calculation formula. is the coordinate of the center of mass in the software coordinate system, is the number of mass points, is the coordinate of the mass point, and the coordinate of the center of mass is obtained by weighted average calculation of the positions of each mass point; The transformation from the software coordinate system to the local coordinate system is done by the homogeneous transformation matrix This matrix uniformly describes spatial rotation and translation transformations, facilitating accurate and reversible transformations between coordinate systems.

[0013] ; Among them, the rotation matrix R is: ; In the rotation matrix, is the rotation angle around the Y axis, is the rotation angle around the Z axis, is the rotation angle around the X axis; After transformation Axis in the original The component on the axis, After transformation Axis in the original The component on the axis, After transformation Axis in the original The component on the axis, After transformation Axis in the original The component on the axis, After transformation Axis in the original The component on the axis, After transformation Axis in the original The component on the axis, After transformation Axis in the original The component on the axis, After transformation Axis in the original The component on the axis, After transformation Axis in the original The components on the axis; the specific form of the rotation matrix is ​​as follows: ; Translation vector : ; Represents the translation along each axis of the original coordinate system; Centroid coordinates in the local coordinate system The centroid coordinates in the software coordinate system After homogeneous transformation, we get: .

[0014] Preferably, step S4 is specifically implemented as follows: By utilizing the correspondence between the feature points with known precise 3D coordinates on the calibration plate and their 2D projection points in the image, based on the pinhole imaging model and lens distortion model, the camera's intrinsic parameters (characterizing the geometric and optical properties of the imaging system) and extrinsic parameters (describing the position transformation of the camera relative to the calibration plate) are solved. The obtained parameters will be used as known conditions for the subsequent calculation of the ballast posture change based on the 2D-3D correspondence.

[0015] Preferably, step S5 includes: Capture the first frame image containing three ArUco codes. By using the complete first frame image containing three ArUco codes, identify and solve the position of each ArUco code in the camera coordinate system; based on this, establish the transformation relationship between each ArUco code's own coordinate system and the local coordinate system, and further derive the pose relationship between the local coordinate system and the camera coordinate system.

[0016] Preferably, the principle of the conversion relationship between the ArUco code's own coordinate system and the local coordinate system, and the conversion relationship between the local coordinate system and the camera coordinate system is as follows: assuming that the motion of any point of a rigid body is divided into translation and rotation, its motion is regarded as the synthesis of translation with the center of mass and rotation around the center of mass. Therefore, expressing the motion of the center of mass is equivalent to expressing the motion of the entire rigid body, as follows: 1) Calculate the rotation relationship between the ArUco code's own coordinate system and the camera's coordinate system: ; Where, is the rotation matrix of the ArUco code, is the fixed transformation for calibration; 2) Calculate the position of the rigid body point in the camera coordinate system : ; Where, is the position of the ArUco code, is the calibration parameter; 3) Calculate the position of the center of mass in the camera coordinate system ; ; Where, is the fixed position of the center of mass; 4) Pass and To fully characterize the transformation relationship, the above rotation matrices are all third-order matrices, in the following form: ; Where, is the rotation angle around the Y axis, For around Axis rotation angle, is the rotation angle around the X axis.

[0017] Preferably, in step S6: Based on the coordinate system conversion relationship established in the first frame, the subsequent frames recognize at least one ArUco code, and then realize the ballast spatial pose estimation through pose solution and error analysis. When at least two ArUco codes are recognized in the subsequent frames, the centroid position corresponding to each mark is independently solved; the Euclidean distance between the two centroid positions is calculated. ;like Less than the preset threshold , the result is determined to be reliable and the average value of the centroid position is taken as the output; if , regarded as dirty data, thus constructing the ballast time-space displacement dataset.

[0018] The beneficial effects of the present invention are significantly reflected in three aspects: Technological Breakthrough: This method overcomes the dual limitations of contact mechanics distortion caused by the non-realistic contours of smart sensors and the lack of rotational dimension in machine vision. It proposes a six-degree-of-freedom pose estimation method based on multi-marker spatial solution (leveraging the rigid body constraint relationship of three non-collinear markers) and sub-pixel vector analysis (using a sub-pixel corner detection algorithm to improve feature positioning accuracy to 0.1 pixel level), enabling accurate capture of the full-dimensional motion of ballast. Superior performance: Laboratory verification shows that this method achieves submillimeter displacement accuracy (±0.5mm) and 0.5° rotation resolution, and has a high continuity rate for pose estimation under partial occlusion conditions, far exceeding traditional marking methods. Foresighted Application: This system provides a full-dimensional, high-precision kinematic dataset for the study of the correlation mechanism between ballast motion and trackbed status, achieving a technological breakthrough in the precise detection of the six-degree-of-freedom spatial posture of ballasted trackbed granular bodies. It also provides a core algorithm module for the intelligent health monitoring system for ballasted railways, supporting preventive maintenance decisions for trackbed status. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 Schematic diagram of the flow of a method for estimating the spatial pose of ballast on a ballasted track bed surface based on machine vision in an embodiment; Figure 2 Schematic diagram of the ArUco code of 7×7 and ID=0 in the embodiment; Figure 3 Schematic diagram of ArUco code attached to the ballast surface in the embodiment; Figure 4 Schematic diagram of 3D scanning ballast profile in an embodiment; Figure 5 Schematic diagram of the ballast profile point cloud model in the embodiment; Figure 6 Schematic diagram of extracting ArUco code corner point coordinates through software processing in the embodiment; Figure 7 Schematic diagram of the local coordinate system in the embodiment; Figure 8 Schematic diagram of the conversion relationship principle in the embodiment; Figure 9a Schematic diagram of single ballast monitoring initialization in the embodiment; Figure 9b Schematic diagram of subsequent frames for single ballast monitoring in an embodiment; Figure 10a Schematic diagram of initialization of ballast group particle monitoring in the embodiment; Figure 10b Schematic diagram of subsequent frames for monitoring ballast group particles in the embodiment; Figure 11a This is a schematic diagram of the recognition results under dark conditions in the embodiment; Figure 11b Schematic diagram of recognition results under natural light conditions in the embodiment. DETAILED DESCRIPTION

[0020] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0021] In order to overcome the technical bottlenecks in the background technology, this paper proposes a method for estimating the spatial pose of surface ballast based on machine vision, which achieves three breakthroughs in core functional objectives: First, through the reconstruction of the three-dimensional spatial coordinates of the ArUco code and the vector analysis algorithm, the ballast displacement and the complete rotation angle around the X, Y, and Z axes are simultaneously calculated, breaking through the limitation of the two-dimensional projection dimension. Secondly, a multi-marker topological association solution mechanism is designed to maintain the continuity of pose estimation based on spatial geometric constraints when the markers are occluded; Finally, by combining sub-pixel corner detection technology, the simultaneous and precise capture of 0.5°-level micro-rotation and millimeter-level spatial displacement of ballast particles can be achieved, completing the quantitative analysis of their spatial posture.

[0022] This method will promote the innovation of ballasted track maintenance mode: based on millimeter-level displacement accuracy and sub-angle rotation analysis capabilities, a non-contact long-term monitoring system will be constructed to fill the shortcomings of existing technologies in terms of motion dimension integrity, environmental robustness and mechanism correlation, and directly support the panoramic perception and preventive decision-making of the trackbed status in the intelligent operation and maintenance of rail transit.

[0023] The present invention provides a technical solution: a method for estimating the spatial posture of ballast on the surface of a ballasted track bed based on machine vision, the process is as follows: Figure 1 As shown, the following steps are included: S1. Paste ArUco codes on the ballast surface: In this example, three 7×7 ArUco codes with IDs of 0, 1, and 2 are pasted on the flat surface of the ballast particles. The ArUco code format is as follows: Figure 2 As shown, the ArUco code is pasted on the ballast surface as shown in the figure. Figure 3 shown.

[0024] The ArUco code is a visual marker with a unique ID. Its structure consists of an outer black border and an inner binary information grid. High-precision recognition is achieved through the ArucoDetection algorithm in a Python environment. The marker establishes an independent coordinate system for each ballast particle, with the upper left corner as the origin, the marker edge as the X and Y axes, and the vertical plane as the Z axis. This provides a spatial position and orientation reference for machine vision.

[0025] S2. 3D scanning of ballast profile: Use a 3D scanner to scan the ballast particles with ArUco codes at multiple angles, such as Figure 4 As shown in , the accurate ballast profile point cloud model is obtained, as shown in Figure 5 shown.

[0026] S3. Establish the local coordinate system of the ballast particles based on the three ArUco code positions.

[0027] Based on the ballast outline point cloud model obtained in step S2, the coordinates of the upper left corner of the three ArUco codes (i.e., the origin of the coordinate system) are identified in the point cloud processing software, and IDs 0, 1, and 2 are assigned to reference points A, B, and C respectively; with point A as the origin, vector AB as the positive direction of the X-axis, the plane uniquely determined by points A, B, and C is the XY plane, and the axis passing through origin A and perpendicular to the XY plane is the Z axis to construct a local coordinate system. Since the mass of the ballast is uniform, the coordinates of the point cloud center of mass O in the local coordinate system can be solved according to the centroid calculation formula. Finally, a local coordinate system with A as the origin is established, and the position of the center of mass O is determined in this coordinate system (see Extraction of corner coordinates). Figure 6 , the coordinate system configuration is shown in Figure 7 ).

[0028] The centroid calculation formula is as follows: ; Where, are the three-dimensional space coordinates of A, B, and C; ; Where, is the unit vector of the coordinate axis, and each vector is obtained through the coordinates of points A, B, and C. The direction of each coordinate axis is obtained according to this formula; ; Since the mass of ballast particles is evenly distributed, the centroid and center of mass coincide in space, and the center of mass coordinates can be solved using the centroid calculation formula. is the coordinate of the center of mass in the software coordinate system, is the number of mass points, is the coordinate of the mass point, and the coordinate of the center of mass is obtained by weighted average calculation of the positions of each mass point; The transformation from the software coordinate system to the local coordinate system is done by the homogeneous transformation matrix This matrix uniformly describes spatial rotation and translation transformations, facilitating accurate and reversible transformations between coordinate systems.

[0029] ; Among them, the rotation matrix R is: ; In the rotation matrix, is the rotation angle around the Y axis, is the rotation angle around the Z axis, is the rotation angle around the X axis; After transformation Axis in the original The component on the axis, After transformation Axis in the original The component on the axis, After transformation Axis in the original The component on the axis, After transformation Axis in the original The component on the axis, After transformation Axis in the original The component on the axis, After transformation Axis in the original The component on the axis, After transformation Axis in the original The component on the axis, After transformation Axis in the original The component on the axis, After transformation Axis in the original The components on the axis; the specific form of the rotation matrix is ​​as follows: ; Translation vector : ; Represents the translation along each axis of the original coordinate system; Centroid coordinates in the local coordinate system The centroid coordinates in the software coordinate system After homogeneous transformation, we get: .

[0030] S4. Camera calibration.

[0031] By utilizing the correspondence between the feature points with known precise 3D coordinates on the calibration plate and their 2D projection points in the image, based on the pinhole imaging model and lens distortion model, the camera's intrinsic parameters (characterizing the geometric and optical properties of the imaging system) and extrinsic parameters (describing the position transformation of the camera relative to the calibration plate) are solved. The obtained parameters will be used as known conditions for the subsequent calculation of the ballast posture change based on the 2D-3D correspondence.

[0032] S5. Establish the connection between the three ArUco code coordinate systems and the local coordinate system through the first frame image.

[0033] Through the first frame image containing three complete ArUco codes, the three-dimensional coordinates of points A, B, and C in the camera coordinate system are identified and obtained, and the rotation-translation transformation from each ArUco code coordinate system to the local coordinate system and the pose transformation from the local coordinate system to the camera coordinate system are calculated accordingly. The conversion principle (such as Figure 8As shown in Figure 2, the motion of a rigid body can be decomposed into the translation of the center of mass and the rotation around the center of mass. That is, the center of mass pose can represent the full-degree-of-freedom motion of the rigid body, as follows: 1) Calculate the rotation relationship between the ArUco code's own coordinate system and the camera's coordinate system: ; Where, is the rotation matrix of the ArUco code, is the fixed transformation for calibration; 2) Calculate the position of the rigid body point in the camera coordinate system : ; Where, is the position of the ArUco code, is the calibration parameter; 3) Calculate the position of the center of mass in the camera coordinate system ; ; Where, is the fixed position of the center of mass; 4) Pass and To fully characterize the transformation relationship, the above rotation matrices are all third-order matrices, in the following form: ; Where, is the rotation angle around the Y axis, For around Axis rotation angle, is the rotation angle around the X axis.

[0034] S6. Ballast pose spatial estimation.

[0035] Based on the coordinate system conversion relationship established in the first frame, the subsequent frames recognize the ArUco code, and the ballast spatial pose is estimated through pose solution and error analysis. When two or more markers are detected, the error analysis is triggered: the corresponding centroid position of each marker is calculated independently, and the Euclidean distance is calculated. ;like ( is the preset threshold), output the mean value of the centroid position; if , is considered as dirty data, thus constructing the ballast time-space displacement dataset. Figure 9a As shown, it is the first frame initialization. Figure 9b As shown, it is the subsequent frame recognition.

[0036] By configuring exclusive ArUco code combinations for different ballast particles, the single-particle pose solution process is reused ( Figure 10a The first frame of multi-track ballast particle initialization is shown. Figure 10b The figure shows the subsequent frame recognition of multiple ballast particles), which enables parallel tracking of the six-degree-of-freedom motion of the ballast particle group.

[0037] Environmental adaptability verification: algorithm testing in complex lighting environments such as strong light, shadow, and weak light ( Figure 11a The following shows the dark working condition of the identification code. Figure 11b The system still maintains high recognition accuracy under natural light conditions, proving its robustness to complex lighting conditions.

[0038] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or apparatus comprising the element.

[0039] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The singular forms "a", "an", "the" and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise.

[0040] It should be understood that the term "and / or" as used herein is merely a description of the relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. Furthermore, the character " / " in this document generally indicates that the associated objects are in an "or" relationship.

[0041] The word "if," as used herein, may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to the determination" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)," depending on the context.

[0042] The references to "first" and "second" in the embodiments merely distinguish similar objects and do not represent a specific ordering of the objects. It is understood that the specific order or precedence of "first" and "second" can be interchanged where appropriate. It should be understood that the objects distinguished by "first" and "second" can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein.

Claims

1. A method for estimating the spatial pose of ballast on a ballasted track bed surface based on machine vision, characterized in that: The steps include: S1. Paste ArUco codes on the ballast surface: Select the flat surface of the ballast particles and paste three 7×7 ArUco codes with IDs of 0, 1, and 2 respectively; S2. 3D scanning of ballast profile: Use a 3D scanning device to scan the marked ballast particles at multiple angles to obtain a point cloud model of the ballast profile; S3, establishing the local coordinate system of the ballast particles based on the three ArUco code positions; S4, perform camera calibration; S5. Establish associations between the three ArUco code coordinate systems and the local coordinate system through the first frame image; S6. Realize the spatial pose estimation of ballast.

2. The method for estimating the spatial pose of ballast on a ballasted track bed surface based on machine vision according to claim 1, characterized in that: In step S1, the ArUco code is a marker with unique information encoding, and its structure includes an outer black border and an inner binary information grid; ArUco codes with different IDs contain unique information, providing position and direction references for machine vision detection; recognition detection is implemented through the ArucoDetection algorithm in the Python environment, and an independent coordinate system is established for each marker: the coordinate system takes the upper left corner of the marker as the origin, the marker edge directions as the X-axis and Y-axis, and the axis perpendicular to the marker plane as the Z-axis.

3. The method for estimating the spatial pose of ballast on a ballasted track bed surface based on machine vision according to claim 1, characterized in that: In step S3, the specific steps include: Based on the ballast profile point cloud model, the positions of the three ArUco codes are identified in the point cloud processing software. The upper left corner point of each ArUco code is selected as the reference point, where ID 0 corresponds to point A, ID 1 corresponds to point B, and ID 2 corresponds to point C. A local coordinate system is established: point A is the origin, vector AB is the X-axis, the plane where points A, B, and C are located is the XY plane, and the axis perpendicular to the plane is the Z-axis. According to the centroid calculation formula, the coordinates of the point cloud centroid in the local coordinate system are solved.

4. The method for estimating the spatial pose of ballast on a ballasted track bed surface based on machine vision according to claim 3, characterized in that: The centroid calculation formula is as follows: ； Where, are the three-dimensional space coordinates of A, B, and C; ； Where, is the unit vector of the coordinate axis, and each vector is obtained through the coordinates of points A, B, and C. The direction of each coordinate axis is obtained according to this formula; ； Since the mass of ballast particles is evenly distributed, the centroid and center of mass coincide with each other in space, and the center of mass coordinates are solved by the centroid calculation formula; where: is the coordinate of the center of mass in the software coordinate system, is the number of mass points, is the coordinate of the mass point, and the coordinate of the center of mass is obtained by weighted average calculation of the positions of each mass point; The transformation from the software coordinate system to the local coordinate system is done by the homogeneous transformation matrix This matrix uniformly describes spatial rotation and translation transformations, facilitating accurate and reversible transformations between coordinate systems: ； Among them, the rotation matrix R is: ； In the rotation matrix, is the rotation angle around the Y axis, is the rotation angle around the Z axis, is the rotation angle around the X axis; After transformation Axis in the original The component on the axis, After transformation Axis in the original The component on the axis, After transformation Axis in the original The component on the axis, After transformation Axis in the original The component on the axis, After transformation Axis in the original The component on the axis, After transformation Axis in the original The component on the axis, After transformation Axis in the original The component on the axis, After transformation Axis in the original The component on the axis, After transformation Axis in the original The components on the axis; the specific form of the rotation matrix is ​​as follows: ； Translation vector : ； Represents the translation along each axis of the original coordinate system; Centroid coordinates in the local coordinate system The centroid coordinates in the software coordinate system After homogeneous transformation, we get: 。 5. The method for estimating the spatial pose of ballast on a ballasted track bed surface based on machine vision according to claim 1, characterized in that: In step S4, the correspondence between the feature points with known precise three-dimensional coordinates on the calibration plate and their two-dimensional projection points in the image is used to solve the intrinsic and extrinsic parameters of the camera based on the pinhole imaging model and the lens distortion model; the obtained parameters will be used as known conditions for the subsequent calculation of the ballast posture change based on the 2D-3D correspondence.

6. The method for estimating the spatial pose of ballast on a ballasted track bed surface based on machine vision according to claim 1, characterized in that: In step S5, the position of each ArUco code in the camera coordinate system is identified and solved by using the first frame image that completely contains three ArUco codes; based on this, the transformation relationship between the coordinate system of each ArUco code itself and the local coordinate system is established, and the pose relationship between the local coordinate system and the camera coordinate system is further derived.

7. The method for estimating the spatial pose of ballast on a ballasted track bed surface based on machine vision according to claim 6, characterized in that: The details include: 1) Calculate the rotation relationship between the ArUco code's own coordinate system and the camera's coordinate system: ； Where, is the rotation matrix of the ArUco code, is the fixed transformation for calibration; 2) Calculate the position of the rigid body point in the camera coordinate system : ； Where, is the position of the ArUco code, is the calibration parameter; 3) Calculate the position of the center of mass in the camera coordinate system ; ； Where, is the fixed position of the center of mass; 4) Pass and To fully characterize the transformation relationship, the above rotation matrices are all third-order matrices, in the following form: ； Where, is the rotation angle around the Y axis, For around Axis rotation angle, is the rotation angle around the X axis.

8. The method for estimating the spatial pose of ballast on a ballasted track bed surface based on machine vision according to claim 1, characterized in that: In step S6, based on the conversion relationship between the ArUco code coordinate system and the local coordinate system established in the first frame, the subsequent frames recognize at least one ArUco code, and realize the ballast spatial pose estimation through pose solution and error analysis.

9. The method for estimating the spatial pose of ballast on a ballasted trackbed surface based on machine vision according to claim 8, characterized in that: The error analysis and judgment includes: a) When at least two ArUco codes are recognized in the subsequent frame, the centroid position corresponding to each marker is independently calculated; b) Calculate the Euclidean distance between the two centroid positions ; c) If Less than the preset threshold , the result is determined to be reliable and the average value of the centroid position is taken as the output; d) If , the judgment result is unreliable and is regarded as dirty data.

Images