Station platform gauge intelligent detection method and system based on computer vision

Abstract

The invention relates to the technical field of computer vision, in particular to a station platform limit intelligent detection method and system based on computer vision, and the method comprises the following steps: obtaining platform region spectral data, extracting spectral features according to the reflectivity value of each pixel point in a spectral range, comparing the spectral features with the features of various known materials, and determining the position of the station platform limit. And identifying material types of the plurality of positions, detecting obstacles, and generating material classification feature values. According to the method, the spectral data is acquired, the reflectivity value is compared to identify the material type, the texture feature and the gradient change rate are utilized to analyze and identify the fuzzy area, misjudgment on the fuzzy identification is reduced, point cloud data and image information are adopted for time-space synchronous registration, and multi-source sensing information is fused to enhance the time-space consistency of three-dimensional reconstruction. The size accuracy of the platform model is improved, the space offset is calculated by comparing engineering design drawing data, a structure deviation quantitative evaluation system is established, a deformation area is positioned, and the foreseeability of safety early warning is improved.

Description

Technical Field

[0001] The present invention relates to the field of computer vision technology, and particularly to an intelligent detection method and system for station platform clearance based on computer vision. Background Art

[0002] The field of computer vision technology involves using computers and algorithms to analyze, understand, and process image and video data. The core contents include image acquisition, feature extraction, object detection, object recognition, image segmentation, 3D reconstruction, and image enhancement. Computer vision technology is widely applied in multiple fields such as intelligent monitoring, autonomous driving, medical image analysis, industrial inspection, and face recognition. Its development mainly relies on deep learning, traditional machine learning, and pattern recognition technologies. By constructing neural networks or rule-based methods, it realizes the automatic analysis and understanding of image and video data. Combining data preprocessing, feature representation, classifier training, and object tracking, different optimization algorithms and computing models are usually adopted for different application scenarios to improve recognition accuracy and computing efficiency.

[0003] Among them, the intelligent detection method for station platform clearance refers to using computer vision technology to automatically detect the clearance situation of the station platform. Through the acquisition of image or video data, combined with feature extraction, object detection, and 3D measurement, elements such as platform edges, obstacles, and clearance signs are identified and located. It includes means such as platform structure reconstruction based on multi-view images, object detection based on deep learning, and distance measurement based on point cloud data. It accurately identifies the platform area and detects its clearance status, and uses image matching and feature point tracking technology to realize the dynamic monitoring of platform facilities and the surrounding environment, and conducts data analysis on factors that may affect train operation safety.

[0004] Traditional station platform clearance detection technologies rely on single visual feature analysis, lack the ability to identify the material characteristics of the platform, are prone to missed detections and misjudgments when the reflective characteristics of the obstacle surface are similar to the background, do not consider the problem of feature degradation caused by natural wear or light interference, the lack of a gradient direction interpolation mechanism limits the positioning accuracy of fuzzy identification areas, the spatio-temporal registration accuracy of point cloud and image data is insufficient, and no time synchronization mechanism for sensor data is established, resulting in three-dimensional model jitter or dislocation during dynamic monitoring, affecting the continuous tracking of structural deformation. The structural offset assessment lacks a comparison benchmark for engineering design data, and it is difficult to distinguish construction errors from later deformations relying only on on-site measurement data, affecting the pertinence of maintenance decisions. The collision risk assessment does not integrate train operation parameters, and the static clearance detection is disconnected from the dynamic operation trajectory, unable to predict changes in safety margins under special working conditions such as curve driving or emergency braking, and there is a risk of lag in safety hazard prediction. Summary of the Invention

[0005] To solve the technical problems existing in the prior art, an embodiment of the present invention provides an intelligent detection method and system for the platform limit of a station based on computer vision. The technical solution is as follows:

[0006] To achieve the above object, the present invention adopts the following technical solution. An intelligent detection method for the platform limit of a station based on computer vision includes the following steps:

[0007] S1: Obtain the spectral data of the platform area. According to the reflectivity values of each pixel point within the spectral range, extract the spectral features and compare them with the features of multiple known materials to identify the material types at multiple positions, detect obstacles, and generate material classification feature values;

[0008] S2: According to the material classification feature values, use an image acquisition device to obtain an image of the platform area, extract the texture features of the limit marking area, identify the blurred area by calculating the direction gradient change rate, and calculate the gradient direction interpolation of the blurred area to obtain the limit marking correction value;

[0009] S3: Based on the limit marking correction value, use a lidar to obtain the point cloud data of the platform area, and combine the image information to perform spatio-temporal synchronous registration on the platform point cloud data to construct a platform structure model;

[0010] S4: Based on the platform structure model, calculate the spatial offset and offset angle at multiple points by comparing with the platform structure data in the engineering design drawings to generate a structure offset value;

[0011] S5: According to the structure offset value, extract the offset point position information, call the outer contour size and running track of the train, calculate the collision risk between the platform limit and the train, and generate a collision risk value.

[0012] As a further solution of the present invention, the material classification feature values are specifically material spectral reflectivity parameters, material type matching results, and obstacle spectral feature values. The limit marking correction value includes the limit marking texture direction gradient change rate, the blurred area gradient interpolation calculation value, and the marking boundary correction parameter. The platform structure model is specifically the platform edge point cloud coordinates, the limit marking point cloud data, and the platform facility point cloud data. The structure offset value includes the change amount of the spatial coordinates of the offset point, the change amount of the rotation angle of the offset point, and the platform structure offset trend data. The collision risk value includes the platform offset distance, the platform offset angle, and the collision risk level.

[0013] As a further solution of the present invention, the step of obtaining the spectral data of the platform area, extracting the spectral features and comparing them with the features of multiple known materials according to the reflectivity values of each pixel point within the spectral range to identify the material types at multiple positions, detect obstacles, and generate material classification feature values is specifically as follows:

[0014] S101: Obtain the spectral data of the platform area, extract the reflectivity values of each pixel point, calculate the spectral peak positions, bandwidth half-height widths, and spectral slope parameters of multiple bands, extract the spectral features at multiple positions, and generate a spectral feature parameter set;

[0015] S102: Based on the spectral feature parameter set, call the spectral database of known materials, calculate the spectral similarity of materials at multiple positions on the platform, calibrate the material types at multiple positions on the platform surface, and obtain the material category distribution data of the platform;

[0016] S103: Call the material category distribution data of the platform, detect the spectral reflection abnormal points in the platform area, identify obstacles, and record the position information in real time to generate a material classification feature value.

[0017] As a further solution of the present invention, the specific formula for calculating the spectral similarity of materials at multiple positions on the platform is:

[0018]

[0019] Calculate the spectral matching degree index;

[0020] where S′ represents the spectral matching degree index between the platform measurement spectrum and the known material spectrum, A k represents the reflectivity value of the platform measurement spectrum at the k-th wavelength point, B k represents the reflectivity value of the known material spectrum at the k-th wavelength point, K represents the total number of wavelength points used to calculate the spectral matching degree, and k represents the wavelength point index in the spectral measurement.

[0021] As a further solution of the present invention, according to the material classification feature value, the steps of obtaining the platform area image by using an image acquisition device, extracting the texture features of the limit marking area, identifying the blurred area by calculating the direction gradient change rate, and calculating the gradient direction interpolation of the blurred area to obtain the limit marking correction value are specifically as follows:

[0022] S201: Obtain the material classification feature value, call the image acquisition device, collect the image data of the platform area, identify the limit marking area, calculate the pixel gray gradient of the target area, and extract the local texture features according to the gradient direction distribution to obtain the limit marking texture feature data;

[0023] S202: Based on the limit marking texture feature data, identify the blurred area by calculating the direction gradient change rate of adjacent pixels, and mark the pixel range of the blurred area to obtain the limit marking blurred area data;

[0024] S203: Based on the limit marking blurred area data, calculate the interpolation parameter of the pixel gradient direction in the blurred area to obtain the limit marking correction value.

[0025] As a further solution of the present invention, based on the limit identification correction value, using a lidar to obtain the point cloud data of the platform area, and combining with the image information, the steps of performing spatio-temporal synchronization registration on the platform point cloud data and constructing a platform structure model are specifically as follows:

[0026] S301: Obtain the limit identification correction value, call the lidar to collect the point cloud data of the platform area, analyze the spatial coordinate information of the point cloud data, and obtain the preliminary platform point cloud data;

[0027] S302: Based on the preliminary platform point cloud data, call the image data of the platform area, calculate the spatial coordinate matching error between the point cloud data and the image data, and adjust the spatial coordinates of the point cloud data according to the alignment of the platform edge points and the limit identification points to obtain the platform matching point cloud data;

[0028] S303: Based on the platform matching point cloud data, extract the structural form information of multiple areas of the platform, construct the spatial model of the platform, and obtain the platform structure model.

[0029] As a further solution of the present invention, based on the platform structure model, by comparing with the platform structure data in the engineering design drawings, the steps of calculating the spatial offset amount and offset angle of multiple points and generating a structure offset value are specifically as follows:

[0030] S401: Obtain the platform structure model, call the engineering design drawings, extract the spatial coordinates of the platform edge, platform facilities, and limit identification in the drawings, and obtain the platform reference coordinates;

[0031] S402: Based on the platform reference coordinates, detect multiple spatial offset points by calculating the spatial position difference between the platform point cloud data and the reference coordinates, and obtain the platform offset point data;

[0032] S403: Based on the platform offset point data, calculate the spatial offset distance of multiple offset points, analyze the offset direction angles of multiple points, and obtain the platform structure offset value.

[0033] As a further solution of the present invention, according to the structure offset value, extract the offset point information, call the outer contour size and running track of the train, and calculate the collision risk between the platform limit and the train. The steps of generating a collision risk value are specifically as follows:

[0034] S501: Obtain the platform structure offset value, extract the spatial coordinates of multiple offset points, calibrate the spatial distribution of the offset points, and obtain the platform limit offset point data;

[0035] S502: Based on the data of the platform limit offset points, call the train outline dimensions and operation trajectory information, calculate the spatial occupancy range of the train when passing through the platform limit area, and obtain the train passage space data;

[0036] S503: Based on the train passage space data, evaluate the collision risk by calculating the distance between the offset points and the train outline, and obtain the collision risk value.

[0037] As a further solution of the present invention, the specific formula for evaluating the collision risk is:

[0038]

[0039] Calculate the collision risk coefficient;

[0040] Where, R c is the collision risk coefficient, d s is the actual measured distance between the s-th offset point and the train outline, d safe is the safety distance in the platform limit design standard, v s is the speed of the train passing through the offset point area, t s is the time required for the train to pass through the offset area, s is the index of the s-th offset point, S is the total number of offset points participating in the calculation of the collision risk, d s is the actual measured distance between the s-th offset point and the train outline, d safe is the platform safety distance standard determined according to the engineering design specifications.

[0041] On the other hand, a computer vision-based intelligent detection system for station platform limits is provided. This system is applied to the computer vision-based intelligent detection method for station platform limits. The system includes:

[0042] The spectral analysis module, based on the spectral data of the platform area, calculates the spectral peak position, bandwidth at half maximum, and spectral gradient change by analyzing the spectral reflectance values of each pixel point, calls the known material spectral database, classifies the platform materials, detects obstacles, and obtains the platform material classification characteristic values;

[0043] The identification optimization module, based on the platform material classification characteristic values, collects the image data of the platform area, extracts the pixel features of the limit identification area, identifies the blurred identification area, and calculates the gradient direction interpolation of the blurred area to obtain the limit identification correction value;

[0044] The structure construction module, based on the limit identification correction value, calls the lidar to collect the point cloud data of the platform area, extracts the platform edge points, facility points, and limit identification points, calculates the matching deviation between the image information and the point cloud data, performs spatio-temporal synchronous registration, adjusts the spatial coordinates of the point cloud data, and constructs the platform structure model;

[0045] The offset calculation module calls the engineering design drawing data based on the platform structure model, compares the offset of the point cloud data with the design standard, calculates the spatial offset of multiple points of the platform, and obtains the platform structure offset value;

[0046] The risk assessment module extracts the offset point within the platform limit based on the platform structure offset value, calls the train outline dimension data and running trajectory, calculates the collision risk, and obtains the collision risk value.

[0047] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0048] By acquiring spectral data and comparing reflectance values ​​to identify material types, using texture features and gradient change rate analysis to identify fuzzy areas, reducing misjudgment of fuzzy signs, using point cloud data and image information for spatiotemporal synchronous registration, integrating multi-source perception information to enhance the spatiotemporal consistency of three-dimensional reconstruction, and improving the dimensional accuracy of the platform model, the spatial offset is calculated by comparing engineering design drawing data, a quantitative evaluation system for structural deviations is established, the deformation area is located, and the predictability of safety warnings is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0050] Figure 1 It is a schematic diagram of the workflow of the present invention;

[0051] Figure 2 It is a system flow chart of the present invention. DETAILED DESCRIPTION

[0052] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0053] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either of the two.

[0054] In the embodiments of the present invention, "image" and "picture" can sometimes be used interchangeably. It should be noted that when the difference is not emphasized, the meanings they convey are the same. "Of", "corresponding", and "corresponding" can sometimes be used interchangeably. It should be noted that when the difference is not emphasized, the meanings they convey are the same.

[0055] In the embodiments of the present invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meanings they convey are the same.

[0056] To make the technical problems, technical solutions, and advantages to be solved by the present invention clearer, the following will be described in detail with reference to the accompanying drawings and specific embodiments.

[0057] Please refer to Figure 1 , the present invention provides a technical solution, an intelligent detection method for the platform boundary of a station based on computer vision, including the following steps:

[0058] S1: Obtain the spectral data of the platform area, extract the spectral features according to the reflectivity values of each pixel point within the spectral range, compare with the features of multiple known materials, identify the material types at multiple positions, detect obstacles, and generate material classification feature values;

[0059] S2: According to the material classification feature values, use an image acquisition device to obtain the platform area image, extract the texture features of the boundary marking area, identify the blurred area by calculating the direction gradient change rate, and calculate the gradient direction interpolation of the blurred area to obtain the boundary marking correction value;

[0060] S3: Based on the boundary marking correction value, use a lidar to obtain the point cloud data of the platform area, combine with the image information, perform spatio-temporal synchronous registration on the platform point cloud data, and construct a platform structure model;

[0061] S4: Based on the platform structure model, by comparing with the platform structure data in the engineering design drawings, calculate the spatial offset and offset angle at multiple points, and generate a structure offset value;

[0062] S5: According to the structure offset value, extract the offset point information, call the outer contour size and running track of the train, calculate the collision risk between the platform boundary and the train, and generate a collision risk value.

[0063] The material classification feature values are specifically the material spectral reflectance parameters, the material type matching results, and the obstacle spectral feature values. The boundary identification correction values include the boundary identification texture direction gradient change rate, the fuzzy area gradient interpolation calculation value, and the identification boundary correction parameter. The platform structure model is specifically the platform edge point cloud coordinates, the boundary identification point cloud data, and the platform facility point cloud data. The structure offset values include the spatial coordinate change amount of the offset point, the rotation angle change amount of the offset point, and the platform structure offset trend data. The collision risk values include the platform offset distance, the platform offset angle, and the collision risk level.

[0064] The steps of obtaining the spectral data of the platform area, extracting the spectral features according to the reflectance values of each pixel point within the spectral range, comparing with the features of multiple known materials, identifying the material types at multiple positions, and detecting obstacles to generate the material classification feature values are specifically as follows:

[0065] S101: Obtain the spectral data of the platform area, extract the reflectance value of each pixel point, calculate the spectral peak positions, bandwidth at half maximum, and spectral slope parameters of multiple bands, extract the spectral features at multiple positions, and generate a spectral feature parameter set;

[0066] Obtain the spectral data of the platform area, collect data using a spectral sensor in multiple bands, where the collection band range covers the visible light, near-infrared, and short-wave infrared regions. For each pixel point, extract the spectral reflectance value and calculate the spectral peak positions of multiple bands. The spectral peak position refers to the wavelength value corresponding to the maximum reflectance of a specific material within the spectral range. The specific calculation method is to calculate the first derivative of the spectral intensity for the spectral curves of different materials and obtain the zero-crossing position to obtain the spectral peak. Use the bandwidth at half maximum parameter to describe the width of the spectral peak, that is, the wavelength difference at half of the peak intensity of the spectral curve. The bandwidth at half maximum can be obtained by finding the two wavelength values corresponding to half of the peak intensity and calculating their difference. For example, when the spectral peak reflectance of a certain material is 75%, its bandwidth at half maximum can be determined by calculating the wavelength position difference at a reflectance of 37.5%. In addition, calculate the spectral slope parameter, which characterizes the change rate of the spectral curve within a specific wavelength range. The specific calculation method is to perform piecewise fitting on the spectral curve, calculate the slopes of each segmented interval, and select the slope values of the key bands. If the change rate of the spectral curve within a certain wavelength range is large, then this band may be a key feature point for material identification. For example, for the surface of a cement platform, its spectral reflectance change rate in the near-infrared band is much higher than that of metal materials, so it can be used as a key feature parameter. The spectral slope calculation formula is as follows:

[0067]

[0068] Among them, S is the spectral slope, R1 and R2 are the reflectivities at two wavelength points respectively, and λ1 and λ2 are the corresponding wavelength positions. Suppose the reflectivity of a certain material is 0.45 at a wavelength of 800 nm and 0.60 at a wavelength of 850 nm, then its spectral slope is calculated as follows:

[0069]

[0070] Finally, the spectral features at multiple positions of the platform are extracted, and a spectral feature parameter set is generated.

[0071] S102: Based on the spectral feature parameter set, call the spectral database of known materials, calculate the spectral similarity of materials at multiple positions of the platform, calibrate the material types at multiple positions on the surface of the platform, and obtain the material category distribution data of the platform;

[0072] The specific formula for calculating the spectral similarity of materials at multiple positions of the platform is:

[0073]

[0074] Calculate the spectral matching degree index;

[0075] Among them, S′ represents the matching degree index between the measured spectrum of the platform and the spectrum of the known material, A k represents the reflectivity value of the measured spectrum of the platform at the k-th wavelength point, B k represents the reflectivity value of the spectrum of the known material at the k-th wavelength point, K represents the total number of wavelength points used to calculate the spectral matching degree, and k represents the wavelength point index in the spectral measurement.

[0076] Formula:

[0077]

[0078] Detailed explanation of the formula and the derivation process of the formula calculation:

[0079] The formula is used to calculate the matching degree between the measured spectrum of the platform and the spectrum of the known material, and the result is used to identify the material types at each position on the surface of the platform;

[0080] Parameter meanings and set values:

[0081] A k represents the reflectivity value of the measured spectrum of the platform at the k-th wavelength point, and it is set that A1 = 0.42, A2 = 0.65, A3 = 0.81;

[0082] B k represents the reflectivity value of the spectrum of the known material at the k-th wavelength point, and it is set that B1 = 0.40, B2 = 0.70, B3 = 0.79;

[0083] K represents the total number of wavelength points used to calculate the spectral matching degree, which is set to 3;

[0084] k represents the wavelength point index in the spectral measurement;

[0085] Substitute the parameters into the formula for calculation:

[0086]

[0087] The spectral matching degree S′ = 0.9971, indicating that the spectral similarity between the platform measurement spectrum and the spectrum of the known concrete material is relatively high. Therefore, the material of this platform area can be determined as concrete. This value can be used as part of the material category distribution data of the platform surface to identify the material classification of the platform surface and to detect whether there are abnormal materials or obstacles in the platform area.

[0088] S103: Call the material category distribution data of the platform, detect the spectral reflection abnormal points in the platform area, identify obstacles, and record the position information in real time to generate the material classification feature value;

[0089] Call the material category distribution data of the platform, analyze whether the spectral reflectance of each measurement point conforms to the matching material type, detect the spectral reflection abnormal points in the platform area. For the pixel points whose reflectance values deviate significantly from the standard material range, record the changes in their spectral peak value and bandwidth at half maximum, calculate the spectral similarity between the abnormal points and the surrounding material points. If the spectral characteristics of the abnormal points are less similar to the spectral characteristics of the adjacent material measurement points than the set threshold of 0.85, then determine that this point may be an obstacle. By using the spatial clustering method, merge multiple continuous abnormal positions to form an obstacle area. At the same time, analyze the contour shape of the obstacle, calculate the change range of the spectral characteristics of the area covered by the obstacle. If the spectral change in the obstacle area is large, such as a large reflectance mutation on the glass surface or a non-linear spectral change in the concrete area, then further confirm that it is a non-platform material, calibrate the center position of the obstacle, record the coordinate information of the obstacle, and store the real-time detection data, calculate the reflectance abnormality degree of the obstacle:

[0090]

[0091] Among them, D r is the reflectance abnormality degree, R obs is the spectral reflectance of the current measurement point, R std is the spectral reflectance of the adjacent known material. For example, if the spectral reflectance of a certain measurement point is 0.75 and the reference reflectance of the adjacent concrete material is 0.50, then the calculation is as follows:

[0092]

[0093] When D rWhen it is >0.3, calibrate that this point may be an obstacle, and further perform obstacle contour extraction to obtain the material classification feature value.

[0094] According to the material classification feature value, use the image acquisition device to obtain the image of the platform area, extract the texture features of the clearance identification area, identify the blurred identification area by calculating the direction gradient change rate, and calculate the gradient direction interpolation of the blurred area to obtain the clearance identification correction value. The specific steps are as follows:

[0095] S201: Obtain the material classification feature value, call the image acquisition device, collect the image data of the platform area, identify the clearance identification area, calculate the pixel gray gradient of the target area, extract the local texture features according to the gradient direction distribution, and obtain the clearance identification texture feature data;

[0096] After obtaining the material classification feature value, call the image acquisition device, select a high-resolution camera to collect images of the platform area, ensure that the lighting conditions are balanced, avoid loss of image details caused by overexposure or underexposure. For the collected image data, locate the clearance identification area of the platform based on the pixel intensity distribution, use the color feature extraction method to identify the boundary of the clearance identification. Usually, the color of the clearance identification has standardized features. For example, there is an obvious contrast between the yellow or red lines and the surrounding background color. Further, use the edge detection method to obtain the contour information of the identification, eliminate the interference pixels in the non-target area, extract the pixel gray values of the target area, calculate the gray gradient to quantify the spatial change of the pixel brightness. The gray gradient calculation uses the Sobel operator. This operator obtains the gradient magnitude and direction by calculating the pixel gray difference in the horizontal and vertical directions. The specific calculation is as follows:

[0097]

[0098] Among them, G θ represents the gray gradient magnitude of the pixel point, G x and G y respectively represent the gray change rates in the horizontal and vertical directions. The calculation methods are as follows:

[0099] G x =I(i + 1,j)-I(i - 1,j), G y =I(i,j + 1)-I(i,j - 1);

[0100] Among them, I(i,j) represents the gray value of the pixel point (i,j). Assuming that the pixel gray values of a certain point are I(3,2)=120, I(5,2)=150, I(4,1)=110, I(4,3)=140 respectively, then calculate:

[0101] G x =150 - 120 = 30;

[0102] G y = 140 - 110 = 30;

[0103]

[0104] For pixel points with a gray - level gradient greater than the set threshold of 40, they are regarded as boundary - marker edge points. Further analyze the gradient - direction distribution, and use local texture descriptors to extract the detailed texture features of the marker area to obtain the boundary - marker texture - feature data.

[0105] S202: Based on the boundary - marker texture - feature data, by calculating the directional - gradient change rate of adjacent pixels, identify the blurred area of the marker, and mark the pixel range of the blurred area to obtain the boundary - marker blurred - area data;

[0106] Based on the boundary - marker texture - feature data, calculate the directional - gradient change rate of adjacent pixel points to determine whether there is a blurred situation in the boundary - marker area. For adjacent pixel points in the image, calculate the change rate of their gray - level gradients. If the change rate is lower than the set threshold, then this area may be blurred. The local - gradient - ratio method is used to calculate the directional - gradient change rate, and the specific calculation is as follows:

[0107]

[0108] where T θ is the directional - gradient change rate, G θ (p) is the gray - level gradient of pixel p, G θ (q) is the gray - level gradient of adjacent pixel q. If the gradient change rate between adjacent pixels is less than the set threshold of 0.1, then this area is blurred. For example, if the gray - level gradients of two adjacent pixels in a certain area are 42.4 and 40.0 respectively, the calculation is as follows:

[0109]

[0110] Since this value is less than 0.1, it is determined that this area is blurred. Further use the connected - component labeling method to label all pixel areas with a gradient change rate lower than the threshold to form a continuous blurred area. By calculating the pixel distribution of the blurred area, delimit the blurred boundary to obtain the boundary - marker blurred - area data.

[0111] S203: Based on the boundary - marker blurred - area data, calculate the interpolation parameter of the pixel - gradient direction within the blurred area to obtain the boundary - marker correction value;

[0112] Based on the data of the bounded identification fuzzy region, interpolation calculation is performed on the pixel gradient direction within the fuzzy region to correct the boundary information of the fuzzy region and improve the recognizability of the bounded identification. The interpolation calculation uses the bilinear interpolation method. Based on the gradient direction information of adjacent pixel points, gradient direction interpolation is performed on the current fuzzy pixel point, and the interpolation parameters are calculated as follows:

[0113]

[0114] Where, G interp is the interpolated gradient direction, G θ (a), G θ (b), G θ (c), G θ (d) are the adjacent gradient values in four directions of this pixel point. Assuming that the four neighborhood gradient values of a fuzzy region pixel point are 38.0, 39.5, 40.2, and 41.0 respectively, the calculation is as follows:

[0115]

[0116] After the interpolation calculation is completed, interpolation calculation is performed on all fuzzy region pixel points to correct their gradient directions, and a corrected bounded identification region is generated. By comparing the gradient distributions before and after interpolation, it is ensured that the boundary of the bounded identification is clearly distinguishable, and the bounded identification correction value is obtained.

[0117] Based on the bounded identification correction value, using a lidar to obtain the point cloud data of the platform area, combined with the image information, the steps for spatio-temporal synchronous registration of the platform point cloud data to construct a platform structure model are as follows:

[0118] S301: Obtain the bounded identification correction value, call the lidar to collect the point cloud data of the platform area, parse the spatial coordinate information of the point cloud data, and obtain the preliminary platform point cloud data;

[0119] After obtaining the bounded identification correction value, call the lidar to scan the platform area, collect three-dimensional point cloud data, parse the spatial coordinate information in the point cloud data. The lidar emits laser pulses and records the time difference from the emission to the return of the pulses. According to the speed of light, the spatial distance of each point is calculated. Set the measurement angle range θ of the lidar to 180°. The spatial coordinates (x, y, z) of each measurement point in the point cloud data are calculated as follows:

[0120] x = dcos(θ)cos(φ), y = dcos(θ)sin(φ), z = dsin(θ);

[0121] Where, d represents the laser ranging value, φ is the horizontal angle, and θ is the vertical angle. Assuming that the laser ranging value d of a certain measurement point is 2.5m, φ = 45°, and θ = 30°, the calculation is as follows:

[0122] x = 2.5 cos(30°) cos(45°) = 1.77 m;

[0123] y = 2.5 cos(30°) sin(45°) = 1.77 m;

[0124] z = 2.5 sin(30°) = 1.25 m;

[0125] Calculate the spatial coordinates of all points in sequence to establish the preliminary platform point cloud data. Remove the noise from the point cloud data, calculate the distance between adjacent points. If the distance between adjacent points is greater than the set threshold of 0.5 m, then remove this point to eliminate the abnormal points and obtain the preliminary platform point cloud data.

[0126] S302: Based on the preliminary platform point cloud data, call the image data of the platform area, calculate the spatial coordinate matching error between the point cloud data and the image data, and adjust the spatial coordinates of the point cloud data according to the alignment of the platform edge points and the limit identification points to obtain the platform matching point cloud data;

[0127] Based on the preliminary platform point cloud data, call the image data of the platform area, perform coordinate projection on each point in the point cloud data, project the point cloud data onto the image plane, and compare it with the platform edge points and the limit identification points in the image to calculate the position deviation between the point cloud projection points and the actual image points. The deviation calculation method uses pixel distance calculation, corresponding to the actual error value in the physical world. For the points with large errors, analyze their relative error distribution to determine whether the deviation is caused by the laser scanning angle, surface material reflection characteristics, or environmental occlusion during the acquisition of the point cloud data. If there is a systematic deviation, then adjust the coordinates according to the overall trend of the point cloud and perform local interpolation correction on the error points. Calculate the spatial coordinates of the platform edge points, and use the platform boundary information in the image to adjust the boundary position of the point cloud data to ensure that the boundary line in the point cloud data is aligned with the boundary line extracted from the image data. The adjustment of the limit identification points is based on the identification boundary in the image data. Calculate the actual deviation of the identification points in the point cloud data and perform global coordinate correction using weighted smoothing to reduce the overall error of the point cloud and make the point cloud data more accurately reflect the true spatial layout of the platform. Finally, the adjusted point cloud data will be used for the subsequent construction of the platform structure model to ensure the spatial accuracy of the platform point cloud data, improve the accurate measurement of the platform limit, and obtain the platform matching point cloud data.

[0128] S303: Based on the platform matching point cloud data, extract the structural form information of multiple areas of the platform, construct the spatial model of the platform, and obtain the platform structure model;

[0129] Based on the platform matching point cloud data, extract the structural form information of multiple areas of the platform, divide the platform structural features by region, calculate the point cloud density of different regions, and the point cloud density D is calculated as follows:

[0130]

[0131] Among them, N p is the number of point clouds per unit area, and A is the area of this region. Assume that the area A of a certain platform region is 4 m 2 , and the number of point clouds N p per unit region = 12000, then the calculation is as follows:

[0132]

[0133] If the point cloud density is lower than the set threshold of 2000 points / m 2 , then supplement the point cloud data of this region to improve the integrity of the model. Fit the point clouds of each region to establish structural surface models such as the platform ground, edge, and columns, and finally obtain the platform structure model.

[0134] Based on the platform structure model, by comparing with the platform structure data in the engineering design drawings, calculate the spatial offset and offset angle of multiple points. The specific steps for generating the structural offset value are as follows:

[0135] S401: Obtain the platform structure model, call the engineering design drawings, extract the spatial coordinates of the platform edge, platform facilities, and limit signs in the drawings, and obtain the platform reference coordinates;

[0136] After obtaining the platform structure model, call the engineering design drawings, parse the spatial coordinate information of the platform edge, platform facilities, and limit signs in the drawings, extract the coordinates of multiple key points from the design drawings, use the three-dimensional coordinate representation method, and set the platform edge point P e (x e , y e , z e ), the platform facility point P s (x s , y s , z s ), and the limit sign point P l (x l , y l , z l ) to form the reference coordinate data, calculate the Euclidean distance between each point, and use the formula:

[0137]

[0138] Among them, d ab is the spatial distance between two points, (x a , y a , z a ) and (x b , yb , z b ) are the coordinates of two points. Assume the edge point P of the platform e (10, 5, 2) and the limit marking point P l (12, 8, 2), then the calculation is as follows:

[0139]

[0140] Calculate the coordinates of all key points in sequence, establish the reference coordinate data of the platform, and store it for subsequent comparative analysis to provide a reference, and obtain the reference coordinates of the platform.

[0141] S402: Based on the reference coordinates of the platform, by calculating the spatial position difference between the point cloud data of the platform and the reference coordinates, detect multiple spatial offset points, and obtain the platform offset point data;

[0142] Based on the reference coordinates of the platform, call the point cloud data of the platform, calculate the spatial position difference between the point cloud data and the reference coordinates, select the same edge points, facility points and limit marking points of the platform in the point cloud data, and calculate the corresponding coordinates P′ e (x′ e , y′ e , z′ e ), P′ s (x′ s , y′ s , z′ s ), P′ l (x′ l , y′ l , z′ l ), and calculate the offset Δd of each point from the reference coordinates:

[0143]

[0144] Assume the detected coordinates of a certain edge point of the platform are P′ e (10.5, 5.8, 2.2), and the corresponding reference coordinates are P e (10, 5, 2), the offset is calculated as follows:

[0145]

[0146] If the offset is greater than the set threshold of 0.5 m, record this point as an offset point, screen multiple platform offset points, and obtain the platform offset point data.

[0147] S403: Based on the platform offset point data, calculate the spatial offset distances of multiple offset points, analyze the offset direction angles of multiple points, and obtain the platform structure offset value;

[0148] Based on the platform offset point data, calculate the spatial offset distances of multiple offset points, and analyze the offset direction angles of multiple points. First, calculate the offset direction angle θ of each offset point. The direction angle calculation formula is as follows:

[0149]

[0150] Suppose the detected coordinates of an offset point on a certain platform are P' e (11, 6, 2), and the corresponding reference coordinates are P e (10, 5, 2). Calculate the direction angle:

[0151]

[0152] Calculate the direction angles of all offset points, and classify the offset points into longitudinal offset, lateral offset, and height offset. Classify them as lateral offset according to the direction angle range of 0° - 45°, and as longitudinal offset according to 45° - 90°. Record the offset direction data of all offset points to obtain the platform structure offset value.

[0153] According to the structure offset value, extract the offset point information, call the train outline dimensions and running trajectory, and calculate the collision risk between the platform clearance and the train. The specific steps for generating the collision risk value are as follows:

[0154] S501: Obtain the platform structure offset value, extract the spatial coordinates of multiple offset points, calibrate the spatial distribution of the offset points, and obtain the platform clearance offset point data;

[0155] After obtaining the platform structure offset value, extract the spatial coordinate information of multiple offset points, represent the position of the offset points using a three-dimensional coordinate system, and set each offset point as P i (x i , y i , z i ), and calculate the spatial distribution of these points within the platform clearance. By calculating the geometric center of all offset points, define the centroid G(x g , y g , z g ):

[0156]

[0157] Among them, N is the total number of offset points, (x i , y i , z i ) is the coordinate of the i-th offset point. Suppose the detected offset points are P1(10, 5, 2), P2(12, 6, 2.5), and P3(11, 5.5, 2.2), then the calculation is as follows:

[0158]

[0159] Calculate the spatial distribution of all offset points, record the coordinates of the boundary offset points, and obtain the data of the platform boundary offset points.

[0160] S502: Based on the data of the platform boundary offset points, call the outer contour dimensions and running trajectory information of the train, calculate the spatial occupancy range of the train when passing through the platform boundary area, and obtain the train passing space data;

[0161] Based on the data of the platform boundary offset points, call the outer contour dimensions and running trajectory information of the train, establish a train passing space model, and set the outer contour dimensions of the train as W t , H t , L t (representing the width, height, and length of the train respectively), the running trajectory is defined by the center line C(x c , y c ), calculate the occupancy range of the train when passing through the platform boundary area, and set the maximum outer contour coordinates P of the train when it travels to the center position of the platform t (x t , y t , z t ), calculate the train envelope range within the platform boundary area. Taking the train width as an example, calculate the maximum horizontal expansion from the center point of the track to the side boundary of the train:

[0162]

[0163] Assume the train width W t = 3m, and the center position of the track x c = 10m, then:

[0164]

[0165] Calculate the maximum outer contour range of the train in the platform area in sequence to obtain the train passing space data.

[0166] S503: Based on the train passing space data, evaluate the collision risk by calculating the distance between the offset points and the train outer contour, and obtain the collision risk value;

[0167] The specific formula for evaluating the collision risk is:

[0168]

[0169] Calculate the collision risk coefficient;

[0170] Among them, R c is the collision risk coefficient, d s is the actual measured distance between the s-th offset point and the train outer contour, d safe is the safety distance in the platform boundary design standard, v sis the speed of the train passing through the offset point area, t s is the time required for the train to pass through the offset area, s is the index of the s-th offset point, S is the total number of offset points participating in the calculation of the collision risk, d s is the actual measured distance between the s-th offset point and the train outline, d safe is the platform safety distance standard determined according to the engineering design specifications.

[0171] Formula:

[0172]

[0173] Detailed explanation of the formula and the derivation process of the formula calculation:

[0174] The formula is used to calculate the collision risk coefficient between the platform limit offset point and the train, and the result is used to evaluate the safety impact of the platform structure deformation or facility offset on the train operation;

[0175] Parameter meaning and set values:

[0176] R c is the collision risk coefficient, reflecting the safety between the platform offset point and the train operation space;

[0177] d s is the actual measured distance between the s-th platform offset point and the train outline, set to 0.48m, 0.55m, 0.62m, 0.51m, 0.45m respectively;

[0178] d safe is the platform limit safety reference value, set to 0.5m;

[0179] v s is the speed of the train passing through this offset point area, and the set value is 15m / s;

[0180] t s is the time required for the train to pass through this offset point area, set t s = 0.5s;

[0181] S is the total number of platform offset points participating in the calculation, set to 5;

[0182] Substitute the parameters into the formula for calculation:

[0183]

[0184]

[0185] The result shows that the collision risk coefficient of the current platform limit offset point is 2.25, indicating that there is a relatively high collision risk at the offset point, and structural adjustment or boundary correction measures need to be taken.

[0186] Please refer to Figure 2 , an intelligent detection system for station platform clearance based on computer vision. The intelligent detection system for station platform clearance based on computer vision is used to execute the above-mentioned intelligent detection method for station platform clearance based on computer vision. The system includes:

[0187] The spectral analysis module is based on the spectral data of the platform area. By analyzing the spectral reflectance values of each pixel point, it calculates the spectral peak position, bandwidth at half maximum, and spectral gradient change, calls the known material spectral database to classify the platform material, detects obstacles, and obtains the classification characteristic values of the platform material;

[0188] The identification optimization module is based on the classification characteristic values of the platform material, collects the image data of the platform area, extracts the pixel characteristics of the clearance identification area, identifies the blurred identification area, and calculates the gradient direction interpolation of the blurred area to obtain the clearance identification correction value;

[0189] The structure construction module is based on the clearance identification correction value, calls the lidar to collect the point cloud data of the platform area, extracts the platform edge points, facility points, and clearance identification points, calculates the matching deviation between the image information and the point cloud data, performs spatio-temporal synchronous registration, adjusts the spatial coordinates of the point cloud data, and constructs the platform structure model;

[0190] The offset measurement module is based on the platform structure model, calls the engineering design drawing data, compares the offset situation between the point cloud data and the design standard, calculates the spatial offset of multiple points on the platform, and obtains the platform structure offset value;

[0191] The risk assessment module is based on the platform structure offset value, extracts the offset points within the platform clearance, calls the train outline size data and the running track, calculates the collision risk, and obtains the collision risk value.

[0192] The above embodiments can be implemented in whole or in part by software, hardware (such as circuits), firmware, or any combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions according to the embodiments of the present invention are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (such as infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or a data center that contains one or more collections of available media. The available media can be magnetic media (such as floppy disks, hard disks, magnetic tapes), optical media (such as DVDs), or semiconductor media. The semiconductor media can be solid-state drives.

[0193] It should be understood that the term "and / or" in this document is merely a description of the association relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. Here, A and B can be singular or plural. In addition, the character " / " in this document generally represents an "or" relationship between the associated objects before and after, but it may also represent an "and / or" relationship, which can be specifically understood with reference to the context before and after.

[0194] In the present invention, "at least one" means one or more, and "a plurality" means two or more. "At least one of the following" or its similar expressions refer to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b, or c can represent: a, b, c, a - b, a - c, b - c, or a - b - c, where a, b, and c can be single or multiple.

[0195] It should be understood that in various embodiments of the present invention, the magnitudes of the serial numbers of the above processes do not mean the order of execution. The order of execution of each process should be determined according to its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

[0196] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in connection with the embodiments disclosed herein can be implemented in electronic hardware, or in a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. Skilled professionals may use different methods for each specific application to implement the described functions, but such implementation should not be considered to exceed the scope of the present invention.

[0197] Those skilled in the art can clearly understand that for the sake of convenience and brevity of description, the specific working processes of the devices, apparatuses, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0198] In several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of the devices or units can be in electrical, mechanical, or other forms.

[0199] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0200] In addition, the functional units in each embodiment of the present invention can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit.

[0201] When the above-mentioned function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this 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 for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs.

[0202] As described above, the above are only specific implementation manners of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A station platform clearance intelligent detection method based on computer vision, characterized in that: The method comprises: S1: Obtain the spectral data of the platform area, extract the spectral features according to the reflectance value of each pixel point in the spectral range, compare them with the features of multiple known materials, identify the material types at multiple locations, detect obstacles, and generate material classification feature values; S2: according to the material classification feature value, using the image acquisition device to obtain the platform area image, extract the texture feature of the limit mark area, identify the mark fuzzy area by calculating the directional gradient change rate, and calculate the gradient direction interpolation of the fuzzy area to obtain the limit mark correction value; S3: Based on the limit mark correction value, the laser radar is used to obtain the point cloud data of the platform area, and the platform point cloud data is synchronously registered in time and space in combination with the image information to construct a platform structure model; S4: Based on the platform structure model, by comparing with the platform structure data in the engineering design drawings, the spatial offset and offset angle of multiple points are calculated to generate a structure offset value; S5: According to the structural offset value, the offset point information is extracted, the train outline dimensions and running track are called, the collision risk between the platform limit and the train is calculated, and a collision risk value is generated.

2. The method for intelligent detection of station platform limits based on computer vision according to claim 1 is characterized in that: The material classification characteristic value specifically includes material spectral reflectance parameters, material type matching results, and obstacle spectral characteristic values. The limit mark correction value includes the limit mark texture direction gradient change rate, the fuzzy area gradient interpolation calculation value, and the mark boundary correction parameter. The platform structure model specifically includes the platform edge point cloud coordinates, limit mark point cloud data, and platform facility point cloud data. The structure offset value includes the change in spatial coordinates of the offset point, the change in rotation angle of the offset point, and platform structure offset trend data. The collision risk value includes the platform offset distance, the platform offset angle, and the collision risk level.

3. The method for intelligent detection of station platform limits based on computer vision according to claim 1 is characterized in that: The steps to obtain the spectral data of the platform area, extract the spectral features according to the reflectance value of each pixel point in the spectral range, compare them with the features of multiple known materials, identify the material types at multiple locations, detect obstacles, and generate material classification feature values ​​are as follows: S101: Acquire the spectral data of the platform area, extract the reflectance value of each pixel point, calculate the spectral peak position, bandwidth half-height width, and spectral slope parameters of multiple bands, extract the spectral features of multiple positions, and generate a spectral feature parameter set; S102: Based on the spectral feature parameter set, calling a spectral database of known materials, calculating the spectral similarity of materials at multiple positions of the platform, calibrating the material types at multiple positions on the platform surface, and obtaining platform material category distribution data; S103: calling the platform material category distribution data, detecting spectral reflectance anomalies in the platform area, identifying obstacles, and recording location information in real time to generate material classification feature values.

4. The computer vision-based intelligent detection method for station platform limits according to claim 3 is characterized in that: The specific formula for calculating the material spectrum similarity of multiple positions of the station is: Calculate the spectral matching index; Among them, S′ represents the matching index between the station measurement spectrum and the known material spectrum, A k represents the reflectivity value of the spectrum measured by the station at the kth wavelength, B k Represents the reflectance value of the known material spectrum at the kth wavelength point, K represents the total number of wavelength points used to calculate the spectrum matching, and k represents the wavelength point index in the spectrum measurement.

5. The method for intelligent detection of station platform limits based on computer vision according to claim 1 is characterized in that: According to the material classification feature value, the platform area image is acquired by using an image acquisition device, the texture feature of the limit mark area is extracted, the blurred area of ​​the mark is identified by calculating the directional gradient change rate, and the gradient direction interpolation of the blurred area is calculated to obtain the limit mark correction value. Specifically, the steps are as follows: S201: Acquire the material classification feature value, call the image acquisition device, collect image data of the platform area, identify the boundary mark area, calculate the pixel grayscale gradient of the target area, extract local texture features according to the gradient direction distribution, and obtain the boundary mark texture feature data; S202: Based on the boundary mark texture feature data, by calculating the directional gradient change rate of adjacent pixels, identifying the mark fuzzy area, marking the pixel range of the fuzzy area, and obtaining the boundary mark fuzzy area data; S203: Based on the boundary mark fuzzy area data, calculate the interpolation parameters of the pixel gradient direction in the fuzzy area to obtain the boundary mark correction value.

6. The method for intelligent detection of station platform limits based on computer vision according to claim 1 is characterized in that: Based on the limit mark correction value, the point cloud data of the platform area is obtained by using laser radar, and the platform point cloud data is synchronously registered in time and space in combination with the image information. The steps of constructing the platform structure model are as follows: S301: Obtain the limit mark correction value, call the laser radar to collect point cloud data of the platform area, analyze the spatial coordinate information of the point cloud data, and obtain preliminary point cloud data of the platform; S302: Based on the preliminary point cloud data of the platform, the image data of the platform area is called, the spatial coordinate matching error between the point cloud data and the image data is calculated, and the spatial coordinates of the point cloud data are adjusted according to the alignment of the platform edge points and the limit identification points to obtain the platform matching point cloud data; S303: Based on the platform matching point cloud data, extract the structural morphological information of multiple areas of the platform, build a spatial model of the platform, and obtain a platform structural model.

7. The method for intelligent detection of station platform limits based on computer vision according to claim 1 is characterized in that: Based on the platform structure model, by comparing with the platform structure data in the engineering design drawings, the spatial offset and offset angle of multiple points are calculated, and the steps of generating the structure offset value are specifically as follows: S401: Acquire the platform structure model, call the engineering design drawing, extract the spatial coordinates of the platform edge, platform facilities, and limit marks in the drawing, and acquire the platform reference coordinates; S402: Based on the platform reference coordinates, a plurality of spatial offset points are detected by calculating the spatial position difference between the platform point cloud data and the reference coordinates, and the platform offset point data is obtained; S403: Based on the platform offset point data, calculate the spatial offset distances of multiple offset points, analyze the offset direction angles of the multiple points, and obtain the platform structure offset value.

8. The method for intelligent detection of station platform limits based on computer vision according to claim 1 is characterized in that: According to the structural offset value, the offset point information is extracted, the train outline dimensions and running track are called, and the collision risk between the platform limit and the train is calculated. The steps of generating the collision risk value are specifically as follows: S501: Obtain the platform structure offset value, extract the spatial coordinates of multiple offset points, calibrate the spatial distribution of the offset points, and obtain platform limit offset point data; S502: Based on the platform clearance offset point data, the train outline dimensions and running track information are called to calculate the space occupied by the train when passing through the platform clearance area, and obtain the train passage space data; S503: Based on the train passage space data, the collision risk is evaluated by calculating the distance between the offset point and the train outline to obtain a collision risk value.

9. The method for intelligent detection of station platform limits based on computer vision according to claim 8 is characterized in that: The specific formula for evaluating the collision risk is: Calculate the collision risk factor; Among them, R c is the collision risk coefficient, d s is the actual measured distance between the sth offset point and the train profile, d safe is the safety distance in the platform clearance design standard, v s is the speed of the train passing through the offset point area, t s is the time required for the train to pass through the offset area, s is the index of the sth offset point, S is the total number of offset points involved in calculating the collision risk, and d s is the actual measured distance between the sth offset point and the train profile, d safe It is the platform safety distance standard determined according to the engineering design specifications.

10. A station platform clearance intelligent detection system based on computer vision, characterized in that: According to the computer vision-based intelligent detection method for station platform limits according to any one of claims 1 to 9, the system comprises: The spectrum analysis module is based on the spectral data of the platform area. By analyzing the spectral reflectance value of each pixel point, the spectral peak position, the half-width of the bandwidth and the spectral gradient change are calculated, and the spectral database of known materials is called to classify the platform materials, detect obstacles, and obtain the classification characteristic values ​​of the platform materials. The identification optimization module collects image data of the platform area based on the platform material classification feature value, extracts pixel features of the boundary identification area, identifies the identification fuzzy area, and calculates the gradient direction interpolation of the fuzzy area to obtain the boundary identification correction value; The structure construction module uses the laser radar to collect point cloud data of the platform area based on the limit mark correction value, extracts the platform edge points, facility points and limit mark points, calculates the matching deviation between the image information and the point cloud data, performs time-space synchronous registration, adjusts the spatial coordinates of the point cloud data, and constructs the platform structure model; The offset calculation module calls the engineering design drawing data based on the platform structure model, compares the offset of the point cloud data with the design standard, calculates the spatial offset of multiple points of the platform, and obtains the platform structure offset value; The risk assessment module extracts the offset point within the platform limit based on the platform structure offset value, calls the train outline dimension data and running trajectory, calculates the collision risk, and obtains the collision risk value.

Images