Multi-polarized light point cloud data fusion method and device for rail appearance status

Abstract

This invention discloses a method and device for fusing multi-polarized light point cloud data of rail appearance. The method comprises: merging data read from a polarization camera into rail profile data; dividing the rail profile data into multiple regions based on rail vertex coordinates and gauge point coordinates; fitting the rail profile data of the multiple regions using the random sampling consistency (RANSAC) algorithm to obtain fitting curves for the multiple regions; fusing the fitting curves of the multiple regions to obtain an optimal rail profile fitting curve; and measuring the rail profile using the optimal rail profile fitting curve. This invention effectively solves the problem of partial underexposure of rail laser cross-sections.

Description

Technical Field

[0001] The present invention relates to the field of structured light imaging technology, and in particular to a method and device for fusing multi-polarized light point cloud data of rail appearance status. Background Art

[0002] This section is intended to provide a background or context to the embodiments of the invention that are recited in the claims. No statement herein is admitted to be prior art by virtue of its inclusion in this section.

[0003] Online measurement of rail appearance, timely understanding of rail health, analysis of rail profile and wear trends, and detection of rail top surface abrasion defects are crucial for railway operations and maintenance. Rail profile measurement involves comparing measured rail profile data with standard rail profile data to determine indicators such as side wear, vertical wear, angle wear, and total wear, providing a data foundation for detecting rail surface damage. Linear structured light vision measurement based on triangulation can acquire three-dimensional cross-sectional data of rail profiles at high speed and in real time. It is a typical non-contact optical measurement method and has become the mainstream dynamic measurement method for rail profile and top surface abrasions, widely used in the railway inspection industry. However, when applied on railway sites, factors such as uneven rail surfaces, surface dirt, rust, and light bands at the wheel-rail interface can affect the light scattering direction and energy intensity on the rail surface, resulting in abnormal energy distribution and seriously affecting imaging quality. For example, due to long-term contact and friction with the wheels, the surface of the rail is very smooth, and the mirror reflection effect of light is significant. Specifically, most of the energy of the incident light is distributed near the mirror reflection direction, and only a small amount of diffuse reflected light is collected by the camera, resulting in very weak image information obtained by the camera, resulting in underexposure (i.e., insufficient exposure). This causes the image contrast and the confidence level of the light strip center to be relatively low, affecting the accurate acquisition of rail profile data, and further causing some areas of the rail to have voids (data loss) after three-dimensional reconstruction. This situation makes it difficult to ensure the accuracy and stability of rail profile measurement, and affects the subsequent profile analysis and rail top surface defect detection results. Adjusting the exposure time is a common solution. Although it solves the problem of underexposure in the light strip area, it will cause overexposure in other areas, resulting in excessively thick light strips, which will cause new problems.

[0004] Based on the information correlation and complementarity between polarization images, some researchers have proposed various polarization image fusion methods, mainly including frequency domain fusion and spatial domain fusion. In the frequency domain, some researchers use discrete wavelet transform to decompose images into low-frequency and high-frequency components of different scales, using the wavelet coefficients in the low-frequency and high-frequency images as features to determine the wavelet coefficients of the fused image. Some researchers have proposed a polarization image fusion algorithm based on two-dimensional discrete wavelet transform to enhance image detail and improve visual quality. In the spatial domain, some researchers have proposed a feature analysis-based polarization image fusion method. This method determines fusion weights based on the grayscale, texture, and shape features of the images and fuses the images to address the problem of detail loss when calculating polarization parametric images. Recently, deep learning-based image fusion methods, such as DeepFuse and FusionGan, have become a hot topic. However, these networks are mostly targeted at natural scenes with rich color and texture features. However, the amount of laser stripe data is small and lacks rich texture and color features. Therefore, these methods are not suitable for laser polarization stripe image fusion.

[0005] How to determine the fusion weight is the key to improving the fusion quality of multi-polarized light images. Some scholars have introduced a light stripe reliability evaluation mechanism and used it to determine the fusion weight of the source image. The light stripe reliability evaluation uses characteristic quantities such as light stripe width, grayscale, and the sum of squared residuals of the light stripes as evaluation indicators. For each source image, the light stripe reliability of each column is calculated. The total pixel intensity or light stripe width of the light stripe is selected separately as an evaluation indicator to calculate the reliability of the light stripe polarization imaging, and the best fusion effect is obtained by continuously adjusting the weights of each component image; the two evaluation indicators can also be used in combination to calculate the light stripe reliability and obtain the image fusion weight. This method effectively overcomes the problem of rail surface reflection, but when applied to the dynamic measurement process of the profile system, there are still some problems to be solved:

[0006] (1) Fusion strategies and weight calculations lack systematic quantitative analysis and rely too much on qualitative analysis results and manual experience thresholds;

[0007] (2) The light strip changes with the train operation. Rail wear, sunlight, and foreign object interference can cause unpredictable changes in the light strip. Therefore, choosing light strip width and brightness as the fusion weight judgment criteria is difficult to adapt to the complex and changing line conditions of the entire railway line.

[0008] (3) The computational time overhead of fusing multiple polarization images is large, which affects the real-time performance of the measurement system. Summary of the Invention

[0009] An embodiment of the present invention provides a method for fusing multi-polarized light point cloud data of rail appearance to address the problem of local underexposure of laser cross-sectional images encountered in rail profile measurement using non-polarized structured light projection. The method includes:

[0010] Merge the polarization camera data read into rail profile data;

[0011] The rail profile data is divided into multiple areas according to the rail vertex coordinates and the gauge point coordinates;

[0012] The rail profile data of multiple regions are fitted using the random sampling consistency RANSAC algorithm to obtain fitting curves of multiple regions;

[0013] Merge the fitting curves of multiple regions to obtain the best rail profile fitting curve;

[0014] The rail profile is measured using the optimal rail profile fitting curve.

[0015] An embodiment of the present invention further provides a multi-polarized light point cloud data fusion device for rail appearance status, which is used to solve the problem of local underexposure of laser cross-section images encountered in rail profile measurement using non-polarized structured light projection. The device includes:

[0016] A data merging module is used to merge the polarization camera data read into rail profile data;

[0017] The area division module is used to divide the rail profile data into multiple areas according to the rail vertex coordinates and the gauge point coordinates;

[0018] The fitting module is used to fit the rail profile data of multiple regions using the random sampling consistency RANSAC algorithm to obtain fitting curves of multiple regions;

[0019] Fusion module, used to fuse the fitting curves of multiple regions to obtain the optimal rail profile fitting curve;

[0020] The measurement module is used to measure the rail profile using the optimal rail profile fitting curve.

[0021] An embodiment of the present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the multi-polarized light point cloud data fusion method of the rail appearance state is implemented.

[0022] An embodiment of the present invention further provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the multi-polarized light point cloud data fusion method of the rail appearance state is implemented.

[0023] An embodiment of the present invention further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the multi-polarization light point cloud data fusion method of the rail appearance state.

[0024] In an embodiment of the present invention, the problem of local underexposure of laser cross-section images faced by rail profile measurement using non-polarized structured light projection is addressed. The polarization camera data read is merged into rail profile data; the rail profile data is divided into multiple regions according to the rail vertex coordinates and the gauge point coordinates; the rail profile data of the multiple regions are fitted using the random sampling consistency RANSAC algorithm to obtain fitting curves for the multiple regions; the fitting curves of the multiple regions are fused to obtain an optimal rail profile fitting curve; and the rail profile is measured using the optimal rail profile fitting curve. The fused image effectively solves the problem of local underexposure of the rail laser cross-section, improves the light stripe contrast and light stripe center confidence, as well as the image quality, and no longer has holes in the rail after three-dimensional reconstruction. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0026] Figure 1 The multi-polarized light point cloud data fusion method for the rail appearance state in the embodiment of the present invention is as follows: Figure 1 ;

[0027] Figure 2 The multi-polarized light point cloud data fusion method for the rail appearance state in the embodiment of the present invention is as follows: Figure 2 ;

[0028] Figure 3 The multi-polarized light point cloud data fusion method for the rail appearance state in the embodiment of the present invention is as follows: Figure 2 ;

[0029] Figure 4 Schematic diagram of the rail profile data acquisition principle based on multi-polarized light fusion in an embodiment of the present invention;

[0030] Figure 5 The four-directional polarization component image, the linear polarization angle image, the linear polarization degree image and the Stokes parameter image in the embodiment of the present invention;

[0031] Figure 6 Schematic diagram of fused point cloud data and segmented fitting results of segmented RANSAC results;

[0032] Figure 7 The 3D reconstruction results of the measured rail light strip images before and after polarization fusion and the corresponding local magnification images are given;

[0033] Figure 8 The structure frame of the multi-polarized light point cloud data fusion device for the rail appearance state in the embodiment of the present invention is shown in FIG. Figure 1 ;

[0034] Figure 9 The structure frame of the multi-polarized light point cloud data fusion device for the rail appearance state in the embodiment of the present invention is shown in FIG. Figure 2 . DETAILED DESCRIPTION

[0035] To make the purpose, technical solutions and advantages of the embodiments of the present invention more clear, the embodiments of the present invention are further described in detail below with reference to the accompanying drawings. Here, the exemplary embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0036] Figure 1 The multi-polarized light point cloud data fusion method for the rail appearance state in the embodiment of the present invention is as follows: Figure 1 ,like Figure 1 As shown, the method includes:

[0037] Step 101: Merging the read polarization camera data into rail profile data;

[0038] Step 102: Divide the rail profile data into multiple regions according to the rail vertex coordinates and the gauge point coordinates;

[0039] Step 103: fitting the rail profile data of multiple regions using the random sampling consistency RANSAC algorithm to obtain fitting curves of the multiple regions;

[0040] Step 104: Fusing the fitting curves of multiple regions to obtain the optimal rail profile fitting curve;

[0041] Step 105: Measure the rail profile using the optimal rail profile fitting curve.

[0042] In an embodiment of the present invention, the polarization camera data includes a four-directional polarization component image, a linear polarization angle image, a linear polarization degree image, and a Stokes parameter image;

[0043] like Figure 2 As shown, step 101 combines the read polarization camera data into rail profile data, including:

[0044] Step 201: Calculate the light stripe centers corresponding to the four-directional polarization component image, the linear polarization angle image, the linear polarization degree image, and the Stokes parameter image respectively;

[0045] Step 202: Calculate and obtain corresponding profile polarization point cloud data based on the corresponding light strip center;

[0046] Step 203: merging the corresponding profile polarization point cloud data into a piece of rail profile data according to the point cloud coordinate arrangement.

[0047] Specifically, for the 9 polarization component images, the four-directional polarization component images I0, I 45 , I 90 , I 135 , linear polarization angle image I AoP , linear polarization image I DoP , the Stokes parameter images S0, S1, S2 are used to find the center of the light strip, and the corresponding profile polarization point cloud data are obtained, which are recorded as P0, P 45 , P 90 , P 135 , P AoP , P DoP , S′0, S′1, S′2. Calculating the corresponding profile polarization point cloud data according to the corresponding light strip center is an existing technology and can be calculated using methods in the existing technology.

[0048] The 9 profile polarization point cloud data are arranged and merged into one profile data P according to the point cloud coordinates, which is recorded as

[0049] P=∪(P0,P 45 ,P 90 ,P 135 ,P AoP ,P DoP ,S′0,S′1,S′2);

[0050] Where P represents the rail profile data; ∪(·) represents the point cloud coordinate data merging operation; P0 represents the profile polarization point cloud data corresponding to the polarization component image I0 with a polarization direction of 0°; P 45 The polarization component image I with a polarization direction of 45° 45 Corresponding profile polarization point cloud data; P 90 Represents the polarization component image I with a polarization direction of 90° 90 Corresponding profile polarization point cloud data; P 135 The polarization component image I with a polarization direction of 135° 135 Corresponding profile polarization point cloud data; P AoP Represents the linear polarization angle image I AoP Corresponding profile polarization point cloud data; P DoP Represents the linear polarization image IDoP Corresponding profile polarization point cloud data; S0′ represents the profile polarization point cloud data corresponding to the first component image S0 of the Stokes parameter image; S1′ represents the profile polarization point cloud data corresponding to the second component image S1 of the Stokes parameter image; S2′ represents the profile polarization point cloud data corresponding to the third component image S2 of the Stokes parameter image.

[0051] In the embodiment of the present invention, the rail profile is divided into five regions according to the rail top gauge point. The rail profile data is divided into multiple regions according to the rail top coordinates and the gauge point coordinates according to the following formula:

[0052]

[0053] Among them, P t , They represent the rail profile data of the five divided regions respectively; segment(·) represents the region segmentation operation; P represents the rail profile data; T x,y Indicates the coordinates of the track vertex; G x,y Indicates the coordinates of the track gauge point.

[0054] In the embodiment of the present invention, the Random Sampling Consensus Algorithm (RANSAC) is executed on each region to obtain the best fitting polynomial equation curve for each region. The rail profile data of multiple regions are fitted using the Random Sampling Consensus Algorithm according to the following formula to obtain the fitting curves of multiple regions:

[0055]

[0056] Among them, C t , Represents the RANSAC fitting polynomial curves of the five divided regions; ξ t , Respectively represent the internal point thresholds of the five divided regions. Points smaller than this value are sampled for fitting, and points larger than this value are filtered out as noise; m t , Respectively represents the number of sampling iterations of the five divided regions; n t , They represent the best polynomial fitting powers of the five divided areas respectively. The value can be obtained by selecting multiple typical profile data from the actual road, constructing a global optimization model, and using the least squares method.

[0057] In the embodiment of the present invention, the fitting curve C of the five segments is t , Combine the fitting curves of multiple regions into a complete half-section rail profile. Use the following formula to merge the fitting curves of multiple regions to obtain the best rail profile fitting curve:

[0058]

[0059] Where C represents the optimal rail profile fitting curve after fusion of multiple polarization point cloud data; C t , They represent the RANSAC fitting polynomial curves of the five divided regions respectively; stitching(·) represents the curve fusion operation.

[0060] In the embodiment of the present invention, Figure 3 As shown, the method further includes:

[0061] Step 301: Determine whether the fitting curves of the multiple regions meet the requirements. If not, adjust the RANSAC algorithm parameters and re-fit. The parameters may include threshold, number of iterations, and optimal polynomial fitting power.

[0062] This paper addresses the problem of localized underexposure in laser cross-section images during rail profile measurement using unpolarized structured light projection. A method for fusing multi-polarized light point cloud data of rail appearance is proposed. Based on a segmented RANSAC algorithm, a point cloud data fusion algorithm for four polarization component images, Stokes parameter images, linear polarization angle images, and linear polarization degree images is constructed. The resulting fused image effectively resolves the issue of localized underexposure in rail laser cross-sections, improves light stripe contrast and light stripe center confidence, and enhances image quality. The resulting 3D reconstructed rails no longer exhibit voids. Specific embodiments

[0064] We selected the Genie Nano M2450Polarized polarization camera, which is manufactured by Teledyne Dalsa of Canada and has good polarization imaging performance. We integrated the camera into the rail profile detection system and Figures 1-4 The process shown in the figure obtains the polarization fusion point cloud data of the track profile, and the following can be obtained: Figure 5 The rail laser cross-section four-directional polarization component images and Stokes parameter images S0, S1, S2, linear polarization angle image I AoP and linear polarization image I DoLP .in, Figure 5 In the figure, (a) represents the polarization component image in the 0° direction; (b) represents the polarization component image in the 135° direction; (c) represents the polarization component image in the 45° direction; (d) represents the polarization component image in the 90° direction; (e) represents the Stokes parameter image S0; (f) represents the Stokes parameter image S1; (g) represents the Stokes parameter image S2; (h) represents the linear polarization angle image I AoP ; (i) represents the linear polarization image I DoP .

[0065] The polarization components are fused using the segmented random sampling consistency polarization point cloud data fusion algorithm. The results are as follows: Figure 6 As shown in Figure 2, (a) represents the fused point cloud data; (b) represents the segmented fitting result.

[0066] Rail profiles were measured near Qinghecheng Station on the K346 line of the Beijing-Kowloon Line's downlink section after the first and second passes of the rail grinding process. The profiles obtained by the detection system were compared with those obtained using the Miniprof tool to confirm system accuracy. The system-derived profiles employed the fusion algorithm proposed in this project. Because the number of points collected by the system differs from that of the Miniprof tool, the distances between point pairs calculated directly cannot be used to directly assess profile differences. Therefore, smoothing was performed before discretization of the profiles to ensure that only corresponding point pairs were compared, allowing for accurate assessment of profile differences. The comparison results are shown in Table 1.

[0067] Table 1 Comparison of profile measurement accuracy before and after fusion

[0068] Silhouette Maximum Minimum average value 95th percentile Before fusion 0.2587 0.0019 0.0542 0.1230 After fusion 0.1202 0.0011 0.0363 0.0800

[0069] As shown in Table 1, the 95th percentile difference between the profile detected by the fused system and the profile measured by a contact detection device (miniprof) with an accuracy of 0.02 mm is controlled within 0.1 mm. In other words, the polarization fusion method proposed in the present invention effectively improves the accuracy of the profile measurement system, and the system measurement accuracy is better than 0.1 mm.

[0070] To further verify the effectiveness of multi-polarized light fusion for 3D imaging of the rail top surface, a section of rail prone to exposure anomalies was placed on an electrically controlled translation stage. A linear structured light profile scanning device was constructed. Laser cross-sectional images of the rail were acquired using the aforementioned polarization camera at a 2mm sampling interval. Three-dimensional reconstruction of the rail was performed through a process involving light strip center extraction and calibration. Figure 7 The 3D reconstruction results of the measured rail light stripe images before and after polarization fusion are shown, with (a) showing the image before polarization fusion; (b) showing a partial magnification of (a); (c) showing the image after polarization fusion; and (d) showing a partial magnification of (c). It can be seen that before polarization fusion, due to abnormal exposure of the total intensity image, some areas of the 3D reconstructed rail exhibit voids (data loss). However, after polarization fusion, the image quality is significantly improved. The rail reconstructed from the fused image exhibits no voids, and the reconstruction results still accurately reflect the rail's true condition.

[0071] The present invention also provides a device for fusion of multi-polarized light point cloud data on rail appearance, as described in the following embodiments. Because the principles underlying the device are similar to those of the multi-polarized light point cloud data fusion method for rail appearance, the implementation of the device can be referenced to the implementation of the multi-polarized light point cloud data fusion method for rail appearance, and any repetitions will not be repeated.

[0072] Figure 8 The structure frame of the multi-polarized light point cloud data fusion device for the rail appearance state in the embodiment of the present invention is shown in FIG. Figure 1 ,like Figure 8 As shown, the multi-polarized light point cloud data fusion device for the rail appearance state includes:

[0073] Data merging module 02, used to merge the read polarization camera data into rail profile data;

[0074] The region division module 04 is used to divide the rail profile data into multiple regions according to the rail vertex coordinates and the gauge point coordinates;

[0075] Fitting module 06, used to fit the rail profile data of multiple regions using the random sampling consistency RANSAC algorithm to obtain fitting curves of the multiple regions;

[0076] Fusion module 08 is used to fuse the fitting curves of multiple regions to obtain the best rail profile fitting curve;

[0077] The measurement module 10 is used to measure the rail profile using the optimal rail profile fitting curve.

[0078] In an embodiment of the present invention, the polarization camera data includes a four-directional polarization component image, a linear polarization angle image, a linear polarization degree image, and a Stokes parameter image;

[0079] The data merging module is specifically used for:

[0080] Calculate the light stripe centers corresponding to the four-directional polarization component images, linear polarization angle images, linear polarization degree images and Stokes parameter images respectively;

[0081] The corresponding profile polarization point cloud data is calculated based on the corresponding light strip center;

[0082] The corresponding profile polarization point cloud data are arranged and merged into a rail profile data according to the point cloud coordinates.

[0083] In an embodiment of the present invention, the polarization camera data read is combined into rail profile data according to the following formula:

[0084] P=∪(P0,P 45 ,P 90 ,P 135,P AoP ,P DoP ,S′0,S′1,S′2);

[0085] Where P represents the rail profile data; ∪(·) represents the point cloud coordinate data merging operation; P0 represents the profile polarization point cloud data corresponding to the polarization component image I0 with a polarization direction of 0°; P 45 The polarization component image I with a polarization direction of 45° 45 Corresponding profile polarization point cloud data; P 90 Represents the polarization component image I with a polarization direction of 90° 90 Corresponding profile polarization point cloud data; P 135 The polarization component image I with a polarization direction of 135° 135 Corresponding profile polarization point cloud data; P AoP Represents the linear polarization angle image I AoP Corresponding profile polarization point cloud data; P DoP Represents the linear polarization image I DoP Corresponding profile polarization point cloud data; S0′ represents the profile polarization point cloud data corresponding to the first component image S0 of the Stokes parameter image; S1′ represents the profile polarization point cloud data corresponding to the second component image S1 of the Stokes parameter image; S2′ represents the profile polarization point cloud data corresponding to the third component image S2 of the Stokes parameter image.

[0086] In an embodiment of the present invention, the rail profile data is divided into multiple regions according to the rail vertex coordinates and the gauge point coordinates according to the following formula:

[0087]

[0088] Among them, P t , They represent the rail profile data of the five divided regions respectively; segment(·) represents the region segmentation operation; P represents the rail profile data; T x,y Indicates the coordinates of the track vertex; G x,y Indicates the coordinates of the track gauge point.

[0089] In an embodiment of the present invention, the rail profile data of multiple regions are fitted using the random sampling consistency RANSAC algorithm according to the following formula to obtain fitting curves of the multiple regions:

[0090]

[0091] Among them, C t , Represents the RANSAC fitting polynomial curves of the five divided regions; ξ t , Respectively represent the internal point thresholds of the five divided regions; m t , Respectively represents the number of sampling iterations of the five divided regions; n t , Respectively represent the best polynomial fitting powers of the five divided regions.

[0092] In the embodiment of the present invention, the fitting curves of multiple regions are fused according to the following formula to obtain the optimal rail profile fitting curve:

[0093]

[0094] Where C represents the optimal rail profile fitting curve; C t , They represent the RANSAC fitting polynomial curves of the five divided regions respectively; stitching(·) represents the curve fusion operation.

[0095] In the embodiment of the present invention, Figure 9 As shown, it also includes:

[0096] The judgment module 12 is used to judge whether the fitting curves of multiple regions meet the requirements. If not, the RANSAC algorithm parameters are adjusted and the fitting is performed again.

[0097] An embodiment of the present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the multi-polarized light point cloud data fusion method of the rail appearance state is implemented.

[0098] An embodiment of the present invention further provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the multi-polarized light point cloud data fusion method of the rail appearance state is implemented.

[0099] An embodiment of the present invention further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the multi-polarization light point cloud data fusion method of the rail appearance state.

[0100] In an embodiment of the present invention, the problem of local underexposure of laser cross-section images faced by rail profile measurement using non-polarized structured light projection is addressed. The polarization camera data read is merged into rail profile data; the rail profile data is divided into multiple regions according to the rail vertex coordinates and the gauge point coordinates; the rail profile data of the multiple regions are fitted using the random sampling consistency RANSAC algorithm to obtain fitting curves for the multiple regions; the fitting curves of the multiple regions are fused to obtain an optimal rail profile fitting curve; and the rail profile is measured using the optimal rail profile fitting curve. The fused image effectively solves the problem of local underexposure of the rail laser cross-section, improves the light stripe contrast and light stripe center confidence, as well as the image quality, and no longer has holes in the rail after three-dimensional reconstruction.

[0101] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0102] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0103] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0104] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0105] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for fusing multi-polarized light point cloud data of rail appearance, characterized in that: include: Merge the polarization camera data read into rail profile data; The rail profile data is divided into multiple areas according to the rail vertex coordinates and the gauge point coordinates; The rail profile data of multiple regions are fitted using the random sampling consistency RANSAC algorithm to obtain fitting curves of multiple regions; Merge the fitting curves of multiple regions to obtain the best rail profile fitting curve; Use the best rail profile fitting curve to measure rail profile; The rail profile data of multiple regions are fitted using the random sampling consistency RANSAC algorithm according to the following formula to obtain the fitting curves of multiple regions: Among them, C t , C w1 , Represents the RANSAC fitting polynomial curves of the five divided regions; ξ t , Respectively represent the internal point thresholds of the five divided regions; m t , Respectively represents the number of sampling iterations of the five divided regions; n t , They represent the best polynomial fitting power of the five divided areas, which is obtained by selecting multiple typical profile data from the actual road, building a global optimization model, and using the least squares method; P t , Respectively represent the rail profile data of the five divided areas; Determine whether the fitting curves of multiple regions meet the requirements. If not, adjust the RANSAC algorithm parameters and re-fit.

2. The multi-polarized light point cloud data fusion method for rail appearance status according to claim 1, characterized in that: The polarization camera data includes four-directional polarization component images, linear polarization angle images, linear polarization degree images and Stokes parameter images; Merge the polarization camera data into rail profile data, including: Calculate the light stripe centers corresponding to the four-directional polarization component images, linear polarization angle images, linear polarization degree images and Stokes parameter images respectively; The corresponding profile polarization point cloud data is calculated based on the corresponding light strip center; The corresponding profile polarization point cloud data are arranged and merged into a rail profile data according to the point cloud coordinates.

3. The multi-polarized light point cloud data fusion method for rail appearance status according to claim 1, characterized in that: The polarization camera data is combined into rail profile data according to the following formula: P=∪(P0,P 45 ,P 90 ,P 135 ,P AoP ,P DoP ,S′0,S′1,S′2); Where P represents the rail profile data; ∪(·) represents the point cloud coordinate data merging operation; P0 represents the profile polarization point cloud data corresponding to the polarization component image I0 with a polarization direction of 0°; P 45 The polarization component image I with a polarization direction of 45° 45 Corresponding profile polarization point cloud data; P 90 Represents the polarization component image I with a polarization direction of 90° 90 Corresponding profile polarization point cloud data; P 135 The polarization component image I with a polarization direction of 135° 135 Corresponding profile polarization point cloud data; P AoP Represents the linear polarization angle image I AoP Corresponding profile polarization point cloud data; P DoP Represents the linear polarization image I DoP Corresponding profile polarization point cloud data; S0′ represents the profile polarization point cloud data corresponding to the first component image S0 of the Stokes parameter image; S1′ represents the profile polarization point cloud data corresponding to the second component image S1 of the Stokes parameter image; S2′ represents the profile polarization point cloud data corresponding to the third component image S2 of the Stokes parameter image.

4. The multi-polarized light point cloud data fusion method for rail appearance status according to claim 1, characterized in that: The rail profile data is divided into multiple areas according to the rail vertex coordinates and gauge point coordinates according to the following formula: Among them, P t , They represent the rail profile data of the five divided regions respectively; segment(·) represents the region segmentation operation; P represents the rail profile data; T x,y Indicates the coordinates of the track vertex; G x,y Indicates the coordinates of the track gauge point.

5. The multi-polarized light point cloud data fusion method for rail appearance status according to claim 1, characterized in that: The fitting curves of multiple regions are fused according to the following formula to obtain the optimal rail profile fitting curve: Where C represents the optimal rail profile fitting curve; C t , C w1 , They represent the RANSAC fitting polynomial curves of the five divided regions respectively; stitching(·) represents the curve fusion operation.

6. A multi-polarized light point cloud data fusion device for rail appearance status, characterized in that: include: A data merging module is used to merge the polarization camera data read into rail profile data; The area division module is used to divide the rail profile data into multiple areas according to the rail vertex coordinates and the gauge point coordinates; The fitting module is used to fit the rail profile data of multiple regions using the random sampling consistency RANSAC algorithm to obtain fitting curves of multiple regions; Fusion module, used to fuse the fitting curves of multiple regions to obtain the optimal rail profile fitting curve; A measurement module is used to measure the rail profile using the optimal rail profile fitting curve; The rail profile data of multiple regions are fitted using the random sampling consistency RANSAC algorithm according to the following formula to obtain the fitting curves of multiple regions: Among them, C t , C w1 , Represents the RANSAC fitting polynomial curves of the five divided regions; ξ t , Respectively represent the internal point thresholds of the five divided regions; m t , Respectively represents the number of sampling iterations of the five divided regions; n t , They represent the best polynomial fitting power of the five divided areas, which is obtained by selecting multiple typical profile data from the actual road, building a global optimization model, and using the least squares method; P t , Respectively represent the rail profile data of the five divided areas; The judgment module is used to judge whether the fitting curves of multiple regions meet the requirements. If not, the RANSAC algorithm parameters are adjusted and the fitting is performed again.

7. The multi-polarized light point cloud data fusion device for rail appearance status according to claim 6, characterized in that: The polarization camera data includes four-directional polarization component images, linear polarization angle images, linear polarization degree images and Stokes parameter images; The data merging module is specifically used for: Calculate the light stripe centers corresponding to the four-directional polarization component images, linear polarization angle images, linear polarization degree images and Stokes parameter images respectively; The corresponding profile polarization point cloud data is calculated based on the corresponding light strip center; The corresponding profile polarization point cloud data are arranged and merged into a rail profile data according to the point cloud coordinates.

8. The multi-polarized light point cloud data fusion device for rail appearance status according to claim 6, characterized in that: The polarization camera data is combined into rail profile data according to the following formula: P=∪(P0,P 45 ,P 90 ,P 135 ,P AoP ,P DoP ,S′0,S′1,S′2); Where P represents the rail profile data; ∪(·) represents the point cloud coordinate data merging operation; P0 represents the profile polarization point cloud data corresponding to the polarization component image I0 with a polarization direction of 0°; P 45 The polarization component image I with a polarization direction of 45° 45 Corresponding profile polarization point cloud data; P 90 Represents the polarization component image I with a polarization direction of 90° 90 Corresponding profile polarization point cloud data; P 135 The polarization component image I with a polarization direction of 135° 135 Corresponding profile polarization point cloud data; P AoP Represents the linear polarization angle image I AoP Corresponding profile polarization point cloud data; P DoP Represents the linear polarization image I DoP Corresponding profile polarization point cloud data; S′0 represents the profile polarization point cloud data corresponding to the first component image S0 of the Stokes parameter image; S′1 represents the profile polarization point cloud data corresponding to the second component image S1 of the Stokes parameter image; S′2 represents the profile polarization point cloud data corresponding to the third component image S2 of the Stokes parameter image.

9. The multi-polarized light point cloud data fusion device for rail appearance status according to claim 6, characterized in that: The rail profile data is divided into multiple areas according to the rail vertex coordinates and gauge point coordinates according to the following formula: Among them, P t , They represent the rail profile data of the five divided regions respectively; segment(·) represents the region segmentation operation; P represents the rail profile data; T x,y Indicates the coordinates of the track vertex; G x,y Indicates the coordinates of the track gauge point.

10. The multi-polarized light point cloud data fusion device for rail appearance status according to claim 6, characterized in that: The fitting curves of multiple regions are fused according to the following formula to obtain the optimal rail profile fitting curve: Where C represents the optimal rail profile fitting curve; C t , C w1 , They represent the RANSAC fitting polynomial curves of the five divided regions respectively; stitching(·) represents the curve fusion operation.

11. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the multi-polarized light point cloud data fusion method of the rail appearance state described in any one of claims 1 to 5 is implemented.

12. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the multi-polarization light point cloud data fusion method of the rail appearance state according to any one of claims 1 to 5 is implemented.

13. A computer program product, characterized in that The computer program product includes a computer program, and when the computer program is executed by a processor, it implements the multi-polarization light point cloud data fusion method of the rail appearance state according to any one of claims 1 to 5.

Images