Track measurement method, device and storage medium based on line structured light

Abstract

This invention relates to a method, apparatus, and storage medium for track measurement based on line structured light. The method includes: Step S1: Filtering using depth threshold filtering and bilateral filtering based on spatial distance and reflection intensity differences; Step S2: Dividing the point cloud into multiple sections, selecting candidate track regions within each section based on track height continuity constraints and normal information; Step S3: Obtaining the track point cloud for each section; Step S4: Determining whether each section is a valid section based on minimum number constraints and height continuity constraints; Step S5: Performing interpolation correction to obtain the track point cloud for each invalid section; Step S6: Registering the track point cloud of each section with a standard track template to obtain the roll and pitch angles of each section. Compared with existing technologies, this invention can achieve accurate extraction and stable identification of track point clouds while ensuring computational efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of track posture detection, in particular to a track measurement method and device based on line structured light and a storage medium. BACKGROUND

[0002] At present, the detection and analysis of railway tracks mainly rely on non-contact three-dimensional acquisition technologies such as laser scanning, stereo vision and structured light measurement. Among them, the laser scanning system can obtain high-precision track geometric data in static or low-speed environment, but when the train is running at high speed, due to the limited scanning frequency, the point cloud density is significantly reduced, and the track surface details are seriously lost, especially in the small-scale component area such as fasteners and sleepers, the contour is discontinuous or the structure is blurred, which is difficult to meet the real-time and precision requirements of high-speed detection. Although the stereo vision method can recover the spatial form of the track through binocular imaging, it is affected by environmental light, track surface reflection and vehicle vibration, etc., and its imaging quality is unstable, and it is easy to miss and miss in complex lighting scenes such as tunnels and subways.

[0003] In recent years, line structured light technology has gradually been applied to railway track detection due to its high measurement accuracy and data acquisition speed. For example, Chinese patent CN119197374A discloses a method and system for comprehensive analysis of existing line state based on a three-dimensional mobile scanning system. This method projects a line laser onto the track surface and recovers the three-dimensional point cloud information of the track according to the deformation characteristics of the stripes in space. However, in the actual high-speed detection environment, the existing track point cloud processing methods based on structured light still have the following main defects:

[0004] (1) The accuracy of track area segmentation is insufficient. Traditional track extraction methods are mostly based on gray level, light intensity or texture features for region recognition. However, line structured light may cause stripe drift, distortion and local spot loss under high-speed motion, resulting in blurred boundaries between track and non-track areas in point cloud, and unstable segmentation results.

[0005] (2) The cross-section point cloud is discontinuous. In the high-speed acquisition process, the interval of structured light scanning lines is affected by train speed and sampling frequency, resulting in data gaps or overlaps between cross-sections, poor spatial continuity of track point cloud, and affecting subsequent overall extraction and reconstruction.

[0006] In summary, the existing track detection technology based on line structured light still has problems such as low track area recognition accuracy, poor point cloud continuity and insufficient algorithm stability under high-speed dynamic acquisition conditions, which cannot meet the needs of real-time detection and high-precision data extraction of railway tracks. Therefore, how to fully utilize the geometric characteristics of line structured light data to achieve accurate extraction and stable identification of track point cloud while ensuring computational efficiency is a technical problem to be solved in the field. SUMMARY

[0007] The purpose of this invention is to address the shortcomings of the prior art by providing a track measurement method, device, and storage medium based on line structured light. Under high-speed operating conditions, it can fully utilize the geometric characteristics of line structured light data and quickly and stably extract the point cloud of the track region by geometrically constraining the point cloud cross-section, thereby reducing environmental interference and improving extraction efficiency and accuracy.

[0008] The objective of this invention can be achieved through the following technical solutions:

[0009] A trajectory measurement method based on line structured light, comprising:

[0010] Step S1: Obtain the raw point cloud data acquired by structured light scanning, and filter it using depth threshold filtering and bilateral filtering based on spatial distance and reflection intensity difference.

[0011] Step S2: Divide all filtered point clouds into multiple sections according to the scanning direction, and obtain the candidate track regions in each section based on the track height continuity constraint and combined with the normal information.

[0012] Step S3: In the candidate orbit regions of each section, the lateral coordinate distribution of the point cloud is statistically analyzed and the density peak is identified. Clustering is performed within a first pre-configured lateral distance range with the density peak as the center to obtain the orbit point cloud of each section.

[0013] Step S4: If the number of track point clouds in any section is greater than the first pre-configured lower limit, and the difference between the maximum height of the track point cloud in the section and the maximum height of the track point cloud in the adjacent section, and the difference between the minimum height of the track point cloud in the section and the minimum height of the track point cloud in the adjacent section are both less than the pre-configured second difference threshold, then the section is considered a valid section; otherwise, the section is considered an invalid section.

[0014] Step S5: For all invalid sections, perform interpolation correction based on the orbit point clouds of their adjacent valid sections to obtain the orbit point cloud of each invalid section;

[0015] Step S6: By registering the track point cloud of each section with the standard track template, the roll angle and pitch angle of each section are obtained.

[0016] Step S1 includes:

[0017] Step S1-1: Obtain the raw point cloud data acquired by structured light scanning;

[0018] Step S1-2: Filter out point clouds located outside the first depth threshold range;

[0019] Steps S1-3: Calculate the spatial distance weight between each point cloud and other point clouds in its neighborhood based on the spatial distance threshold, and calculate the reflection intensity weight between each point cloud and other point clouds in its neighborhood based on the reflection intensity difference threshold.

[0020] Step S1-4: Fuse the spatial distance weight and the reflection intensity weight to obtain the fusion weight of each point cloud and other point clouds in its neighborhood, and based on the fusion weight, superimpose the position information of other point clouds in its neighborhood onto the position of each point cloud to complete the position update of all point clouds;

[0021] Step S1-5: Determine whether there are isolated noise points with abnormal reflection intensity or abnormal spatial location in all updated point clouds. If yes, proceed to step S1-6; otherwise, proceed to step S2.

[0022] Step S1-6: Adjust the spatial distance threshold and / or the reflection intensity difference threshold, and return to step S1-3.

[0023] Steps S1-5 include:

[0024] Step S1-5-1: Calculate the distance between each point cloud and its nearest neighbor point cloud as the nearest neighbor distance for each point cloud, and calculate the mean and standard deviation of the nearest neighbor distance for the global point cloud. Point clouds that satisfy the following conditions are identified as isolated noise points with spatial anomalies:

[0025]

[0026] in: The nearest neighbor distance of the point cloud. The mean of the nearest neighbor distances of the global point cloud. is the standard deviation of the nearest neighbor distance in the global point cloud;

[0027] Step S1-5-2: Calculate the reflection intensity of each point cloud, compare the difference between the reflection intensity of each point cloud and the average reflection intensity of all point clouds in its neighborhood, and if it is greater than the pre-configured first difference threshold, then this point cloud is regarded as an abnormal point in reflection intensity.

[0028] Step S1-5-3: Determine whether there are isolated noise points with abnormal reflection intensity or abnormal spatial location in all updated point clouds. If yes, proceed to step S1-6; otherwise, proceed to step S2.

[0029] Step S2 includes:

[0030] Step S2-1: Divide all filtered point clouds into multiple sections according to the scanning direction;

[0031] Step S2-2: Perform height statistics on the point cloud in all sections, and filter the preliminary candidate areas of the track in the point cloud in each section based on the track height continuity constraint. The track height continuity constraint is: the absolute value of the difference between the height of the lowest point of the track in any section and the height of the lowest point of the track in the adjacent section is less than 15 mm.

[0032] Step S2-3: Sort all point clouds in the preliminary candidate area of ​​the track in the cross section according to the horizontal coordinate from smallest to largest, and calculate the unit normal vector of all point clouds in the preliminary candidate area of ​​the track in each cross section to form a normal sequence.

[0033] Step S2-4: Process the normal sequence using a sliding window with length L and step size 1. Calculate the average change amplitude of the elements in the normal sequence within each sliding window step. If the average change amplitude is less than the pre-configured amplitude threshold, then include the point cloud corresponding to the elements in the normal sequence within that sliding window step into the orbit candidate region of the section.

[0034] Step S2-2 includes:

[0035] Step S2-2-1: Perform height statistics on the point clouds in all sections, and calculate the average height of the lowest point clouds in each section as the first effective height of the section.

[0036] Step S2-2-2: Initialize the first state value of all cross sections to state one;

[0037] Step S2-2-3: Select a section with a first state value of state one, and determine whether the absolute value of the difference between its first effective height and the first effective height of the adjacent section is less than 15 mm. If yes, proceed to step S2-2-4; otherwise, proceed to step S2-2-5.

[0038] Step S2-2-4: Set the first state value of the currently selected section to state two, and determine whether there is a section with the first state value to state one. If yes, return to step S2-2-3; otherwise, use the remaining point cloud of all sections as the preliminary candidate area of ​​the track for that section and execute step S2-3.

[0039] Step S2-2-5: Delete the point cloud of the currently selected section and its adjacent sections, so that the absolute value of the difference between the first effective height of the currently selected section and the first effective height of the adjacent sections is less than 15 mm. Update the first effective height of the currently selected section and its adjacent sections, and set the first state value of the currently selected section to state two. Initialize the first state value of all other sections to state one, and return to step S2-2-3.

[0040] The average change range is:

[0041]

[0042] in: The average change range, The length of the sliding window. The normal sequence within the sliding window i One element, The normal sequence within the sliding window i +1 element.

[0043] The interpolation correction method in step S5 is linear interpolation.

[0044] The process of obtaining the roll and pitch angles of a single cross-section in step S6 includes:

[0045] Step S6-1: Set the initial roll and pitch angles of the standard track template to 0, and align the center of the track point cloud of the standard track template and the cross-section.

[0046] Step S6-2: Change the roll angle and pitch angle, and combine the fixed step size translation of the standard track template to calculate the sum of the distances from each track point cloud in the section to the standard track template. The roll angle and pitch angle of the standard track template when the sum of the distances is the smallest are taken as the track roll angle and pitch angle of the section.

[0047] A track measurement device based on line structured light includes a memory, a processor, and a program stored in the memory, wherein the processor executes the program to implement the method described above.

[0048] A storage medium having a program stored thereon, which, when executed, implements the method described above.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] 1. Under high-speed operating conditions, it can make full use of the geometric characteristics of line structured light data, and through geometric constraints on the point cloud cross-section, it can quickly and stably achieve accurate extraction of point clouds in the orbital region, reduce environmental interference, and improve extraction efficiency and accuracy.

[0051] 2. Using a single frame point cloud as the basic processing unit, depth thresholding and bilateral filtering are applied sequentially to remove noise and smooth the data. Depth thresholding is used to initially remove invalid points and anomalous reflection points outside the structured light projection range. Its threshold can be estimated based on the depth range of the measured object and the camera's internal and external parameters. Bilateral filtering is used to smooth the point cloud in local areas while preserving boundary features. Its basic idea is to combine the spatial distance and reflection intensity differences to perform a weighted average of the points, thereby effectively suppressing high-frequency noise.

[0052] 3. A specific anomaly detection algorithm is introduced, including threshold judgments based on nearest neighbor distance and reflection intensity differences, which can accurately identify and isolate isolated noise points with abnormal reflection intensity and spatial location anomalies. This detection mechanism prevents anomalies from interfering with subsequent track screening, improving the cleanliness and consistency of point cloud data. By statistically analyzing the characteristics of the global point cloud, this method exhibits strong adaptability and can effectively address issues such as missing light spots or stripe distortion in high-speed acquisition, enhancing the overall system's anti-interference capability and reliability.

[0053] 4. By fully utilizing the spatial features of the track, including its height, parallelism, and longitudinal continuity, and through segment-by-segment screening, clustering, and continuity determination, accurate identification and extraction of track point clouds are achieved. The sliding window processing using normal sequences captures smooth geometric features of the track surface, reducing false detections in non-track areas. This method improves the accuracy of track point cloud extraction, especially under high-speed conditions when the point cloud is discontinuous, effectively maintaining segmental continuity and enhancing the stability and efficiency of track identification.

[0054] 5. Through iterative height statistics and state management mechanisms, the point cloud selection within the cross-section is dynamically adjusted to ensure that the track height continuity constraint (e.g., the height difference between adjacent cross-sections is less than 15 mm) is met. This adaptive processing can effectively handle local data loss or anomalies, improving the reliability of the initial candidate track area. By progressively deleting inconsistent point clouds and updating the effective height, the algorithm's adaptability to track geometric changes is enhanced, error accumulation is reduced, and the continuity and accuracy of track extraction are guaranteed.

[0055] 6. A formula for calculating the average variation amplitude is defined, providing a quantitative and objective standard for evaluating the smoothness of the normal sequence. This formula, by calculating the average change of the normal dot product, can effectively identify uniform regions on the track surface, avoiding misjudgments caused by noise or irregular shapes. This mathematical processing makes the normal information screening more consistent and repeatable, improves the accuracy of track candidate region screening, and optimizes the effect of subsequent clustering extraction. Attached Figure Description

[0056] Figure 1 This is a schematic diagram of the hardware infrastructure upon which the method of this invention depends;

[0057] Figure 2 This is a schematic diagram of the track separation results;

[0058] Figure 3 This is a schematic diagram of the track cross-section;

[0059] Figure 4 A schematic diagram illustrating the principle of template registration;

[0060] Figure 5 A diagram showing the track matching error;

[0061] Figure 6 This is a schematic diagram of the test results;

[0062] Figure 7 This is a schematic diagram of the main steps of the method of the present invention;

[0063] The components include: 1. Structured light sensor, 2. Mounting bracket, 3. Adapter plate, and 4. Track to be measured. Detailed Implementation

[0064] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0065] A track measurement method based on line structured light acquires real-time 3D point cloud data of the track surface by installing a structured light 3D imaging device on the lower side of the train body while the train is running at approximately 50 km / h. To address noise and outliers in the point cloud, a combination of depth threshold filtering and bilateral filtering is used to improve the accuracy and continuity of track geometry information. Based on the spatial geometric constraints and cross-sectional coherence of the point cloud, an automatic track point cloud extraction method is proposed. Through cross-sectional grouping, geometric filtering, and coherence determination, accurate extraction and cross-sectional identification of the track area point cloud are achieved. Using a standard track template and applying fixed rotation and translation optimization constraints, attitude correction and error calculation are performed on the track cross-section, thereby obtaining high-precision reconstruction results of the track geometry. This invention enables non-contact, high-precision measurement of track morphology under high-speed operating conditions, providing reliable technical support for track condition detection and geometric deformation assessment.

[0066] like Figure 1 As shown, the hardware structure upon which the method of this application relies includes a structured light sensor 1, a mounting bracket 2, and an adapter plate 3, which are installed as a whole at the front of the track inspection vehicle for non-contact, high-precision measurement of the geometric state of the track 4 being measured.

[0067] Mounting bracket 2 is a crucial load-bearing structure connecting the structured light sensor 1 to the vehicle body, providing both rigid support and angle adjustment. The bracket is secured to the adapter plate with four M10 bolts and is constructed entirely of high-strength metal, ensuring high structural stability. The bracket can be adjusted ±10° left and right in the horizontal plane for rapid calibration based on track conditions, vehicle posture, or sensor viewing angle requirements.

[0068] The adapter plate 3 is located between the mounting bracket 2 and the rail vehicle platform, serving as a vibration damping buffer and structural transition. The bottom of the adapter plate 3 has three plates that can be adjusted laterally or longitudinally according to pre-drilled holes to accommodate different vehicle models and the range of structured light scanning. A 10 mm thick shock-absorbing rubber pad is installed between the adapter plate 3 and the rail vehicle platform to effectively absorb high-frequency vibrations generated during track operation.

[0069] like Figure 7 As shown, it includes:

[0070] Step S1: Acquire the raw point cloud data obtained from structured light scanning, and filter it using depth threshold filtering and bilateral filtering methods based on spatial distance and reflection intensity differences, including:

[0071] Step S1-1: Obtain the raw point cloud data acquired by structured light scanning;

[0072] Step S1-2: Filter out point clouds located outside the first depth threshold range. Generally, the first depth threshold range is [z min ,z max ];

[0073] Steps S1-3: Calculate the spatial distance weight between each point cloud and other point clouds in its neighborhood based on a spatial distance threshold, and calculate the reflection intensity weight between each point cloud and other point clouds in its neighborhood based on a reflection intensity difference threshold. The radius of the neighborhood is r, and the spatial distance weights are:

[0074] ,

[0075] in: For point clouds And point cloud Spatial distance weight, This is the spatial distance threshold;

[0076] The reflection intensity weight is:

[0077] ,

[0078] in: For point clouds And point cloud The reflection intensity weight, For point clouds The intensity of reflection, For point clouds The intensity of reflection, This is the threshold for the difference in reflection intensity.

[0079] Step S1-4: The spatial distance weight and reflection intensity weight are fused to obtain the fusion weight of each point cloud and other point clouds in its neighborhood. Based on the fusion weight, the position information of other point clouds in its neighborhood is superimposed on the position of each point cloud to complete the position update of all point clouds. In this embodiment, the fusion method is multiplication.

[0080] Step S1-5: Determine whether there are isolated noise points with abnormal reflection intensity or abnormal spatial location in all updated point clouds. If yes, proceed to step S1-6; otherwise, proceed to step S2. Specifically, this includes:

[0081] Step S1-5-1: Calculate the distance between each point cloud and its nearest neighbor point cloud as the nearest neighbor distance for each point cloud, and calculate the mean and standard deviation of the nearest neighbor distance for the global point cloud. Point clouds that satisfy the following conditions are identified as isolated noise points with spatial anomalies:

[0082]

[0083] in: The nearest neighbor distance of the point cloud. The mean of the nearest neighbor distances of the global point cloud. is the standard deviation of the nearest neighbor distance in the global point cloud;

[0084] Step S1-5-2: Calculate the reflection intensity of each point cloud, compare the difference between the reflection intensity of each point cloud and the average reflection intensity of all point clouds in its neighborhood, and if it is greater than the pre-configured first difference threshold, then this point cloud is regarded as an abnormal point in reflection intensity.

[0085] Step S1-5-3: Determine whether there are isolated noise points with abnormal reflection intensity or abnormal spatial location in all updated point clouds. If yes, proceed to step S1-6; otherwise, proceed to step S2.

[0086] Step S1-6: Adjust the spatial distance threshold and / or the reflection intensity difference threshold, and return to step S1-3. Generally, this can be done by zooming in or zooming out. In this embodiment, the spatial distance threshold and the reflection intensity difference threshold are adjusted simultaneously.

[0087] During structured light scanning, the acquired raw point cloud data often contains outliers, isolated points, and measurement error points due to factors such as differences in surface reflectivity, illumination conditions, sensor noise, and the complexity of the target surface geometry. To improve the accuracy of subsequent reconstruction and analysis, this application uses a single frame of point cloud as the basic processing unit when filtering structured light point clouds, and sequentially employs depth threshold filtering and bilateral filtering to remove noise and smooth the data.

[0088] Among them, depth threshold filtering is used to initially remove invalid points and reflection anomalies outside the structured light projection range. Its threshold can be estimated based on the depth range of the object being measured and the internal and external parameters of the camera. Bilateral filtering is used to smooth the point cloud in a local area while preserving boundary features. Its basic idea is to combine the spatial distance and the difference in reflection intensity to perform a weighted average of the points, thereby effectively suppressing high-frequency noise.

[0089] Step S2: Divide all filtered point clouds into multiple sections according to the scanning direction. Based on the orbital height continuity constraint and combined with normal information, obtain the orbital candidate regions within each section, including:

[0090] Step S2-1: Divide all filtered point clouds into multiple sections according to the scanning direction, with the scanning direction as the longitudinal reference. Divide the point cloud into continuous sections at fixed intervals. Each section represents the local cross-sectional structure of the track at that location, providing basic data for subsequent analysis. Each section is displayed in a different color, such as... Figure 3 As shown

[0091] Step S2-2: Perform height statistics on the point clouds within all cross-sections. Based on the track height continuity constraint, filter the point clouds within each cross-section to obtain preliminary candidate areas for the track. The track height continuity constraint is: the absolute value of the difference between the lowest point of the track within any cross-section and the lowest point of the track within an adjacent cross-section is less than 15 mm. Specifically, this includes:

[0092] Step S2-2-1: Perform height statistics on the point clouds in all sections, and calculate the average height of the lowest point clouds in each section as the first effective height of the section.

[0093] Step S2-2-2: Initialize the first state value of all cross sections to state one;

[0094] Step S2-2-3: Select a section with a first state value of state one, and determine whether the absolute value of the difference between its first effective height and the first effective height of the adjacent section is less than 15 mm. If yes, proceed to step S2-2-4; otherwise, proceed to step S2-2-5.

[0095] Step S2-2-4: Set the first state value of the currently selected section to state two, and determine whether there is a section with the first state value to state one. If yes, return to step S2-2-3; otherwise, use the remaining point cloud of all sections as the preliminary candidate area of ​​the track for that section and execute step S2-3.

[0096] Step S2-2-5: Delete the point cloud of the currently selected section and its adjacent sections, so that the absolute value of the difference between the first effective height of the currently selected section and the first effective height of the adjacent sections is less than 15 mm. Update the first effective height of the currently selected section and its adjacent sections, and set the first state value of the currently selected section to state two. Initialize the first state value of all other sections to state one, and return to step S2-2-3.

[0097] Step S2-3: Sort all point clouds in the preliminary candidate area of ​​the track in the cross section according to the horizontal coordinate from smallest to largest, and calculate the unit normal vector of all point clouds in the preliminary candidate area of ​​the track in each cross section to form a normal sequence.

[0098] Step S2-4: Process the normal sequence using a sliding window with length L and step size 1. Calculate the average change amplitude of the elements in the normal sequence within each sliding window step. If the average change amplitude is less than the pre-configured amplitude threshold, then include the point cloud corresponding to the elements in the normal sequence within that sliding window step into the orbit candidate region of the section.

[0099] In this embodiment, the average variation range is:

[0100]

[0101] in: The average change range, The length of the sliding window. The normal sequence within the sliding window i One element, The normal sequence within the sliding window i +1 element.

[0102] To accurately separate the orbital region point cloud from the raw point cloud obtained by structured light scanning, this invention proposes an orbital point cloud separation algorithm based on spatial geometric constraints and cross-sectional continuity. This algorithm fully utilizes the height, parallelism, and longitudinal continuity features of the orbit in its spatial morphology, achieving accurate identification and extraction of the orbital point cloud through cross-sectional filtering, clustering, and continuity determination. The orbital separation results are shown below. Figure 2 As shown.

[0103] Step S3: In the candidate track regions of each section, the lateral coordinate distribution of the point cloud is statistically analyzed and the density peak is identified. Clustering is performed within a first pre-configured lateral distance range centered on the density peak to obtain the track point cloud of each section. In this embodiment, the first pre-configured lateral distance range is ±80 mm.

[0104] Step S4: If the number of track point clouds in any section is greater than the first pre-configured lower limit, and the difference between the maximum height of the track point cloud in the section and the maximum height of the track point cloud in the adjacent section, and the difference between the minimum height of the track point cloud in the section and the minimum height of the track point cloud in the adjacent section are both less than the pre-configured second difference threshold, then the section is considered a valid section; otherwise, the section is considered an invalid section.

[0105] In this embodiment, the minimum number of the first pre-configured items is 150, and the second pre-configured difference threshold is 15 mm.

[0106] Step S5: For all invalid sections, interpolate and correct the orbit point clouds based on the orbit point clouds of their adjacent valid sections to obtain the orbit point clouds of each invalid section, ensuring the continuity and integrity of the orbit point clouds in the longitudinal direction.

[0107] In this embodiment, the interpolation correction method is linear interpolation.

[0108] Step S6: By registering the track point cloud of each section with the standard track template, the roll angle and pitch angle of each section are obtained.

[0109] The process of obtaining the track roll and pitch angles for a single cross-section in step S6 includes:

[0110] Step S6-1: Set the initial roll and pitch angles of the standard track template to 0, and align the center of the track point cloud of the standard track template and the cross-section.

[0111] The standard track template is specifically constructed based on the design parameters and standard geometric features of the track cross-section. The template is represented in the form of a two-dimensional discrete point cloud and is generated by pre-imported standard cross-section data, serving as the geometric reference for track matching.

[0112] Each cross-section includes the distribution of feature points at the top, waist, and bottom of the track, which can fully reflect the geometry of the track at that location.

[0113] Step S6-2: Change the roll angle and pitch angle, and combine the fixed step size translation of the standard track template to calculate the sum of the distances from each track point cloud in the section to the standard track template. The roll angle and pitch angle of the standard track template when the sum of the distances is the smallest are taken as the track roll angle and pitch angle of the section. The distance from the track point cloud to the standard track template can be selected as the distance of the point closest to the standard track template.

[0114] Specifically, after separating the orbit point cloud, to achieve high-precision identification and matching of orbit cross-sections, the separated orbit point cloud is automatically registered with a standard orbit template to realize attitude correction and error analysis of the orbit cross-sections, thereby ensuring the geometric accuracy of structured light measurement. The attitude optimization results of multiple orbit cross-sections are as follows:Figure 4 As shown.

[0115] After determining the optimal matching posture, the distances from all track points within the cross-section to the corresponding points on the template are calculated, resulting in an error distribution in millimeters. This invention performs error coloring on all track points without cropping or limiting, to fully reflect the overall deviation of the track cross-section. The resulting error distribution map can visually display the geometric deviations and wear characteristics of the track surface, such as... Figure 5 As shown.

[0116] Finally, the attitude angle, translation parameters, and corresponding error statistics for each cross-section are output, and the changes in each cross-section are continuously analyzed in the longitudinal direction to form a complete description of the orbital geometry, such as... Figure 6 As shown.

[0117] This application introduces template matching and attitude optimization techniques based on the extracted orbit point cloud results, achieving accurate registration and error visualization of orbit cross sections. This effectively improves the geometric consistency and detection accuracy of structured light measurements, providing a reliable data foundation for orbit three-dimensional deformation monitoring and health assessment.

[0118] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a 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 instructions 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 invention. 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.

Claims

1. A line structured light based track measurement method, characterized in that, The method comprises the following steps: Step S1: obtaining original point cloud data acquired by structured light scanning, and filtering the original point cloud data by using a depth threshold filtering method and a bilateral filtering method based on spatial distance and reflection intensity difference; Step S2: dividing all the filtered point clouds into multiple sections according to a scanning direction, and screening a track candidate area in each section based on track height continuity constraint and in combination with normal information; Step S3: in the track candidate area of each section, the transverse coordinate distribution of the point clouds is counted and a density peak value is identified, and the track point cloud of each section is obtained by clustering extraction within a first preconfigured transverse distance range centered on the density peak value; Step S4: if the number of track point clouds in any section is greater than a first preconfigured lower limit, and the difference between the maximum height of the track point clouds in the section and the maximum height of the track point clouds in the adjacent section, and the difference between the minimum height of the track point clouds in the section and the minimum height of the track point clouds in the adjacent section are all less than a second preconfigured difference threshold, the section is regarded as a valid section, otherwise, the section is regarded as an invalid section; Step S5: for all invalid sections, the track point clouds of the invalid sections are obtained by interpolation correction based on the track point clouds of the adjacent valid sections; Step S6: the track roll angle and pitch angle of each section are obtained by registering the track point clouds of each section with a standard track template; The step S1 comprises: Step S1-1: obtaining original point cloud data acquired by structured light scanning; Step S1-2: filtering out point clouds located outside a first depth threshold range; Step S1-3: calculating the spatial distance weight of each point cloud and other point clouds in its neighborhood according to spatial distance based on a spatial distance threshold, and calculating the reflection intensity weight of each point cloud and other point clouds in its neighborhood according to reflection intensity difference based on a reflection intensity difference threshold; Step S1-4: fusing the spatial distance weight and the reflection intensity weight to obtain the fusion weight of each point cloud and other point clouds in its neighborhood, and superimposing the position information of other point clouds in the neighborhood of each point cloud to the position of each point cloud based on the fusion weight, to complete the position update of all point clouds; Step S1-5: determining whether there are reflection intensity abnormal points or spatial position abnormal isolated noise points in all updated point clouds, if yes, executing step S1-6, otherwise, executing step S2; Step S1-6: adjusting the spatial distance threshold and / or the reflection intensity difference threshold, and returning to step S1-3.

2. The track measurement method based on line structured light according to claim 1, characterized in that, The step S1-5 comprises: Step S1-5-1: calculating the distance between each point cloud and its nearest neighbor point cloud as the nearest neighbor distance of each point cloud, and calculating the mean value and standard deviation of the nearest neighbor distance of all point clouds, and regarding the point cloud satisfying the following condition as a spatial position abnormal isolated noise point: wherein: is the nearest neighbor distance of the point cloud, is the mean of the nearest neighbor distances of the global point cloud, is the standard deviation of the nearest neighbor distances of the global point cloud; Step S1-5-2: calculating the reflection intensity of each point cloud, comparing the difference between the reflection intensity of each point cloud and the mean value of the reflection intensity of all point clouds in its neighborhood, and regarding the point cloud as a reflection intensity abnormal point if the difference is greater than a first preconfigured difference threshold. Step S1-5-3: judging whether there are abnormal reflection intensity points or spatial position abnormal isolated noise points in all updated point clouds, if yes, performing step S1-6, otherwise performing step S2.

3. The track measurement method based on line structured light according to claim 1, characterized in that, The step S2 comprises: Step S2-1: dividing all filtered point clouds into multiple sections according to the scanning direction; Step S2-2: performing height statistics on the point clouds in all sections, and screening the track preliminary candidate region from the point clouds in each section based on the track height continuity constraint, wherein the track height continuity constraint is that the absolute value of the height difference between the lowest part of the track in any section and the lowest part of the track in the adjacent section is less than 15 mm; Step S2-3: sorting all point clouds in the track preliminary candidate region of the section according to the transverse coordinates from small to large, and calculating the unit normal vector of all point clouds in the track preliminary candidate region of each section to form a normal sequence; Step S2-4: processing the normal sequence by using a sliding window with a length of L and a step of 1, calculating the average change amplitude of the elements in each step sliding window of the normal sequence, and if the average change amplitude is less than a preconfigured amplitude threshold, the point cloud corresponding to the elements in the step sliding window of the normal sequence is included in the track candidate region of the section.

4. The track measurement method based on line structured light according to claim 3, characterized in that, The step S2-2 comprises: Step S2-2-1: performing height statistics on the point clouds in all sections, and calculating the average height of the lowest multiple point clouds in the point clouds in each section as the first effective height of the section; Step S2-2-2: initializing the first state value of all sections as state one; Step S2-2-3: selecting a section with a first state value of state one, judging whether the absolute value of the height difference between the first effective height of the section and the first effective height of the adjacent section is less than 15 mm, if yes, performing step S2-2-4, otherwise, performing step S2-2-5; Step S2-2-4: setting the first state value of the currently selected section as state two, judging whether there is a section with a first state value of state one, if yes, returning to step S2-2-3, otherwise, taking the remaining point clouds of all sections as the track preliminary candidate region of the section, and performing step S2-3; Step S2-2-5: deleting the point clouds of the currently selected section and its adjacent sections, so that the absolute value of the height difference between the first effective height of the currently selected section and the first effective height of the adjacent section is less than 15 mm, updating the first effective height of the currently selected section and its adjacent sections, updating the first state value of the currently selected section as state two, initializing the first state value of all other sections as state one, and returning to step S2-2-3.

5. The track measurement method based on line structured light according to claim 3, characterized in that, The average change amplitude is: wherein: is the average variation amplitude, is the length of the sliding window, is the element of the normal sequence within the sliding window at position i is the element of the normal sequence within the sliding window at position is the element of the normal sequence within the sliding window at position i +1.

6. The track measurement method based on line structured light according to claim 1, characterized in that, The interpolation correction method in the step S5 is linear interpolation.

7. The track measurement method based on line structured light according to claim 1, wherein, The process of obtaining the track roll angle and pitch angle of a single section in the step S6 comprises: Step S6-1: setting the initial roll angle and pitch angle of the standard track template as 0, and aligning the centers of the standard track template and the track point cloud of the section; Step S6-2: transform the roll angle and the pitch angle, and combine the fixed step translation standard rail template, respectively calculate the sum of distances from each rail point cloud in the section to the standard rail template, and take the roll angle and the pitch angle of the standard rail template when the sum of distances is the smallest as the rail roll angle and the pitch angle of the section.

8. A line structured light based track measuring device comprising a memory, a processor, and a program stored in the memory, wherein the program comprises the following steps: The program, when executed by the processor, implements the method of any one of claims 1-7.

9. A storage medium having stored thereon a program, characterized by The program, when executed, implements the method of any one of claims 1-7.

Images