CN116934680A - Method and equipment for detecting track surface block drop of turnout rail piece - Google Patents

Method and equipment for detecting track surface block drop of turnout rail piece Download PDF

Info

Publication number
CN116934680A
CN116934680A CN202310570814.8A CN202310570814A CN116934680A CN 116934680 A CN116934680 A CN 116934680A CN 202310570814 A CN202310570814 A CN 202310570814A CN 116934680 A CN116934680 A CN 116934680A
Authority
CN
China
Prior art keywords
rail
block
falling
starting point
profile
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202310570814.8A
Other languages
Chinese (zh)
Inventor
王玉林
黄珂
洪晓杰
王海涛
刘永奎
刘伟
张长领
王威
何明德
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
CENTRAL PLAINS LEADER RAILWAY TRACK TECHNOLOGY DEVELOPMENT CO LTD
Zhejiang Tiantai Hezhixiang Investment Co ltd
Zhejiang Yinlun Intelligent Equipment Co ltd
Original Assignee
CENTRAL PLAINS LEADER RAILWAY TRACK TECHNOLOGY DEVELOPMENT CO LTD
Zhejiang Tiantai Hezhixiang Investment Co ltd
Zhejiang Yinlun Intelligent Equipment Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by CENTRAL PLAINS LEADER RAILWAY TRACK TECHNOLOGY DEVELOPMENT CO LTD, Zhejiang Tiantai Hezhixiang Investment Co ltd, Zhejiang Yinlun Intelligent Equipment Co ltd filed Critical CENTRAL PLAINS LEADER RAILWAY TRACK TECHNOLOGY DEVELOPMENT CO LTD
Priority to CN202310570814.8A priority Critical patent/CN116934680A/en
Publication of CN116934680A publication Critical patent/CN116934680A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84—Systems specially adapted for particular applications
    • G01N21/88—Investigating the presence of flaws or contamination
    • G01N21/95—Investigating the presence of flaws or contamination characterised by the material or shape of the object to be examined
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B61—RAILWAYS
    • B61K—AUXILIARY EQUIPMENT SPECIALLY ADAPTED FOR RAILWAYS, NOT OTHERWISE PROVIDED FOR
    • B61K9/00—Railway vehicle profile gauges; Detecting or indicating overheating of components; Apparatus on locomotives or cars to indicate bad track sections; General design of track recording vehicles
    • B61K9/08—Measuring installations for surveying permanent way
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84—Systems specially adapted for particular applications
    • G01N21/88—Investigating the presence of flaws or contamination
    • G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/80—Analysis of captured images to determine intrinsic or extrinsic camera parameters, i.e. camera calibration
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
    • G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30108—Industrial image inspection

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Computing Systems (AREA)
  • Multimedia (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Medical Informatics (AREA)
  • Databases & Information Systems (AREA)
  • Biophysics (AREA)
  • Immunology (AREA)
  • Mathematical Physics (AREA)
  • General Engineering & Computer Science (AREA)
  • Chemical & Material Sciences (AREA)
  • Molecular Biology (AREA)
  • Biochemistry (AREA)
  • Analytical Chemistry (AREA)
  • Biomedical Technology (AREA)
  • Pathology (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Signal Processing (AREA)
  • Mechanical Engineering (AREA)
  • Quality & Reliability (AREA)
  • Machines For Laying And Maintaining Railways (AREA)

Abstract

The invention provides a method and equipment for detecting a falling block of a track surface of a turnout track piece, wherein the method comprises a turnout track piece identification step, a two-dimensional image extraction step, a track surface falling block identification step, a falling block effective point cloud generation step and a falling block characteristic size calculation step; due to the fact that the camera image and the track profile data acquired by the line laser sensor are combined, the method can efficiently identify the falling block from the image by utilizing the neural network, and further extracts the corresponding profile data to generate an effective point cloud for calculating the characteristic dimensions such as the depth of the falling block, so that the detection data of the falling block on the track surface can be obtained more comprehensively, the dimensions such as the depth of the falling block do not need to be manually complemented, the method can be applied to actual field detection, and the detection efficiency is improved.

Description

Method and equipment for detecting track surface block drop of turnout rail piece
Technical Field
The invention belongs to the technical field of switch rail piece detection, and particularly relates to a switch rail piece rail surface block drop detection method and device.
Background
The peeling off block of the rail surface of the rail member is a common damage, when the rail surface block is serious, the riding experience of passengers and even the running safety of a train can be influenced, and the rail surface block can possibly indicate further other problems, such as the problems of the material of the rail member, small curve radius of a track line, over-high running frequency, deviation of the geometric shape of the track and the like, so that the rail surface block is necessary to be detected and further analyzed.
There are many kinds of rail parts in the switch, including the point rail, wing rail, guard rail, etc., and it is necessary to detect the condition of the falling block of the rail surface of these rail parts. Because the turnout structure is more complicated, rail piece types are many, at present in actual on-site measurement, mainly still artifical observation rail face goes up the piece condition that falls, adopts corresponding instrument such as dipperstick to measure the specific size of falling piece and record again. The detection mode has low efficiency, and the manual detection mode is easy to miss due to the characteristics of random distribution, different sizes and discontinuous distribution of the rail surface falling blocks. Although some automatic detection methods for rail surface block fall exist in the prior art, the methods are difficult to be used for on-site detection of a turnout: on the one hand, in the switch detection, not only the rail surface falling blocks are required to be identified, but also the identified rail surface falling blocks are required to be positioned to corresponding rail pieces, and the prior method does not provide such a scheme; secondly, the existing method mainly utilizes an image algorithm or a neural network to identify the rail surface falling blocks from the camera image, the method cannot extract the depth of the falling blocks, and the depth of the falling blocks is an important index, so that in actual field detection, even if the existing method is adopted, the existing method still needs to manually supplement and measure the dimension data such as the falling block depth, and the like, and the efficiency in the field detection is limited.
Therefore, in order to improve the efficiency and coverage rate of the rail surface block drop detection in the switch detection, a new rail surface block drop detection method of the switch rail member is needed.
Disclosure of Invention
The invention aims to solve the problems, and aims to provide a detection method and equipment capable of automatically positioning a target turnout rail piece, detecting the peeling and block-falling condition of the rail surface and obtaining the corresponding block-falling characteristic size, wherein the detection method and equipment adopt the following technical scheme:
the invention provides a method for detecting the falling blocks of the rail surface of a turnout rail piece, which is used for detecting the falling blocks of the rail surface of the turnout rail piece and is characterized by detecting based on a two-dimensional image acquired by a camera, rail profile data acquired by a line laser sensor and corresponding mileage information, and comprises the following steps: the method comprises the following steps of: identifying the starting point of the turnout rail piece from the two-dimensional image or the multi-frame profile data based on the profile characteristics of the starting point of the turnout rail piece and obtaining the corresponding mileage information; a two-dimensional image extraction step: extracting a plurality of frames of two-dimensional images corresponding to the turnout rail piece as images to be detected based on the mileage information of the starting point of the turnout rail piece and the preset extension length of the turnout rail piece; rail surface block falling identification: inputting the images to be detected into a rail surface block falling identification model in sequence, and outputting block falling position information of rail surface block falling by the model, wherein the rail surface block falling identification model is a neural network model after training; a step of generating a block dropping effective point cloud: extracting a plurality of frames corresponding to the rail surface block from the plurality of frames of profile data based on the block-falling position information and calibration parameters of the camera and the line laser sensor, and stacking the frames to generate a block-falling effective point cloud; calculating the characteristic size of the falling block: and calculating the characteristic size of the track surface block based on the block falling position information and the block falling effective point cloud.
The method for detecting the track surface falling blocks of the turnout rail piece provided by the invention can also have the technical characteristics that the turnout rail piece comprises a center rail and wing rails arranged on two sides of the center rail, and the turnout rail piece identification step comprises the following steps: based on the mileage information, carrying out profile matching identification on the collected profile data frame by frame to identify the starting point of the section of the heart rail, and obtaining the corresponding mileage information; acquiring the profile data of a preset frame number connected after the starting point of the section of the heart rail based on mileage information of the starting point of the section of the heart rail, and stacking the profile data to generate a point cloud; based on the point cloud, identifying a theoretical starting point of the heart rail in a fitting plane mode, and obtaining corresponding mileage information; extracting profile data in a preset position range before the theoretical starting point of the heart rail based on the mileage information of the theoretical starting point of the heart rail; and in the two-dimensional image extraction step, a plurality of frames of two-dimensional images covering the heart rail and the wing rail are extracted based on the mileage information of the heart rail theoretical starting point, the preset extending length of the heart rail, the mileage information of the wing rail starting point and the preset extending length of the wing rail.
The method for detecting the falling blocks of the track surface of the turnout track piece provided by the invention can also have the technical characteristics that the method for identifying the theoretical starting point of the core track by a fitting plane mode comprises the following steps: fitting a plane between the starting point of the section of the heart rail and the theoretical starting point of the heart rail through a RANSAC algorithm based on the point cloud; fitting an inclined plane after the theoretical starting point of the heart rail through a RANSAC algorithm based on the point cloud; and calculating coordinates of points on an intersection line of the fitted plane and the fitted inclined plane, and obtaining corresponding mileage information based on the coordinates, or respectively calculating a central line of the fitted plane and a central line of the fitted inclined plane, calculating coordinates of an intersection point of the two central lines, and obtaining corresponding mileage information based on the coordinates.
The method for detecting the falling blocks of the track surface of the turnout rail piece provided by the invention can also have the technical characteristics that the profile matching of the starting point of the section of the turnout rail comprises the following steps: judging whether a profile exists at a preset position in the profile data based on a preset starting point matching template of the section of the heart rail, intercepting the profile according to a preset profile intercepting algorithm when the profile is judged to exist, obtaining the profile of the upper end face of the heart rail, judging whether the intercepted profile is a horizontal straight line segment with a preset length, identifying the starting point of the section of the heart rail when the intercepted profile is judged to be horizontal, taking mileage information corresponding to the profile data of the frame as the mileage information of the starting point of the section of the heart rail, and matching the profile of the starting point of the wing rail comprises the following steps: judging whether a profile exists at a preset position in the profile data based on a preset profile starting point matching template, intercepting the profile according to a preset profile intercepting algorithm when the profile exists, obtaining the profile of the rail head parts of the wing rails at two sides, matching the intercepted profile with a preset standard wing rail starting point profile, calculating the similarity, judging that the wing rail starting point is identified when the similarity reaches a preset similarity threshold, and taking the mileage information corresponding to the profile data of the frame as the mileage information of the wing rail starting point.
The method for detecting the falling blocks of the rail surface of the turnout rail piece provided by the invention can also have the technical characteristics that the rail surface falling block area identification model is an improved YOLOv5 neural network model, and the method comprises the following steps: the back box module is used for extracting the characteristics of the two-dimensional image, and an up-sampling layer is additionally arranged behind the back box module; the PAN+Bi-FPN module fuses the extracted features to obtain fusion features; and the Head module is used for identifying a block dropping area and a rail surface area in the two-dimensional image based on the fusion characteristics and outputting the pixel coordinates and the first confidence coefficient of the corresponding block dropping anchor frame and the pixel coordinates and the second confidence coefficient of the rail surface anchor frame.
The method for detecting the falling blocks of the rail surface of the turnout rail piece provided by the invention can also have the technical characteristics that the step of identifying the falling blocks of the rail surface further comprises the following steps: and removing the block dropping anchor frame with the first confidence coefficient lower than a preset confidence coefficient threshold value and the block dropping anchor frame positioned outside the rail surface anchor frame from the identified block dropping anchor frame.
The method for detecting the falling blocks of the track surface of the turnout track piece provided by the invention can also have the technical characteristics that the step of generating the falling block effective point cloud comprises the following steps: extracting corresponding multiframes from the profile data based on the mileage information corresponding to the image to be detected, and stacking the multiframes to generate a track point cloud; cutting the track part point cloud based on the distance between the line laser sensor and the track surface of the turnout track part to obtain a track surface point cloud; converting pixel coordinates of the block dropping anchor frame into camera coordinates based on camera internal parameters of the camera, and converting the camera coordinates of the block dropping anchor frame into world coordinates based on a calibration relation between the camera and the line laser sensor; and extracting a block falling effective point cloud corresponding to the block falling of the rail surface from the rail surface point cloud based on the world coordinates of the block falling anchor frame.
The method for detecting the falling blocks of the track surface of the turnout rail piece provided by the invention can also have the technical characteristics that the step of calculating the characteristic dimension of the falling blocks comprises the following steps: calculating the length and the width of the block falling anchor frame based on the world coordinates of the block falling anchor frame, and taking the maximum value as the length of the rail surface block falling; fitting a rail surface plane through a RANSAC algorithm based on the block falling effective point cloud; calculating the inclination angle of the fitted rail surface plane, and horizontally correcting the effective point cloud of the block falling based on the inclination angle; traversing in the horizontal corrected effective point cloud of the block falling, finding out the maximum value and the minimum value of Z-axis coordinate values of traversed points, and calculating the difference value between the maximum value and the minimum value as the depth of the block falling of the rail surface.
The method for detecting the falling blocks of the track surface of the turnout rail piece provided by the invention can also have the technical characteristics that after the step of calculating the characteristic dimension of the falling blocks, the method further comprises the following steps: and judging the severity of the block drop based on the characteristic size and a preset rail surface block drop severity standard, wherein the rail surface block drop severity standard comprises a plurality of severity levels and corresponding threshold information.
The invention provides a detecting device for detecting the falling blocks of the rail surface of a turnout rail piece, which is characterized by detecting based on a two-dimensional image collected by a camera, rail profile data collected by a line laser sensor and corresponding mileage information, and comprises the following components: the turnout rail piece identification part is used for identifying the starting point of the turnout rail piece from a plurality of frames of two-dimensional images or a plurality of frames of profile data based on the profile characteristics of the starting point of the turnout rail piece and obtaining corresponding mileage information; a two-dimensional image extraction unit configured to extract a plurality of frames of the two-dimensional image corresponding to the switch rail piece based on the mileage information of the start point of the switch rail piece and a predetermined extension length of the switch rail piece; the rail surface block falling recognition part is used for sequentially inputting the extracted two-dimensional images into a rail surface block falling recognition model, and outputting block falling position information of rail surface blocks, wherein the rail surface block falling recognition model is a neural network model after training; the block-falling effective point cloud generating part is used for extracting a plurality of frames corresponding to the rail surface block-falling from the plurality of frames of profile data based on the block-falling position information and calibration parameters of the camera and the line laser sensor, and stacking the frames to generate a block-falling effective point cloud; and a block-falling feature size calculation unit configured to calculate a feature size of the rail surface block based on the block-falling effective point cloud.
The actions and effects of the invention
According to the method and the device for detecting the falling blocks of the track surface of the turnout rail piece, the method comprises a turnout rail piece identification step, a two-dimensional image extraction step, a track surface falling block identification step, a falling block effective point cloud generation step and a falling block characteristic size calculation step, wherein the starting point of the turnout rail piece is identified in profile data based on the profile characteristics of the turnout rail piece, and corresponding two-dimensional images are extracted based on the starting point of the turnout rail piece and the preset extension length to identify the falling blocks of the track surface, so that the identified falling blocks of the track surface can correspond to the turnout rail piece of the corresponding type; due to the fact that the camera image and the track profile data acquired by the line laser sensor are combined, the rail surface falling block can be efficiently identified from the camera image by utilizing the neural network, corresponding profile data can be further extracted to generate an effective point cloud, the characteristic dimensions such as the falling block depth and the like can be calculated based on the effective point cloud, accordingly more comprehensive detection data of the rail surface falling block can be obtained, the dimension such as the falling block depth and the like can be not needed to be manually complemented, the rail surface falling block detection method can be applied to actual field detection, and the detection efficiency is improved. In addition, only partial profile data corresponding to the identified rail surface falling blocks are extracted to generate point clouds for further analysis, so that the profile data of the whole track is not needed to be analyzed, the operation amount is relatively small, and the application of field detection is very beneficial.
Drawings
FIG. 1 is a flow chart of a method for detecting a block drop on a rail surface of a switch rail member in an embodiment of the present invention;
FIG. 2 is a perspective view of a track inspection vehicle in accordance with an embodiment of the present invention;
FIG. 3 is a schematic view showing the distribution of sensors in a body unit according to an embodiment of the present invention;
FIG. 4 is a perspective view of a load bearing wheel assembly in an embodiment of the present invention;
FIG. 5 is a cross-sectional view of a load bearing wheel assembly in an embodiment of the present invention;
FIG. 6 is a side view of a compression assembly in an embodiment of the invention;
FIG. 7 is a perspective view of a compression assembly in accordance with an embodiment of the present invention;
FIG. 8 is a flowchart of the switch rail identification step in an embodiment of the present invention;
FIG. 9 is a schematic diagram of center rail section start profile identification in accordance with an embodiment of the present invention;
FIG. 10 is a flow chart of identifying a theoretical starting point of a heart rail by fitting a planar approach in an embodiment of the present invention;
FIG. 11 is a block diagram of a rail surface block identification model in an embodiment of the invention;
FIG. 12 is a flowchart of a step of identifying a block-down region in an embodiment of the invention;
FIG. 13 is a flowchart of a step of generating a block-down valid point cloud in an embodiment of the present invention;
FIG. 14 is a flowchart of a step of obtaining a block size in an embodiment of the present invention;
FIG. 15 is a block diagram of a railroad switch rail piece rail face block detection apparatus in accordance with an embodiment of the present invention;
FIG. 16 is a schematic view of a single switch in an embodiment of the invention;
FIG. 17 is a cross-sectional view of a center rail and its two side rails in accordance with an embodiment of the present invention;
FIG. 18 is a side view of a center rail near the start of the center rail in accordance with an embodiment of the present invention.
Reference numerals:
track inspection vehicle 100; a vehicle body 20; a vehicle body unit 21; a housing 211; bevel end 2111a; a camera 215; a refractive mirror 216; a first line laser sensor 217a; a second line laser sensor 217b; a third line laser sensor 217c; a fourth line laser sensor 217d; a carrying mechanism 30; a carrier wheel assembly 31; a carrier wheel bracket 311; a carrier wheel spindle 312; a carrier wheel 313; encoder accommodating groove 3133; a bearing 314; a brake member 315; a mileage encoder 317; a hold down assembly 32; a pressing bracket 321; a mounting block 3211; a guide rod 3212; a guide member 322; a guide rail 3221; a slider 3222; an elastic member 323; wheel body bracket 324; a pinch roller 326; a locking member 327; a lock fitting 3271; a locking member 3272; a wrench member 328; a pushing mechanism 40; the turnout rail piece rail surface block falling detection equipment 60; a model storage unit 601; a detection information storage unit 602; a switch drop block detection communication unit 603; a switch rail identification portion 604; a two-dimensional image extraction unit 605; a rail surface block-off recognition unit 606; a block-dropping effective point cloud generation unit 607; a block-down feature size calculation unit 608; a switch drop block information storage unit 609; a switch drop detection control unit 610; a single switch 90; a wing rail 96; a core rail 97; a theoretical starting point 971 for the heart rail; plane 972; an inclined surface 973; a core rail section start 974.
Detailed Description
In order to make the technical means, creation characteristics, achievement purposes and effects of the present invention easy to understand, the present invention's switch rail surface block drop detection method and apparatus are specifically described below with reference to the embodiments and the accompanying drawings.
Example 1
The embodiment provides a method and equipment for detecting falling blocks of a rail surface of a rail piece of a turnout, which are used for detecting falling blocks of the surface of the rail piece of the turnout, wherein the rail piece at the turnout comprises a plurality of stock rails, a point rail, a connecting rail, a point rail, wing rails, guard rails and the like.
Fig. 16 is a schematic view of the structure of the single turnout in this embodiment.
Fig. 17 is a cross-sectional view of the center rail and its both side rails.
As shown in fig. 16 to 17, taking the single switch 90 as an example, a center rail 97 (frog center) thereof is a variable cross-section rail member, a starting point thereof is in a triangular pointed shape, and a cross-section width thereof gradually becomes larger from the starting point to the other end thereof; the center rail 97 is provided with wing rails 96 on both sides, the middle part of the wing rails 96 is relatively close to the center rail 97, both ends of the wing rails 96 are bent outwards (towards the direction far away from the center rail), and the section shape of the wing rails 96 is similar to that of a stock rail.
Fig. 18 is a side view of the center rail near the start point of the center rail of the present embodiment.
As shown in fig. 18, between a center rail section starting point 974 and a center rail theoretical starting point 971, the upper end surface of the center rail 97 is a plane 972, and after the center rail theoretical starting point 971, the upper end surface of the center rail is an inclined surface 973 with a smaller size, and the intersection position of the plane 972 and the inclined surface 973 is the center rail theoretical starting point 971.
Fig. 1 is a flowchart of a method for detecting a block drop of a rail surface of a switch rail member in the present embodiment.
As shown in fig. 1, the method for detecting the track surface block drop of the turnout rail comprises the following steps:
the turnout data acquisition step S1: two-dimensional images of the turnout are collected through a camera, profile data of the turnout are collected through a line laser sensor, and mileage information is collected through a mileage encoder.
Switch rail piece identification step S2: and identifying the starting point of the turnout rail piece from the track profile data based on the profile characteristics of the starting point of the turnout rail piece to be detected, and obtaining corresponding mileage information.
Two-dimensional image extraction step S3: and extracting a two-dimensional image corresponding to the turnout piece based on mileage information of the turnout piece starting point and the preset extending length of the turnout piece.
The step S4 of identifying the block dropping area: the extracted two-dimensional images are sequentially input into a rail surface block-falling identification model, and the model outputs block-falling position information.
A step S5 of generating a block-dropping effective point cloud: and extracting a plurality of frames corresponding to the rail surface block from the multi-frame profile data based on the block-dropping position information and the calibration relation of the camera and the line laser sensor, and stacking the frames to generate a block-dropping effective point cloud.
A step S6 of obtaining the size of the falling block: and calculating the characteristic size of the track surface block based on the block-falling position information and the block-falling effective point cloud.
A block dropping degree judging step S7: and judging the severity of the rail surface falling block based on the length and the depth of the rail surface falling block and a preset rail piece falling block severity standard.
The above steps will be described in detail below.
The turnout data acquisition step S1: two-dimensional images of the turnout are collected through a camera, profile data of the turnout are collected through a line laser sensor, and mileage information is collected through a mileage encoder.
As one of examples, a track inspection vehicle provided with a camera, a line laser sensor, and a mileage encoder is shown below.
Fig. 2 is a perspective view of the rail inspection vehicle in the present embodiment.
As shown in fig. 2, the track inspection vehicle 100 includes a vehicle body 20, a carrying mechanism 30, and a pushing mechanism 40, wherein the vehicle body 20 is movably carried on the two normal rails 9 by the carrying mechanism 30 at the bottom, and the pushing mechanism 40 is mounted on the vehicle body 20 for a person to be inspected to push the vehicle body 20.
The body 20 is of a two-part construction and comprises two mirror-symmetrical body units 21. Each of the vehicle body units 21 includes a housing 211, and a sensor, an electronic control assembly, and the like provided in the housing 211.
Fig. 3 is a schematic diagram showing the distribution of the sensors in the vehicle body unit in the present embodiment.
As shown in fig. 3, a camera 215, a refractive mirror 216, a first line laser sensor 217a, a second line laser sensor 217b, a third line laser sensor 217c, and an illumination unit (not shown) are provided in the housing 211 of each vehicle body unit 21. The housing 211 has an opening below, through which the camera and the line laser sensor can detect the underlying positive line rail 9.
The camera 215 is mounted on the inside top of the housing 211 by a corresponding bracket, and is mounted laterally with its lens oriented in the horizontal direction. The refractive mirror 216 is a 45 ° refractive mirror, is also mounted on the top inside the housing 211 by a corresponding mount, and is mounted in front of the lens of the camera 215. By refraction of the refractor 216, the camera 215 is able to take a picture of the rail underneath it. By doing so, the mounting of the camera 215 is more stable and blocking of the line laser sensor can be avoided.
The lighting unit includes a plurality of lamps, also mounted on the inner top of the housing 211 through corresponding brackets, for providing sufficient illumination for photographing of the camera 215, and shielding of the housing 211 can reduce the influence of the change of the external ambient lighting conditions on photographing of the camera 215.
The first line laser sensor 217a is mounted in the housing 211 by a corresponding bracket and is located in the beveled end 2111a of the housing 211. The first line laser sensor 217a is mounted laterally. When the track inspection vehicle 100 is mounted on the stock rail, the first line laser sensor 217a is located on the outside of the stock rail, slightly higher than the stock rail, and faces the stock rail, and the line laser projected therefrom covers the inactive side of the stock rail, the jaw portion on the inactive side, and a part of the rail surface, so that corresponding point position data can be acquired.
The second line laser sensor 217b is obliquely mounted on the inside top of the housing 211 by a corresponding bracket. When the rail detection vehicle 100 is mounted on the stock rail, the second line laser sensor 217b is located above the outer side of the straight line rail and faces the stock rail, and the line laser projected by the second line laser sensor covers the inactive side, the waist portion on the inactive side, and the rail surface of the stock rail, so that corresponding point position data can be acquired.
The third line laser sensor 217c is obliquely mounted at the inside top of the housing 211 by a corresponding bracket, and is closer to the middle of the vehicle body 20 than the second line laser sensor 217 b. When the rail detection vehicle 100 is mounted on the stock rail, the third line laser sensor 217c is located above the inner side of the stock rail and faces the stock rail, and the line laser projected by the third line laser sensor covers the active side of the stock rail, the waist portion on the active side, and the rail surface, so that corresponding point position data can be acquired. When a switch rail member such as a tongue 92 is present inside the stock rail 91, the line laser projected by the third line laser sensor 217c can cover the upper end surface of the rail member.
The fourth line laser sensor 217d is also mounted obliquely to the inside top of the housing 211 by a corresponding bracket at approximately the same angle as the third laser sensor 217c, and is mounted closer to the middle of the vehicle body than the third laser sensor 217 c. When the rail detection vehicle 100 is mounted on the stock rail, the fourth line laser sensor 217d is located above the inside of the stock rail, near a position between the two stock rails, and toward a position beside the rail bottom of the stock rail, and when other switch rail pieces exist inside the stock rail, the line laser projected by the fourth line laser sensor 217d covers the upper end face and the inside face of the switch rail pieces. It can be seen that, taking the point rail 92 as an example, the line laser of the fourth line laser sensor 217d can cover the point rail 92 in the attached state (attached to the stock rail) and the point rail 92' in the detached state.
In this embodiment, the several line laser sensors on the rail inspection car 100 are all the same type.
As shown in fig. 2, the carrying mechanism 30 includes four carrying wheel assemblies 31 and four pressing assemblies 32. Four carrier wheel assemblies 31 are respectively mounted at both side lower portions in the width direction of the vehicle body 20, and are mounted in pairs at positions near one end in the length direction of the vehicle body 20, corresponding to the two rails 9, respectively. Four hold-down assemblies 32 are also mounted at both lower sides in the width direction of the vehicle body 20, and are respectively located beside the four carrier wheel assemblies 31. The track inspection device 100 is movably mounted on two rails 9 by four carrier wheel assemblies 31, and at this time, four hold-down assemblies 32 are respectively located inside the rails 9 on the corresponding sides.
Fig. 4 is a perspective view of the load wheel assembly in this embodiment.
Fig. 5 is a cross-sectional view of the load wheel assembly in this embodiment.
As shown in fig. 4 to 5, the carrier wheel assembly 31 includes a carrier wheel bracket 311, a carrier wheel rotation shaft 312, a carrier wheel 313, a bearing 314, and a brake member 315.
The two ends of the carrier wheel spindle 312 are mounted on the carrier wheel bracket 311 via two bearings 314, respectively. The carrier wheel 313 is mounted on a carrier wheel spindle 312 and is restrained by two bearings 314. In this embodiment, the carrying wheel 313 is made of an insulating material, preferably ceramic or POM plastic, and the insulation resistance value of the carrying wheel 313 is not less than 1mΩ, so as to avoid the detection device 100 from energizing the two rail connections and interfering with the circuitry of the rails.
The bearing wheels 313 have encoder receiving grooves 3133 inside, and distance detecting encoders 317 are provided in the encoder receiving grooves 3133 of two of the bearing wheels 313 for acquiring corresponding mileage information when the bearing wheels 313 roll along the rails 9. The size of the encoder accommodating groove 3133 is slightly larger than the external size of the distance detecting encoder 317, so that the distance detecting encoder 317 is not in direct contact with the carrier wheel 313, and the rotation of the carrier wheel 313 is prevented from being affected by friction.
The structure of the vehicle body 20 and the mounting position of the line laser sensors are such that a plurality of line laser sensors can scan the same cross-sectional profile data of the positive line rail 9. According to the control of the corresponding controller, the plurality of line laser sensors synchronously scan every time a preset time or a preset distance is passed in the moving process of the cart. Each frame of point location data (point cloud) acquired by the line laser sensor contains hundreds to thousands of coordinate points, and each coordinate point contains mileage information (Y axis), height information (Z axis), width information (X axis) and brightness information.
Similarly, according to the control of the corresponding controller, the camera performs shooting once after a predetermined distance is passed or the predetermined distance is moved in the moving process of the cart, so as to obtain a frame of two-dimensional image. In this embodiment, a two-dimensional image is acquired at intervals of 500mm, and a frame of contour point location data is acquired at intervals of 2 mm.
In addition, before data acquisition is started, the camera and all the line laser sensors are calibrated, and corresponding calibration parameters are obtained.
Fig. 6 is a side view of the pressing assembly in this embodiment.
Fig. 7 is a perspective view of the pressing assembly in the present embodiment.
As shown in fig. 6 to 7, the pressing unit 32 includes a pressing bracket 321, a guide member 322, an elastic member 323, a wheel body bracket 324, a pressing wheel spindle (not shown), a pressing wheel 326, a lock member 327, and a wrench member 328.
The pressing bracket 321 includes a mounting block 3211 and a guide rod 3212. The guide rod 3212 is a cylindrical rod extending in the same direction as the longitudinal direction of the vehicle body 20.
The guide member 322 includes a guide rail 3221 and a slider 3222, the guide rail 3221 being fixedly mounted on the housing 211 in the same direction as the guide rod 3212. A slider 3222 is slidably mounted on rail 3221.
The elastic member 323 is a spring, and is also sleeved on the guide rod 3212, one end of the spring is abutted against the mounting block 3211, and the other end of the spring is embedded into a cylindrical groove on the wheel body bracket 324 and abutted against the bottom of the groove.
The pinch roller axle is mounted in a notch in the end of the wheel mount 324, and pinch roller 326 is rotatably mounted on the pinch roller axle and partially embedded in the notch. The pinch roller 326 is also made of the insulating material described above.
Therefore, the wheel body bracket 324 and the pressing wheel 326 thereon can be pressed toward the inner side of the rail under the action of the spring force of the elastic member 323, so that the pressing wheel 326 is closely attached to the active edge of the rail. In this embodiment, when the rail car 100 is mounted on a rail, the pinch roller 326 is located 16cm below the rail surface of the rail.
The lock member 327 includes a lock mating member 3271 and a lock member 3272.
The locking members 3272 are locking screws mounted to the mounting block 3211 at corresponding locking member bores along which screw ends are movable.
The locking fitting 3271 has a bar shape, and a plurality of circular holes are formed along a length direction thereof. One end of the locking fitting 3271 is mounted on the wheel body bracket 324, and the other end passes through a locking fitting hole in the mounting block 3211 and is movable along the hole. The locking piece hole is communicated with the locking matching hole. When the screw end portion of the lock 3272 protrudes downward, the screw end portion passes through the lock hole to reach the lock engagement hole, and passes through the circular hole on the end of the lock engagement member 3271, thereby fixing (locking) the relative positions of the wheel body bracket 324 and the mounting block 3211.
In this embodiment, in the two pressing assemblies 32 corresponding to one of the rails 9, the locking member 3272 is in an unlocked state, and the elastic member 323 thereof is freely movable; in the other two pressing assemblies 32, the locking member 3272 is locked, and the wheel body bracket 324 and the mounting block 3211 are fixed in position, that is, the pressing wheel 326 is fixed in position relative to the vehicle body 20. That is, in this embodiment, two hold-down assemblies 32 on one side are in the form of fixed side wheels and two on the other side are in the form of spring side wheels.
The wrench member 328 includes two combined links for allowing a inspector to adjust the relative position of the pinch roller 326 with respect to the vehicle body 20 to further facilitate placement of the rail inspection vehicle 100 onto a rail.
The pushing mechanism 40 includes a cart lever, and a notebook bracket mounted at an end of the cart lever. The angle of the trolley rod is adjustable, and the angle of the notebook bracket relative to the trolley rod is also adjustable.
In step S1, the inspector mounts the rail inspection vehicle 100 on the switch rail, starts the on-vehicle camera, the line laser sensor, the data acquisition device, and the like, and then pushes the rail inspection vehicle 100 along the switch rail, and the rail inspection vehicle 100 automatically acquires rail data according to the setting during the movement.
Switch rail piece identification step S2: based on the profile characteristics of the starting point of the switch rail piece to be detected, the starting point of the switch rail piece is identified from the profile data, and corresponding mileage information is obtained.
As described above, in the present embodiment, the switch rail member to be detected includes the center rail and the wing rails on both sides thereof, the start points of the center rail are first identified and positioned, and then the start points of the wing rails on both sides thereof are further identified and positioned based on the identified start points of the center rail.
Fig. 8 is a flowchart of the switch rail identification step in this embodiment.
As shown in fig. 8, the switch rail member identification step S2 specifically includes the steps of:
step S2-1: based on mileage information, carrying out profile matching identification on the starting point of the section of the heart rail frame by frame on the acquired profile data, identifying the starting point of the section of the heart rail and obtaining corresponding mileage information.
Fig. 9 is a schematic diagram showing the identification of the start point of the section of the center rail in this embodiment.
As shown in fig. 9, the contour of the heart rail is square at the beginning of the cross-section of the heart rail. In step S2-1, based on a predetermined rail section start point matching template, it is determined whether or not there is a profile at a predetermined position (middle portion between two wing rails) in profile data, when it is determined that there is a profile, the profile is intercepted according to a predetermined profile interception algorithm to obtain a profile (substantially horizontal straight line segment) of an upper end face of the rail, it is determined whether or not the intercepted profile is a horizontal straight line segment having a predetermined length, and when it is determined that there is a profile, the rail section start point is identified, and mileage information corresponding to the profile data frame is used as mileage information of the rail section start point.
Step S2-2: based on mileage information of the starting point of the section of the heart rail, profile data of a continuous preset frame number after the starting point of the section of the heart rail is obtained, and the profile data are stacked to generate a point cloud.
Each frame of point location data is data of an X-Z plane coordinate system, each point contains corresponding mileage information, and multiple frames of point location data are stacked along the Y-axis direction according to the mileage information and converted into three-dimensional point clouds according to a corresponding algorithm.
Step S2-3: and identifying a theoretical starting point of the heart rail by a fitting plane mode based on the generated point cloud, and obtaining corresponding mileage information.
Fig. 10 is a flowchart for identifying a theoretical starting point of a heart rail by fitting a plane in the present embodiment.
As shown in fig. 10, the step S2-3 specifically includes the following sub-steps:
step S2-3-1: based on the point cloud, a plane between a heart rail section starting point and a heart rail theoretical starting point is fitted through a RANSAC algorithm.
Step S2-3-2: based on the point cloud, fitting an inclined plane after the theoretical starting point of the heart rail through a RANSAC algorithm.
Step S2-3-3: and calculating Y-axis coordinates of points on an intersection line of the fitted inclined plane and the fitted plane, and obtaining corresponding mileage information based on the Y-axis coordinates of the points of the intersection line, wherein the mileage information is used as mileage information of a theoretical starting point of the heart track.
Alternatively, the center line of the fitted inclined plane and the center line of the fitted plane (both the center lines along the Y-axis direction) may be calculated, and the Y-axis coordinates of the intersection point of the two center lines may be calculated, so as to obtain the corresponding mileage information.
The fitting plane through the RANSAC algorithm is specifically as follows:
at least three points (the minimum number of points that can fit a planar model) are randomly sampled in the corresponding three-dimensional point cloud as initial points. An initial equation ax+by+cz+d=0 of the plane model is set, and the plane model is fitted based on a plurality of the initial points, to obtain values of the plane model parameters A, B, C, D. In the three-dimensional point cloud, points other than the initial point are traversed, and the distance d between the traversed points and the plane model is calculated:
wherein x is 0 、y 0 、z 0 Is the coordinates of the points traversed. And entering the points with the distance d smaller than a preset distance threshold into an inner group, calculating the proportion of the points of the inner group to the total points of the three-dimensional point cloud, and obtaining updated plane model parameters based on the inner group points when the proportion is larger than the preset proportion threshold so as to complete fitting of corresponding planes.
Step S2-4: and extracting profile data in a preset position range before the theoretical starting point of the heart rail based on mileage information of the theoretical starting point of the heart rail.
The start points of the wing rails on both sides of the head rail are located at a predetermined distance in front of the theoretical start point of the head rail, and therefore, the approximate position range of the start point of the wing rail is inferred based on the identified theoretical start point of the head rail, and only a small amount of profile data in the position range is extracted to further identify the start point of the positioning wing rail.
Step S2-5: and carrying out profile matching identification on the extracted profile data frame by frame on the starting points of the wing rails, identifying the starting points of the wing rails and acquiring corresponding mileage information.
Before the start point of the wing rail, the collected profile has no wing rail profile (a gap between the connecting rail and the wing rail), and the collected profile further has the profile of the wing rails at the two sides at the start point of the wing rail. The profile of the wing rail is similar to that of the stock rail, and the distance between the wing rails at the starting point of the wing rail is a preset value, so that the profile data frame of the starting point of the wing rail can be identified based on the profile of the wing rail and the distance between the wing rails at the two sides.
Specifically, based on a preset wing rail starting point matching template, judging whether a profile exists in two preset position ranges (positions corresponding to two wing rail profiles at the wing rail starting point) at first; when judging that the profile exists, intercepting the profiles in two preset position ranges according to a preset profile intercepting algorithm, and intercepting the head part profile in the two wing rail profiles; and then, carrying out profile matching on the cut-out head part profile and a standard wing rail starting point profile, calculating the similarity, if the calculated similarity reaches a preset similarity threshold value, judging that the wing rail starting point position is identified in the current profile data frame, and taking mileage information corresponding to the frame profile data as mileage information of a wing rail starting point.
Two-dimensional image extraction step S3: based on mileage information of a starting point of a turnout rail piece and the preset extending length of the turnout rail piece, a plurality of two-dimensional images corresponding to the turnout rail piece are extracted and used as images to be detected.
Based on the mileage value of the starting point of the heart rail, the preset extending length of the heart rail, the mileage value of the starting point of the wing rail and the preset extending length of the wing rail, the mileage range from the starting point of the heart rail to the ending point of the wing rail and the mileage range from the starting point of the wing rail to the ending point of the wing rail can be calculated, and all two-dimensional images in the mileage range are extracted, wherein the two-dimensional images cover the whole section of the heart rail and the wing rail.
The step S4 of identifying the block dropping area: the extracted images to be detected are sequentially input into a rail surface block falling identification model, and the model outputs the block falling position information of the identified rail surface block falling.
Fig. 11 is a block diagram of a rail surface block identification model in the present embodiment.
As shown in fig. 11, in this embodiment, the rail surface block falling area recognition model is a model improved based on a YOLOv5 neural network model, and includes a back bone module, a pan+bi-FPN module, and a Head module, and compared with the existing YOLOv5 convolutional neural network model, the improvement lies in: an up-sampling layer is added behind the Backbone module to improve the resolution of the feature map, so that the recall rate of the small-size block dropping can be improved. The original FPN structure is replaced by a PAN+Bi-FPN structure to deepen the network, so that the feature fusion performance of the model is improved.
When training the model, a large number of rail face block images, corresponding block labels and rail face labels are prepared first and used as training data. The steel rail surface block dropping image is a two-dimensional image of the steel rail, which is acquired in the same way as the above way, and the block dropping label and the steel rail surface label are marked manually by a detection personnel with abundant experience. Then, data enhancement is performed on the training data to improve the robustness and generalization capability of the trained model. And then training the rail surface block falling area recognition model by using the enhanced training data.
In this embodiment, the data enhancement algorithm includes performing geometric transformation on the two-dimensional image of the track, including clipping, scaling, translation, inversion, rotation, etc., to obtain more data sets with different sizes in a physical sense, and then performing data transformation of color layer transformation on the two-dimensional image of the track, including noise addition, color disturbance, blurring, etc. By means of the data enhancement algorithm, a new data set several times larger than the original training data set is obtained through generalization.
The training is used for reasoning by using a rail surface falling block area identification model, the input of the training is a rail two-dimensional image acquired by a camera, the model fuses the extracted features through a PAN and Bi-FPN structure of the training, then the fused features are output to a Head module, the Head module comprises a multi-scale detection Head, and the Head module outputs the position coordinates of a falling block anchor frame for identifying the falling block area in the two-dimensional image and the confidence coefficient thereof (marked as first confidence coefficient) and the position coordinates of the rail surface anchor frame for identifying the rail surface area of the rail part and the confidence coefficient thereof (marked as second confidence coefficient). Specifically, the anchor frame is a square frame, and its position coordinates include its upper left corner coordinates and its lower right corner coordinates, where the coordinates refer to pixel coordinates of the two-dimensional image.
It will be appreciated that the drop block anchor frame should be located within the rail anchor frame. Therefore, if a certain output falling block anchor frame is positioned outside the rail surface anchor frame, the falling block anchor frame is filtered out.
Fig. 12 is a flowchart of the step of identifying a block-dropping area in the present embodiment.
As shown in fig. 12, the block-dropping area identifying step S4 specifically includes the steps of:
step S4-1: and preprocessing the extracted image to be detected.
Step S4-2: and inputting the preprocessed image to be detected into a rail surface block dropping recognition model, and outputting a block dropping anchor frame and a first confidence coefficient thereof and a rail surface anchor frame and a second confidence coefficient thereof by the model.
Step S4-3: and removing the block-dropping anchor frame which is positioned outside the rail surface anchor frame and has the first confidence coefficient lower than a preset confidence bottom threshold value.
A step S5 of generating a block-dropping effective point cloud: and extracting multi-frame profile data corresponding to the track surface block drop from the profile data based on the block drop position information and calibration parameters of the camera and the line laser sensor, and generating a block drop effective point cloud.
Fig. 13 is a flowchart of the step of generating a block-down valid point cloud in the present embodiment.
As shown in fig. 13, the block-dropping valid point cloud generating step S5 specifically includes the following steps:
step S5-1: based on mileage information corresponding to one frame of image to be detected, extracting corresponding multi-frame profile data from all the acquired profile data.
Step S5-2: and stacking the extracted multi-frame profile data along the Y axis according to mileage information thereof, and converting the multi-frame profile data into track point clouds through a corresponding conversion algorithm.
Step S5-3: and cutting the track point cloud based on the distance between the line laser sensor and the track surface of the turnout track piece to obtain the track surface point cloud.
Step S5-4: and converting the pixel coordinates of the falling block anchor frame into camera coordinates based on camera internal parameters, and converting the camera coordinates of the falling block anchor frame into world coordinates based on the calibration relation of the camera and the line laser sensor.
That is, the coordinates of the pixel points are corrected through the camera internal reference matrix to obtain the actual pixel positions, and then the pixel positions are projected into the world coordinate system of the three-dimensional point cloud.
Wherein, let camera internal reference matrix be:
the coordinates (u, v) of the pixel points in the anchor frame are converted into camera coordinates (X, Y, Z) according to the internal reference matrix:
the camera coordinates (X, Y, Z) can be obtained as:
further, camera coordinates P C Generating world coordinates P by rotating matrix R and offset matrix t W ：
P C ＝RP W +t
The rotation matrix R and the offset matrix t can be obtained according to calibration parameters.
Solving world coordinates P by the following matrix W ：
Step S5-5: and extracting partial point clouds corresponding to the falling blocks of the track surface from the track surface point clouds based on the world coordinates of the falling block anchor frames, and taking the partial point clouds as falling block effective point clouds.
Through the step, the small-range point cloud corresponding to the block dropping area is extracted for further analysis, and compared with the point cloud of the whole section of steel rail piece, the data volume is greatly reduced.
A step S6 of obtaining the size of the falling block: and calculating the characteristic size of the track surface block based on the block-falling position information and the block-falling effective point cloud.
In this embodiment, the length and depth of the rail face drop block are calculated.
Fig. 14 is a flowchart of the block size acquisition step in the present embodiment.
As shown in fig. 14, the block size obtaining step S6 specifically includes the steps of:
step S6-1: based on the world coordinates of the falling block anchor frame, calculating the length and the width of the falling block anchor frame, and taking the maximum value as the length of the falling block of the rail surface.
Step S6-2: and fitting a rail plane by a RANSAC algorithm based on the block-falling effective point cloud.
Step S6-3: and calculating the inclination angle of the fitted rail plane along the Y axis, and horizontally correcting the effective point cloud of the falling block based on the inclination angle.
Step S6-4: traversing in the horizontal corrected effective point cloud of the falling block, finding out the maximum value and the minimum value in Z-axis coordinate values of traversed points, and calculating the difference value between the maximum value and the minimum value as the depth of the falling block of the rail surface.
The method of fitting the rail plane by using the RANSAC algorithm is similar to the above method, that is, a plane model equation is set, and parameters A, B, C, D of the plane model are updated after the inner group points with a sufficient proportion are obtained. After this, in step S6-3, the tilt angle of the fitted rail plane along the Y-axis is calculated according to the following formula:
And obtaining a rotation matrix of the effective point cloud of the falling block along the Y axis based on the inclination angle theta:
and through the rotation matrix, the effective point cloud P of the falling block is obtained raw Horizontal correction is carried out:
P rotate ＝R y (θ)P raw
a block dropping degree judging step S7: and judging the severity of the rail surface falling block based on the length and the depth of the rail surface falling block and a preset rail piece falling block severity standard.
The standard of the severity of the falling block of the rail member includes the severity level of the falling block and the corresponding threshold value information, and in this embodiment, the standard is specifically shown in the following table 1.
TABLE 1 Rail surface blocking severity threshold information Table
Thus, based on the acquired set of length and depth data for the chunking region, a determination is made as to which threshold range in the table the set of data falls within, i.e., the severity of the chunking region.
In this embodiment, the two-dimensional image, mileage information, a track surface point cloud, a block-drop effective point cloud, the length and depth of the track surface block and the severity corresponding to each identified track surface block are stored correspondingly, and are used as the block-drop information of the switch track piece. Thus, the detection personnel can conveniently check the automatically identified rail surface falling blocks and the information thereof and further analyze the rail surface falling blocks. For example, a detector can screen out the block falling information with serious injury from the stored block falling information, and the block falling information is preferentially processed; or a detection person can call out the two-dimensional image of the block dropping area and point cloud data for rechecking; or the detection personnel can position the corresponding position on the steel rail piece according to the mileage information corresponding to the block dropping area and go to the position for further manual detection.
Fig. 15 is a block diagram of the railroad switch rail piece face block detection apparatus in the present embodiment.
As shown in fig. 15, the present embodiment also provides a railroad switch rail piece face drop block detection apparatus 60 corresponding to the above-described method, which includes a model storage 601, a detection information storage 602, a railroad switch drop block detection communication 603, a railroad switch rail piece identification 604, a two-dimensional image extraction 605, a face drop block identification 606, a drop block valid point cloud generation 607, a drop block feature size calculation 608, a railroad switch drop block information storage 609, and a railroad switch drop block detection control 610.
The model storage unit 601 stores the above-described rail surface block recognition model (neural network model). The detection information storage unit 602 stores parameters necessary for detecting the track surface drop of the switch rail. The switch drop detection communication unit 603 is configured to communicate with other devices, and includes acquiring two-dimensional images, profile data, and mileage information acquired from a data acquisition device of the track inspection vehicle 100. The switch rail identifying unit 604 identifies the start point of the switch rail from the two-dimensional image or profile data in the manner described in step S2. The two-dimensional image extracting unit 605 extracts a plurality of frames of two-dimensional images corresponding to the switch rail members for detecting a block drop in the method of step S3. The rail surface block identification unit 606 inputs the extracted two-dimensional image into the rail surface block identification model in the above-described method of step S4, and identifies a rail surface block and acquires the positional information thereof. The block-falling effective point cloud generation unit 607 extracts a block-falling effective point cloud based on the position information of the identified rail surface block according to the method of step S5. The block-off feature size calculating unit 608 calculates the feature size of the track-face block according to the method of step S6. The switch piece information storage unit 609 is configured to store the switch piece information. The switch drop detection control unit 610 is used to control the operations of the above-described functional units.
As one example, the rail surface drop detection device 60 is a notebook computer provided with a corresponding detection program, which is mounted on a notebook carrier of the rail detection vehicle 10 and connected to a vehicle-mounted data acquisition device by a cable. The detection personnel can conveniently obtain the detection result through the notebook computer.
In this embodiment, the portions not described in detail are known in the art.
Example operation and Effect
According to the method and the device for detecting the falling blocks of the track surface of the turnout rail piece, the method comprises a turnout rail piece identification step, a two-dimensional image extraction step, a track surface falling block identification step, a falling block effective point cloud generation step and a falling block characteristic size calculation step, wherein the starting point of the turnout rail piece is identified in profile data based on the profile characteristics of the turnout rail piece, and the corresponding two-dimensional image is extracted based on the starting point of the turnout rail piece and the preset extension length to identify the falling blocks of the track surface, so that the identified falling blocks of the track surface can correspond to the turnout rail piece of the corresponding type; due to the fact that the camera image and the track profile data acquired by the line laser sensor are combined, the rail surface falling blocks are efficiently identified from the camera image by utilizing the neural network, corresponding profile data can be further extracted to generate effective point clouds, the characteristic dimensions such as the falling block depth and the like can be calculated based on the effective point clouds, accordingly more comprehensive detection data of the rail surface falling blocks can be obtained, the dimension such as the falling block depth and the like does not need to be manually complemented, the rail surface falling block detection method can be applied to actual field detection, and the detection efficiency is improved. In addition, only partial profile data corresponding to the identified rail surface falling blocks are extracted to generate point clouds for further analysis, so that the profile data of the whole track is not needed to be analyzed, the operation amount is relatively small, and the application of field detection is very beneficial.
In the embodiment, the rail data is collected on site by adopting the rail detection vehicle with the camera, the line laser sensor and the mileage encoder, the detection personnel only need to start the equipment and move the trolley along the rail to collect the data, the operation is convenient, and the trolley can be used for acquiring various data required by detection after one-time walking. Further, the camera is arranged in the shell of the detection vehicle, and the lighting unit is further arranged in the shell, so that the influence of external light condition change on camera shooting can be reduced, enough lighting can be provided for camera shooting, and the quality of two-dimensional images acquired by the camera is improved.
Further, a notebook computer is further arranged on the trolley handle of the track detection trolley, and the notebook computer is stored with a preset detection program and is connected with the camera and the sensor through the data acquisition device, so that detection personnel can conveniently acquire the condition of real-time data acquisition and the detection result of the rail surface falling block through the notebook computer, discover problems in time and adjust or re-measure the problems.
In an embodiment, an improved YOLOv5 neural network model is used to automatically identify the chunking region. The last up-sampling layer is additionally arranged in the back bone module of the model, so that the resolution of the feature map can be improved, and the recall rate of the small-size block dropping is improved; the FPN structure in the original model is optimized to be a PAN+Bi-FPN structure, the network is deepened, and the characteristic fusion performance of the model is improved, so that the rail surface block falling identification is performed by adopting the model, and the identification accuracy is higher.
In the embodiment, a theoretical starting point and a wing rail starting point of a point rail are identified based on profile matching and plane fitting modes, so that multi-frame two-dimensional images covering rail pieces can be extracted based on the identified starting point of the rail pieces and the preset extending length of the rail pieces for detecting falling blocks, the detected falling blocks can be corresponding to the rail pieces of corresponding types, and automatic detection of falling blocks on rail surfaces of the multi-rail pieces in the point region can be realized. In profile matching, the intercepting and matching recognition algorithm related to each frame of profile data is relatively simple, the calculated amount is small, the starting point of the corresponding track piece can be rapidly recognized, and the recognized starting point position is accurate due to the fact that the collected profile data are relatively dense.
In the embodiment, in all the acquired profile data, only a small-range track surface point cloud is extracted as an effective point cloud according to the block dropping anchor frame, and the calculation of the block dropping length and depth is performed, so that the related point cloud data volume is small, the calculation amount is small, the detection result can be obtained quickly during on-site detection, and the on-site stop line detection time is reduced.
Further, the severity of the falling block is automatically judged based on the length and the depth of the falling block area and a preset severity standard, and the two-dimensional image, the point cloud data, the calculated size, the severity and other information corresponding to the falling block area are stored correspondingly, so that a detector can check and confirm the identified rail face falling block conveniently or analyze the identified rail face falling block further, and the falling block area with a serious injury level can be screened out for preferential treatment, so that the efficiency and the accuracy of rail face falling block detection can be further improved, the automation degree is high, and the burden of the detector can be reduced.
The above examples are only for illustrating the specific embodiments of the present invention, and the present invention is not limited to the description scope of the above examples.
In the above embodiments, the point rail at the switch and the wing rails on both sides thereof are specifically described as examples, and it will be understood that the method can be applied to other rail members at the switch as well.
In the above embodiment, the starting point of the switch rail member is identified from the multi-frame profile data based on the profile features of the switch rail member, so that the identified rail surface drop block can be corresponding to the switch rail member, in an alternative scheme, the starting point or type of the switch rail member can be identified from the camera two-dimensional image based on the structural features of the switch rail member in the image, for example, the two-dimensional image is input into a trained switch rail member classification model (neural network model), the model outputs the type and probability score of the rail member in the two-dimensional image, for example, the point starting point is identified from the two-dimensional image, and the mileage information corresponding to the frame two-dimensional image is taken as the mileage information of the point starting point.
In the above embodiment, the minor injury, major injury grades and corresponding threshold information of the rail surface falling blocks are set, and in alternative schemes, more grades and corresponding threshold information can be set according to actual detection requirements.
In the above embodiment, for convenience of description, the steps of two-dimensional image extraction, rail surface block drop recognition and the like are performed after the turnout data acquisition step is completed, and it can be understood that the steps of two-dimensional image extraction, rail surface block drop recognition and the like can be performed on the acquired data in the data acquisition process under the condition of sufficient calculation force, so that the real-time performance of detection is further improved.

Claims (10)

1. The utility model provides a switch rail spare rail face fall piece detection method for detect the rail face fall piece of switch rail spare, its characterized in that detects based on the profile data that the camera gathered, line laser sensor gathered and corresponding mileage information, this method includes:
the method comprises the following steps of: identifying the starting point of the turnout rail piece from the two-dimensional image or the multi-frame profile data based on the profile characteristics of the starting point of the turnout rail piece and obtaining the corresponding mileage information;
a two-dimensional image extraction step: extracting a plurality of frames of two-dimensional images corresponding to the turnout rail piece as images to be detected based on the mileage information of the starting point of the turnout rail piece and the preset extension length of the turnout rail piece;
Rail surface block falling identification: inputting the images to be detected into a rail surface block falling identification model in sequence, and outputting block falling position information of rail surface block falling by the model, wherein the rail surface block falling identification model is a neural network model after training;
a step of generating a block dropping effective point cloud: extracting a plurality of frames corresponding to the rail surface block from the plurality of frames of profile data based on the block-falling position information and calibration parameters of the camera and the line laser sensor, and stacking the frames to generate a block-falling effective point cloud;
calculating the characteristic size of the falling block: and calculating the characteristic size of the track surface block based on the block falling position information and the block falling effective point cloud.
2. The switch rail piece rail face block detection method according to claim 1, wherein:
wherein the turnout rail piece comprises a center rail and wing rails arranged at two sides of the center rail,
the switch rail piece identification step comprises the following steps:
based on the mileage information, carrying out profile matching identification on the collected profile data frame by frame to identify the starting point of the section of the heart rail, and obtaining the corresponding mileage information;
acquiring the profile data of a preset frame number connected after the starting point of the section of the heart rail based on mileage information of the starting point of the section of the heart rail, and stacking the profile data to generate a point cloud;
Based on the point cloud, identifying a theoretical starting point of the heart rail in a fitting plane mode, and obtaining corresponding mileage information;
extracting profile data in a preset position range before the theoretical starting point of the heart rail based on the mileage information of the theoretical starting point of the heart rail;
performing profile matching identification of the wing rail starting point on the extracted profile data frame by frame, identifying the wing rail starting point, obtaining the corresponding mileage information,
in the two-dimensional image extraction step, a plurality of frames of the two-dimensional image covering the center rail and the wing rail are extracted based on the mileage information of the center rail theoretical starting point, a predetermined extension length of the center rail, the mileage information of the wing rail starting point, and a predetermined extension length of the wing rail.
3. The switch rail piece rail face block detection method according to claim 2, wherein:
wherein, the identifying the theoretical starting point of the heart rail by a fitting plane mode comprises the following steps:
fitting a plane between the starting point of the section of the heart rail and the theoretical starting point of the heart rail through a RANSAC algorithm based on the point cloud;
fitting an inclined plane after the theoretical starting point of the heart rail through a RANSAC algorithm based on the point cloud;
And calculating coordinates of points on an intersection line of the fitted plane and the fitted inclined plane, and obtaining corresponding mileage information based on the coordinates, or respectively calculating a central line of the fitted plane and a central line of the fitted inclined plane, calculating coordinates of an intersection point of the two central lines, and obtaining corresponding mileage information based on the coordinates.
4. The switch rail piece rail face block detection method according to claim 2, wherein:
wherein, the profile matching of the starting point of the section of the heart rail comprises:
judging whether a profile exists at a preset position in the profile data based on a preset starting point matching template of the section of the heart rail, intercepting the profile according to a preset profile intercepting algorithm when the profile is judged to exist, obtaining the profile of the upper end face of the heart rail, judging whether the intercepted profile is a horizontal straight line segment with a preset length, identifying the starting point of the section of the heart rail when the intercepted profile is judged to be horizontal, taking mileage information corresponding to the profile data of the frame as mileage information of the starting point of the section of the heart rail,
the profile matching of the wing rail starting point comprises the following steps:
judging whether a profile exists at a preset position in the profile data based on a preset profile starting point matching template, intercepting the profile according to a preset profile intercepting algorithm when the profile exists, obtaining the profile of the rail head parts of the wing rails at two sides, matching the intercepted profile with a preset standard wing rail starting point profile, calculating the similarity, judging that the wing rail starting point is identified when the similarity reaches a preset similarity threshold, and taking the mileage information corresponding to the profile data of the frame as the mileage information of the wing rail starting point.
5. The switch rail piece rail face block detection method according to claim 1, wherein:
wherein the rail face chunking region identification model is a modified YOLOv5 neural network model, comprising:
the back box module is used for extracting the characteristics of the two-dimensional image, and an up-sampling layer is additionally arranged behind the back box module;
the PAN+Bi-FPN module fuses the extracted features to obtain fusion features; and
and the Head module is used for identifying a block dropping area and a rail surface area in the two-dimensional image based on the fusion characteristics and outputting the pixel coordinates and the first confidence coefficient of the corresponding block dropping anchor frame and the pixel coordinates and the second confidence coefficient of the rail surface anchor frame.
6. The method for detecting the rail surface block drop of the turnout rail piece according to claim 5, wherein the method comprises the following steps:
wherein, the step of identifying the rail surface falling blocks further comprises the following steps:
and removing the block dropping anchor frame with the first confidence coefficient lower than a preset confidence coefficient threshold value and the block dropping anchor frame positioned outside the rail surface anchor frame from the identified block dropping anchor frame.
7. The method for detecting the rail surface block drop of the turnout rail piece according to claim 5, wherein the method comprises the following steps:
The step of generating the effective point cloud comprises the following steps:
extracting corresponding multiframes from the profile data based on the mileage information corresponding to the image to be detected, and stacking the multiframes to generate a track point cloud;
cutting the track part point cloud based on the distance between the line laser sensor and the track surface of the turnout track part to obtain a track surface point cloud;
converting pixel coordinates of the block dropping anchor frame into camera coordinates based on camera internal parameters of the camera, and converting the camera coordinates of the block dropping anchor frame into world coordinates based on a calibration relation between the camera and the line laser sensor;
and extracting a block falling effective point cloud corresponding to the block falling of the rail surface from the rail surface point cloud based on the world coordinates of the block falling anchor frame.
8. The method for detecting the rail surface block drop of the turnout rail piece according to claim 5, wherein the method comprises the following steps:
the step of calculating the characteristic size of the block dropping comprises the following steps:
calculating the length and the width of the block falling anchor frame based on the world coordinates of the block falling anchor frame, and taking the maximum value as the length of the rail surface block falling;
fitting a rail surface plane through a RANSAC algorithm based on the block falling effective point cloud;
calculating the inclination angle of the fitted rail surface plane, and horizontally correcting the effective point cloud of the block falling based on the inclination angle;
Traversing in the horizontal corrected effective point cloud of the block falling, finding out the maximum value and the minimum value of Z-axis coordinate values of traversed points, and calculating the difference value between the maximum value and the minimum value as the depth of the block falling of the rail surface.
9. The switch rail member rail face drop block detection method of claim 1, further comprising, after the drop block feature size calculation step:
a block-off severity determination step of determining a severity of the rail-face block-off based on the characteristic size and a predetermined rail-face block-off severity criterion,
wherein the rail face drop severity criteria includes a plurality of severity levels and corresponding threshold information.
10. The utility model provides a switch rail spare rail face falls piece check out test set for detect the rail face of switch rail spare and fall the piece, its characterized in that, detect based on the profile data and the mileage information that the camera gathered of two-dimensional image, line laser sensor gathered, this equipment includes:
the turnout rail piece identification part is used for identifying the starting point of the turnout rail piece from a plurality of frames of two-dimensional images or a plurality of frames of profile data based on the profile characteristics of the starting point of the turnout rail piece and obtaining corresponding mileage information;
A two-dimensional image extraction unit configured to extract a plurality of frames of the two-dimensional image corresponding to the switch rail piece based on the mileage information of the start point of the switch rail piece and a predetermined extension length of the switch rail piece;
the rail surface block falling recognition part is used for sequentially inputting the extracted two-dimensional images into a rail surface block falling recognition model, and outputting block falling position information of rail surface blocks, wherein the rail surface block falling recognition model is a neural network model after training;
the block-falling effective point cloud generating part is used for extracting a plurality of frames corresponding to the rail surface block-falling from the plurality of frames of profile data based on the block-falling position information and calibration parameters of the camera and the line laser sensor, and stacking the frames to generate a block-falling effective point cloud; and
and the block falling characteristic size calculating part is used for calculating the characteristic size of the rail surface block falling based on the block falling effective point cloud.
CN202310570814.8A 2023-05-19 2023-05-19 Method and equipment for detecting track surface block drop of turnout rail piece Pending CN116934680A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310570814.8A CN116934680A (en) 2023-05-19 2023-05-19 Method and equipment for detecting track surface block drop of turnout rail piece

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310570814.8A CN116934680A (en) 2023-05-19 2023-05-19 Method and equipment for detecting track surface block drop of turnout rail piece

Publications (1)

Publication Number Publication Date
CN116934680A true CN116934680A (en) 2023-10-24

Family

ID=88393224

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310570814.8A Pending CN116934680A (en) 2023-05-19 2023-05-19 Method and equipment for detecting track surface block drop of turnout rail piece

Country Status (1)

Country Link
CN (1) CN116934680A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117232435A (en) * 2023-11-14 2023-12-15 北京科技大学 Device and method for measuring abrasion value and reduction value of switch tongue

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117232435A (en) * 2023-11-14 2023-12-15 北京科技大学 Device and method for measuring abrasion value and reduction value of switch tongue
CN117232435B (en) * 2023-11-14 2024-01-30 北京科技大学 Device and method for measuring abrasion value and reduction value of switch tongue

Similar Documents

Publication Publication Date Title
CN111855664B (en) Adjustable three-dimensional tunnel defect detection system
US10480939B2 (en) High speed stereoscopic pavement surface scanning system and method
CN105699381B (en) A kind of railway rail clip detection device based on machine vision
CN105158257B (en) Slide plate measurement method and device
US20040095585A1 (en) Wheel profile inspection apparatus and method
CN107743577B (en) Method and apparatus for detecting vibration information of electric railway vehicle
CN116934680A (en) Method and equipment for detecting track surface block drop of turnout rail piece
CN1920896A (en) Real-time image monitoring and genuine-fake identification system for vehicle with automatic color regulating and light filling license plate
CN112950532B (en) Train pantograph state detection method
CN105571508A (en) Deformation detecting method and deformation detecting system for pantograph of overhead contact system
CA3163192A1 (en) Method for in-situ and real-time collection and processing of geometric parameters of railway lines
CN116934682A (en) Method and equipment for detecting light band on rail surface of steel rail
CN103481842B (en) A kind of changing method of moving vehicles detection and tracking pattern
JPH0879904A (en) Abnormality detector for pantograph
CN101680748A (en) Equipment for measuring abrasion of trolley wire by image processing
CN108280894A (en) Not parking method for inspecting for power equipment
CA3066752A1 (en) High speed stereoscopic pavement surface scanning system and method
CN112964195A (en) Power supply rail geometric parameter comprehensive detection method and system based on laser triangulation
CN110306413A (en) A kind of high-precision track measuring device and measuring method based on monocular
CN116923491A (en) Switch rail abrasion detection method and device
CN111640155B (en) Pantograph head inclination angle measurement method and system based on 3D imaging
JP3629568B2 (en) Overhead wire inspection method and overhead wire inspection apparatus
CN116989881A (en) Contact line vibration measurement system and method based on line laser target identification
CN116934681A (en) Method and equipment for detecting falling blocks of rail surface of steel rail
CN116519787A (en) Third rail power receiving boot detection system and method

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination