Railway ballast detection method and railway ballast detection system

Abstract

The invention relates to the field of rail transit, and discloses a railway ballast detection method and a railway ballast detection system.The method comprises the steps that navigation data in vehicle movement are obtained through an inertial navigation unit, and the navigation data are combined with GPS positioning data to conduct fusion calculation to obtain POS data; the method comprises the following steps: scanning a ballast bed, a railway ballast and a sleeper through a laser radar to obtain original three-dimensional point cloud data, and fitting POS data with the original three-dimensional point cloud data to obtain standard point cloud data; preprocessing the obtained standard point cloud data to obtain orbit data, and performing rotation correction on the orbit data through a data correction module; and calibration is performed based on the height difference between the sleeper and the track, an included angle between the laser radar and the ground is obtained, target detection is performed based on the included angle between the laser radar and the ground, and structured measurement data is generated and output.

Description

Technical Field

[0001] The present invention relates to the field of rail transportation, and in particular to a ballast detection method and a ballast detection system. Background Art

[0002] With the increase in heavy-duty railway freight volume and the increase in train speed, the requirements for railway line safety are becoming increasingly higher. The poor drainage along some sections during the rainy season leads to uncompacted ballast and water accumulation in the roadbed, which in turn causes grooving, mud oozing and uneven settlement of the ballast in the lower layer of the roadbed.

[0003] After the tamping machine compacts the ballast, the density of the ballast increases, the bearing capacity of the roadbed is enhanced, and the drainage capacity of the road is also improved to prevent water erosion on the roadbed. However, after the ballast is compacted, the vertical height of the ballast relative to the top surface of the sleeper increases, and the slope of the roadbed changes, which has an adverse effect on line safety and normal operation.

[0004] At present, the detection of ballast missing or defective is mainly done manually, which is very labor-intensive, inefficient and poses a safety hazard.

[0005] Currently, there is no effective detection system to monitor these changes and provide timely repair data. Summary of the invention

[0006] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a ballast detection method, comprising the following steps:

[0007] Step 1: Obtain navigation data during vehicle movement through the inertial navigation unit, and combine it with GPS positioning data to obtain POS data through fusion calculation;

[0008] Step 2: Scan the track bed, ballast and sleepers with a laser radar to obtain original 3D point cloud data, and obtain standard point cloud data by fitting the POS data with the original 3D point cloud data;

[0009] Step 3: preprocess the obtained standard point cloud data, including noise removal, smoothing and format conversion, and then extract key geometric features from the preprocessed standard point cloud data to obtain track data, which is then rotated and corrected by a data correction module;

[0010] Step 4: Calibrate based on the height difference between the sleeper and the track to obtain the angle between the lidar and the ground. Perform target detection based on the angle between the lidar and the ground, including measuring the top surface of the ballast bed, the width of the ballast bed, the height of the ballast shoulder, the slope of the ballast side slope, and the height difference between the ballast and the sleeper, and generate and output structured measurement data.

[0011] Furthermore, it also includes an infrared monitoring camera, which synchronously collects the roadbed image and aligns it with the point cloud data.

[0012] Furthermore, the roadbed, ballast and sleepers are scanned by laser radar to obtain original three-dimensional point cloud data, and standard point cloud data is obtained by fitting POS data with the original three-dimensional point cloud data, including:

[0013] By receiving the original lidar data and parsing it to generate point cloud data, the pbAI::messageToAI message structure is used to parse the received binary data into recognizable structured data.

[0014] Furthermore, the obtained standard point cloud data is preprocessed, including noise removal, smoothing and format conversion, including:

[0015] The point cloud data is preprocessed, noise is removed and downsampled, outliers are removed by calculating the number of points in the neighborhood of each point in the point cloud, and MeanK=50 and StddevMulThresh=1.0 are set to delete noise points that do not meet statistical characteristics.

[0016] Furthermore, the key geometric features are extracted from the pre-processed standard point cloud data to obtain the track data, including:

[0017] For the filtered point cloud data, according to the range of the x coordinates in the point cloud data, the point cloud in the track area is screened out, and the minimum value minX=-5 and the maximum value maxX=5 of the x coordinate of the track area are set;

[0018] Use the FindRailwayTracks function to extract the left and right track points of the track; by calculating the x-coordinates of the center points of the left and right track points, check whether the distance of the track is within the set range. If the distance is within the set range, the track detection is successful.

[0019] Furthermore, the rotation correction of the orbital data by the data correction module includes:

[0020] Rotation correction: Calculate the rotation angle of the track and rotate the point cloud along the Y axis by this angle so that the track is parallel to the horizontal plane in the point cloud data. The rotation angle is calculated by the atan function, and the formula is:

[0021]

[0022] For each (x,y,z), the coordinate after rotation around the Y axis is calculated as:

[0023] x′=x·cos(θ)+z·sin(θ)

[0024] z′=-x·sin(θ)+z·cos(θ)

[0025] Among them, the y coordinate remains unchanged;

[0026] Track height calibration: By recalculating the corrected point cloud data, determine whether the left and right track point clouds are at the same height. If the left and right track point clouds are at the same height, the calibration is completed, otherwise readjustment is required.

[0027] Furthermore, the calibration based on the height difference between the sleeper and the track to obtain the angle between the laser radar and the ground includes:

[0028] The corrected and rotated point cloud set is screened according to the middle value of the X coordinates of the left and right rails, and the middle 80% of the point cloud data of the left and right rails is taken to screen out the point cloud set of the sleeper or ballast;

[0029] Calculate the standard deviation of the middle point cloud and judge whether it is smooth by the standard deviation. If the standard deviation is small and smooth, it is a sleeper; if the standard deviation is large and not smooth, it is ballast. The formula is as follows:

[0030] Calculate the mean. This part calculates the cumulative sum of all z coordinates and divides it by the number of points to get the mean, where the z coordinate of each point is {p1,p2,…,p n}, its mean (z i ) is (μ z )for:

[0031]

[0032] Variance σ 2 for:

[0033]

[0034] The standard deviation σ is:

[0035]

[0036] The angle between the laser radar and the ground is:

[0037]

[0038] Furthermore, the target detection based on the angle between the laser radar and the ground includes measuring the top surface of the ballast bed, the width of the ballast bed, the height of the ballast shoulder, the slope of the ballast side slope, and the height difference between the ballast and the sleeper, generating and outputting structured measurement data, including:

[0039] Determine the top surface and overall width of the roadbed through the X value range of the point cloud data; obtain the width of the roadbed by calculating the difference between the maximum and minimum X values;

[0040] Ballast shoulder height measurement: For the point cloud of the ballast shoulder area, calculate the average height of the point cloud of the ballast shoulder area relative to the reference plane or track plane. Obtain the ballast shoulder height;

[0041] Ballast slope measurement: Select point cloud data in the slope area and use linear regression analysis to calculate the slope. The linear regression formula is as follows:

[0042]

[0043] Vertical height measurement of the ballast on the top surface of the trackbed and the top surface of the sleeper: By comparing the average Z value of the point cloud of the top surface of the sleeper and the top surface of the ballast, the height difference between the ballast on the top surface of the trackbed and the top surface of the sleeper is calculated.

[0044] A ballast detection system, characterized in that it applies the ballast detection method described above, and comprises an inertial navigation unit, a laser radar, an infrared monitoring camera, a data processing module, a communication module and a ground terminal module;

[0045] The inertial navigation unit, laser radar, infrared monitoring camera and communication module are respectively connected to the data processing module, and the ground terminal module is communicatively connected to the communication module.

[0046] The beneficial effect of the present invention is that by using laser radar technology, the present invention can obtain the three-dimensional point cloud data of the roadbed in real time, and accurately calculate the vertical height difference between the ballast and the sleeper by comparing it with the known height difference of the sleeper. This process ensures that the point cloud data collected by the laser radar has high accuracy and high reliability through the fusion of inertial navigation data and positioning data, so that abnormal conditions such as settlement and looseness of the ballast can be quickly discovered.

[0047] The present invention can accurately measure the slope of the roadbed slope by extracting point cloud data of the slope area and applying a linear regression analysis algorithm. This measurement helps to detect the stability of the roadbed slope, especially in the rainy season or on sections with complex geology, and timely discover and warn of the risk of slope sliding or collapse that may be caused.

[0048] The system can simultaneously measure multiple key parameters such as the width of the top surface of the track bed and the height of the ballast shoulder, and through data processing and feature extraction technology, it can achieve a comprehensive assessment of the overall structural status of the track bed. For example, the precise measurement of the ballast shoulder height allows the system to identify the accumulation or loss of ballast, providing important reference data for maintenance personnel to replenish or clean up the ballast in a timely manner.

[0049] The system design of the present invention allows the inspection vehicle to collect data in real time while driving, and immediately process and analyze it. Through high-speed data preprocessing and format conversion modules, the system can effectively remove noise and retain key data. As a result, the system can instantly identify abnormalities in the roadbed state, and through image and data synchronization technology, visualize the analysis results, so that maintenance personnel can take quick action.

[0050] The measurement results generated by the system are output in a structured data format for further analysis and reporting. By comparing with standard measurement values, the system can evaluate performance and automatically adjust algorithm parameters to ensure the accuracy and consistency of subsequent inspections. This adaptive optimization function greatly improves the inspection efficiency and accuracy of the system, making the roadbed maintenance work more scientific and standardized. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 A schematic diagram of a ballast detection method is shown;

[0052] Figure 2 Schematic diagram of the structure of the ballast dynamic detection system;

[0053] Figure 3 is the pre-processing flow chart;

[0054] Figure 4 It is the feature extraction flow chart;

[0055] Figure 5 Correction flow chart for data;

[0056] Figure 6 It is the calibration flow chart;

[0057] Figure 7 This is the target detection flow chart;

[0058] Figure 8 This is the result output and correction flow chart. DETAILED DESCRIPTION

[0059] The technical solution of the present invention is further described in detail below in conjunction with the accompanying drawings, but the protection scope of the present invention is not limited to the following.

[0060] The features and performance of the present invention are further described in detail below in conjunction with the embodiments.

[0061] like Figure 1 As shown, a ballast detection method comprises the following steps:

[0062] Step 1: Obtain navigation data during vehicle movement through the inertial navigation unit, and combine it with GPS positioning data to obtain POS data through fusion calculation;

[0063] Step 2: Scan the track bed, ballast and sleepers with a laser radar to obtain original 3D point cloud data, and obtain standard point cloud data by fitting the POS data with the original 3D point cloud data;

[0064] Step 3: preprocess the obtained standard point cloud data, including noise removal, smoothing and format conversion, and then extract key geometric features from the preprocessed standard point cloud data to obtain track data, which is then rotated and corrected by a data correction module;

[0065] Step 4: Calibrate based on the height difference between the sleeper and the track to obtain the angle between the lidar and the ground. Perform target detection based on the angle between the lidar and the ground, including measuring the top surface of the ballast bed, the width of the ballast bed, the height of the ballast shoulder, the slope of the ballast side slope, and the height difference between the ballast and the sleeper, and generate and output structured measurement data.

[0066] It also includes an infrared monitoring camera, which synchronously collects roadbed images and aligns them with the point cloud data.

[0067] The laser radar is used to scan the track bed, ballast and sleepers to obtain original three-dimensional point cloud data, and the standard point cloud data is obtained by fitting the POS data with the original three-dimensional point cloud data, including:

[0068] By receiving the original lidar data and parsing it to generate point cloud data, the pbAI::messageToAI message structure is used to parse the received binary data into recognizable structured data.

[0069] The preprocessing of the obtained standard point cloud data, including noise removal, smoothing and format conversion, includes:

[0070] The point cloud data is preprocessed, noise is removed and downsampled, and outliers are removed by calculating the number of points in the neighborhood of each point in the point cloud. MeanK = 50 and StddevMulThresh = 1.0 are set to delete noise points that do not meet statistical characteristics.

[0071] The key geometric feature extraction of the pre-processed standard point cloud data to obtain the track data includes:

[0072] For the filtered point cloud data, according to the range of the x coordinates in the point cloud data, the point cloud in the track area is screened out, and the minimum value minX=-5 and the maximum value maxX=5 of the x coordinate of the track area are set;

[0073] Use the FindRailwayTracks function to extract the left and right track points of the track; by calculating the x-coordinates of the center points of the left and right track points, check whether the distance of the track is within the set range. If the distance is within the set range, the track detection is successful.

[0074] The rotation correction of the orbit data by the data correction module includes:

[0075] Rotation correction: Calculate the rotation angle of the track and rotate the point cloud along the Y axis by this angle so that the track is parallel to the horizontal plane in the point cloud data. The rotation angle is calculated by the atan function, and the formula is:

[0076]

[0077] For each (x,y,z), the coordinate after rotation around the Y axis is calculated as:

[0078] x′=x·cos(θ)+z·sin(θ)

[0079] z′=-x·sin(θ)+z·cos(θ)

[0080] Among them, the y coordinate remains unchanged;

[0081] Track height calibration: By recalculating the corrected point cloud data, determine whether the left and right track point clouds are at the same height. If the left and right track point clouds are at the same height, the calibration is completed, otherwise readjustment is required.

[0082] The calibration based on the height difference between the sleeper and the track to obtain the angle between the laser radar and the ground includes:

[0083] The corrected and rotated point cloud set is screened according to the middle value of the X coordinates of the left and right rails, and the middle 80% of the point cloud data of the left and right rails is taken to screen out the point cloud set of the sleeper or ballast;

[0084] Calculate the standard deviation of the middle point cloud and judge whether it is smooth by the standard deviation. If the standard deviation is small and smooth, it is a sleeper; if the standard deviation is large and not smooth, it is ballast. The formula is as follows:

[0085] Calculate the mean. This part calculates the cumulative sum of all z coordinates and divides it by the number of points to get the mean, where the z coordinate of each point is {p1,p2,…,p n}, its mean (z i ) is (μ z )for:

[0086]

[0087] Variance σ 2 for:

[0088]

[0089] The standard deviation σ is:

[0090]

[0091] The angle between the laser radar and the ground is:

[0092]

[0093] The target detection based on the angle between the laser radar and the ground includes measuring the top surface of the ballast bed, the width of the ballast bed, the height of the ballast shoulder, the slope of the ballast side slope, and the height difference between the ballast and the sleeper, and generating and outputting structured measurement data, including:

[0094] Determine the top surface and overall width of the roadbed through the X value range of the point cloud data; obtain the width of the roadbed by calculating the difference between the maximum and minimum X values;

[0095] Ballast shoulder height measurement: For the point cloud of the ballast shoulder area, calculate the average height of the point cloud of the ballast shoulder area relative to the reference plane or track plane. Obtain the ballast shoulder height;

[0096] Ballast slope measurement: Select point cloud data in the slope area and use linear regression analysis to calculate the slope. The linear regression formula is as follows:

[0097]

[0098] Vertical height measurement of the ballast on the top surface of the trackbed and the top surface of the sleeper: By comparing the average Z value of the point cloud of the top surface of the sleeper and the top surface of the ballast, the height difference between the ballast on the top surface of the trackbed and the top surface of the sleeper is calculated.

[0099] A ballast detection system, applying the ballast detection method, comprises an inertial navigation unit, a laser radar, an infrared monitoring camera, a data processing module, a communication module and a ground terminal module;

[0100] The inertial navigation unit, laser radar, infrared monitoring camera and communication module are respectively connected to the data processing module, and the ground terminal module is communicatively connected to the communication module.

[0101] Specifically, LiDAR installation and data collection

[0102] Laser radar installation, the laser radar is installed on the roof of the inspection vehicle, and its angle with the ground is adjustable. In order to ensure the accuracy of data collection, the installation angle of the laser radar needs to be calibrated.

[0103] Collection of inertial navigation and positioning data,After the inspection vehicle is started, the inertial navigation system begins to collect navigation data, and the GPS module obtains the vehicle's positioning data.

[0104] like Figure 2The structural diagram of the ballast dynamic detection system, which obtains 3D point cloud data by scanning the roadbed, ballast and rails through the laser radar. The angle between the laser radar and the ground is calculated and calibrated by the distance between the top surface of the rail and the sleeper. The distance from the top surface of the rail to the top surface of the sleeper is known, and combined with the 3D point cloud data of the ballast, the vertical height between the top surface of the roadbed and the top surface of the sleeper and the slope of the roadbed side slope are calculated.

[0105] Specific implementation methods and steps. The working principle and data acquisition mode of LiDAR. LiDAR is a sensor that obtains distance information by emitting lasers and measuring the time it takes for them to reflect back. Its shooting distance is about 6 to 7 meters, and it works in line scanning mode. In line scanning mode, the LiDAR scans along the roadbed to collect three-dimensional information of terrain and objects with high precision. Specifically, the data collected by the LiDAR includes the three-dimensional coordinates (x, y, z) of each point in the point cloud and the intensity value (Intensity) of the laser signal.

[0106] The hard trigger data collection method of the inspection vehicle. In actual applications, the inspection vehicle uses hard triggering to collect data. Every 250 mm of driving, the system will trigger a data collection to ensure the spatial resolution of data collection. Each collection contains approximately 1,500 point cloud data. In order to ensure the spatiotemporal continuity and integrity of the data, the collection frequency of the lidar is set to greater than 120 frames per second (fps), which means that more than 120 data can be collected per second.

[0107] Technical requirements and applications of data collection, this high-frequency, high-density point cloud data collection method can effectively capture the slight changes in the roadbed and surrounding environment, and provide rich basic data for subsequent data processing and analysis. These data can not only be used for the inspection and maintenance of the roadbed structure, but also for the generation of high-precision three-dimensional maps.

[0108] Data parsing and reception, the system receives the original lidar data and generates point cloud data after parsing for subsequent processing. The data is serialized through protobuf and transmitted over the network. The system first deserializes the data and converts the binary data into a pcl::PointCloud point cloud data structure. The specific process includes:

[0109] Data deserialization: Use the pbAI::messageToAI message structure to parse the received binary data into recognizable structured data.

[0110] Timestamp recording and correspondence: Processing of generated timestamps and received timestamps to ensure the real-time nature of data.

[0111] Point cloud data preprocessing, such as Figure 3Preprocessing flowchart; point cloud data is preprocessed, noise is removed and downsampled.

[0112] Main filtering methods:

[0113] Voxel Grid Filter: Downsample the point cloud by setting the cube voxel size (0.01f) to reduce the amount of data.

[0114] Statistical Outlier Removal: Remove outliers by counting the number of points in the neighborhood of each point in the point cloud. Set MeanK = 50 and StddevMulThresh = 1.0 to remove noise points that do not meet statistical characteristics.

[0115] Extract track information such as Figure 4 Feature extraction flow chart. After the point cloud is filtered, the key information of the railway track is extracted.

[0116] The main process includes:

[0117] Rough filtering by region: Filter the point cloud in the area where the track is located according to the range of the x coordinates in the point cloud. Set the minimum x coordinate of the track area minX = -5 and the maximum x coordinate maxX = 5.

[0118] Track Identification: Use the FindRailwayTracks function to extract the left and right track points of the track. The algorithm is based on the height information and classifies the points that meet the track height threshold into left track points and right track points.

[0119] Track distance check: By calculating the x coordinates of the center points of the left and right track points, check whether the track distance is within a reasonable range (the national standard track gauge is 1435mm). If the distance is in compliance, the track detection is considered successful.

[0120] Rotation Correction and Orbit Calibration

[0121] Purpose: In order to improve the detection accuracy, the system performs orbital rotation correction and height calibration: Figure 5 The data correction flow chart includes the following main processes:

[0122] Rotation correction: Calculate the rotation angle of the track (based on the average height difference between the two sides of the track) and rotate the point cloud along the Y axis by this angle so that the track is parallel to the horizontal plane in the point cloud data. The rotation angle is calculated by the atan function in degrees, and the formula is:

[0123]

[0124] For each (x,y,z), the coordinate after rotation around the Y axis is calculated as:

[0125] x′=x·cos(θ)+z·sin(θ)

[0126] z′=-x·sin(θ)+z·cos(θ)

[0127] Among them, the y coordinate remains unchanged because it is carried out around the y axis and the y direction will not be affected.

[0128] Track height calibration: By recalculating the corrected point cloud data, determine whether the left and right track point clouds are at the same height. If the left and right track point clouds are at the same height, it means that the correction effect is good, otherwise it needs to be readjusted.

[0129] Calibration

[0130] Purpose: Using the characteristics of the certain height difference between the track and the sleeper, the angle is first calibrated. The angle here refers to the angle between the laser radar and the track plane, which is prepared in order to obtain the vertical physical distance. Figure 6 The calibration flow chart is shown.

[0131] The main process includes:

[0132] Locating the sleeper: Filter the corrected and rotated point cloud set according to the middle value of the X coordinates of the left and right rails, take the middle 80% of the point cloud data of the left and right rails, and filter out the point cloud set of the sleeper or ballast.

[0133] Determine whether it is a sleeper or ballast: Calculate the standard deviation of the middle point cloud and determine whether it is smooth by the standard deviation. If the standard deviation is small and smooth, it is a sleeper; if the standard deviation is large and not smooth, it is ballast. The formula is as follows:

[0134] First, calculate the mean. This part calculates the cumulative sum of all z coordinates and divides it by the number of points to get the mean. The mean formula is:

[0135] For a set of points, each of which has a z coordinate of {p1,p2,…,p n}, its mean (z i ) can be expressed as (μ z )The specific formula is:

[0136]

[0137] Then calculate the variance, which represents the degree of dispersion of the data points from their mean. For a set of points' `z` coordinates, the variance (σ 2 ) is:

[0138]

[0139] Finally, calculate the standard deviation. The standard deviation formula is:

[0140] The standard deviation is the square root of the variance and is used to measure the degree of dispersion of the data. The standard deviation (σ) is calculated as:

[0141]

[0142] Once we find the sleepers, we use the inverse sine trigonometric function to calculate the angle between the lidar and the ground based on the fixed height difference between the track and the sleepers. This calculation ensures the high accuracy of the lidar data.

[0143]

[0144] Update the configuration file After calculating the angle, update the configuration file, such as Figure 7 The target detection flow chart shown in the figure records the calculated angle and the height of the sleeper. These data will be used for the subsequent calculation of the ballast height difference to ensure the accuracy of measurement and analysis.

[0145] Purpose: To extract key geometric features from point cloud data and measure the top surface and overall width of the ballast bed, the height of the ballast shoulder, the slope of the ballast side slope, and the vertical height between the ballast on the top surface of the ballast bed and the top surface of the sleeper.

[0146] The main process includes:

[0147] Track bed top surface and overall width measurement: Use the X value range of the point cloud data to determine the track bed top surface and overall width. By calculating the difference between the maximum and minimum X values, the track bed width is obtained.

[0148] Ballast shoulder height measurement: For the point cloud of the ballast shoulder area, calculate its average height relative to the reference surface or track plane. This height reflects the ballast shoulder height.

[0149] Ballast slope gradient measurement: Select point cloud data in the slope area and apply linear regression analysis to calculate the slope. This step can quantify the inclination of the ballast slope.

[0150] The linear regression formula is as follows:

[0151]

[0152] Vertical height measurement of the ballast on the top of the track bed and the top of the sleeper: By comparing the average Z value of the point cloud of the top of the sleeper and the top of the ballast, the height difference between the two is calculated. This height difference is an important indicator for evaluating the status of the track bed.

[0153] Result output and correction, such as Figure 8 The target detection flowchart shown:

[0154] Purpose: Output measurement results and optimize algorithm parameters based on actual measurement data.

[0155] The main process is as follows:

[0156] Result structured output: Organize the measurement results (such as width, height, slope, etc.) into a structured data format for further analysis and reporting. This step ensures the readability and ease of use of the data.

[0157] Performance evaluation and calibration: Evaluate the accuracy and reliability of the algorithm by comparing it with standard measurement values. Based on the evaluation results, adjust the algorithm parameters and calibrate the system to improve measurement accuracy.

[0158] The above is only a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the form disclosed herein, and should not be regarded as excluding other embodiments, but can be used in various other combinations, modifications and environments, and can be modified within the scope of the concept described herein through the above teachings or the technology or knowledge of the relevant field. The changes and modifications made by those skilled in the art shall not deviate from the spirit and scope of the present invention, and shall be within the scope of protection of the claims attached to the present invention.

Claims

1. A ballast detection method, characterized in that: The steps include: Step 1: Obtain navigation data during vehicle movement through the inertial navigation unit, and combine it with GPS positioning data to obtain POS data through fusion calculation; Step 2: Scan the track bed, ballast and sleepers with a laser radar to obtain original 3D point cloud data, and obtain standard point cloud data by fitting the POS data with the original 3D point cloud data; Step 3: preprocess the obtained standard point cloud data, including noise removal, smoothing and format conversion, and then extract key geometric features from the preprocessed standard point cloud data to obtain track data, which is then rotated and corrected by a data correction module; Step 4: Calibrate based on the height difference between the sleeper and the track to obtain the angle between the lidar and the ground. Perform target detection based on the angle between the lidar and the ground, including measuring the top surface of the ballast bed, the width of the ballast bed, the height of the ballast shoulder, the slope of the ballast side slope, and the height difference between the ballast and the sleeper, and generate and output structured measurement data.

2. A ballast detection method according to claim 1, characterized in that: It also includes an infrared monitoring camera, which synchronously collects roadbed images and aligns them with the point cloud data.

3. A ballast detection method according to claim 1, characterized in that: The laser radar is used to scan the track bed, ballast and sleepers to obtain original three-dimensional point cloud data, and the standard point cloud data is obtained by fitting the POS data with the original three-dimensional point cloud data, including: By receiving the original lidar data and parsing it to generate point cloud data, the pbAI::messageToAI message structure is used to parse the received binary data into recognizable structured data.

4. A ballast detection method according to claim 3, characterized in that: The preprocessing of the obtained standard point cloud data, including noise removal, smoothing and format conversion, includes: The point cloud data is preprocessed, noise is removed and downsampled, and outliers are removed by calculating the number of points in the neighborhood of each point in the point cloud. MeanK = 50 and StddevMulThresh = 1.0 are set to delete noise points that do not meet statistical characteristics.

5. A ballast detection method according to claim 4, characterized in that: The key geometric feature extraction of the pre-processed standard point cloud data to obtain the track data includes: For the filtered point cloud data, according to the range of the x coordinates in the point cloud data, the point cloud in the track area is screened out, and the minimum value minX=-5 and the maximum value maxX=5 of the x coordinate of the track area are set; Use the FindRailwayTracks function to extract the left and right track points of the track; by calculating the x-coordinates of the center points of the left and right track points, check whether the distance of the track is within the set range. If the distance is within the set range, the track detection is successful.

6. A ballast detection method according to claim 5, characterized in that: The rotation correction of the orbit data by the data correction module includes: Rotation correction: Calculate the rotation angle of the track and rotate the point cloud along the Y axis by this angle so that the track is parallel to the horizontal plane in the point cloud data. The rotation angle is calculated by the atan function, and the formula is: For each (x, y, z), the coordinate after rotation around the Y axis is calculated as: x′=x·cos(θ)+z·sin(θ) z′=-x·sin(θ)+z·cos(θ) Among them, the y coordinate remains unchanged; Track height calibration: By recalculating the corrected point cloud data, determine whether the left and right track point clouds are at the same height. If the left and right track point clouds are at the same height, the calibration is completed, otherwise readjustment is required.

7. A ballast detection method according to claim 6, characterized in that: The calibration based on the height difference between the sleeper and the track to obtain the angle between the laser radar and the ground includes: The corrected and rotated point cloud set is screened according to the middle value of the X coordinates of the left and right rails, and the middle 80% of the point cloud data of the left and right rails is taken to screen out the point cloud set of the sleeper or ballast; Calculate the standard deviation of the middle point cloud and judge whether it is smooth by the standard deviation. If the standard deviation is small and smooth, it is a sleeper; if the standard deviation is large and not smooth, it is ballast. The formula is as follows: Calculate the mean. This part calculates the cumulative sum of all z coordinates and divides it by the number of points to get the mean, where the z coordinate of each point is {p1, p2, ..., p n }, its mean (z i ) is (μ z )for: Variance σ 2 for: The standard deviation σ is: The angle between the laser radar and the ground is:

8. A ballast detection method according to claim 7, characterized in that: The target detection based on the angle between the laser radar and the ground includes measuring the top surface of the ballast bed, the width of the ballast bed, the height of the ballast shoulder, the slope of the ballast side slope, and the height difference between the ballast and the sleeper, and generating and outputting structured measurement data, including: Determine the top surface and overall width of the roadbed through the X value range of the point cloud data; obtain the width of the roadbed by calculating the difference between the maximum and minimum X values; Ballast shoulder height measurement: For the point cloud of the ballast shoulder area, calculate the average height of the point cloud of the ballast shoulder area relative to the reference plane or track plane. Obtain the ballast shoulder height; Ballast slope measurement: Select point cloud data in the slope area and use linear regression analysis to calculate the slope. The linear regression formula is as follows: Vertical height measurement of the ballast on the top surface of the trackbed and the top surface of the sleeper: By comparing the average Z value of the point cloud of the top surface of the sleeper and the top surface of the ballast, the height difference between the ballast on the top surface of the trackbed and the top surface of the sleeper is calculated.

9. A ballast detection system, characterized in that: A ballast detection method according to any one of claims 1 to 8, comprising an inertial navigation unit, a laser radar, an infrared monitoring camera, a data processing module, a communication module and a ground terminal module; The inertial navigation unit, laser radar, infrared monitoring camera and communication module are respectively connected to the data processing module, and the ground terminal module is communicatively connected to the communication module.

Images