WO2023019709A1 - 基于车载移动激光点云的接触网导高与拉出值自动检测方法 - Google Patents

基于车载移动激光点云的接触网导高与拉出值自动检测方法 Download PDF

Info

Publication number
WO2023019709A1
WO2023019709A1 PCT/CN2021/124110 CN2021124110W WO2023019709A1 WO 2023019709 A1 WO2023019709 A1 WO 2023019709A1 CN 2021124110 W CN2021124110 W CN 2021124110W WO 2023019709 A1 WO2023019709 A1 WO 2023019709A1
Authority
WO
WIPO (PCT)
Prior art keywords
point cloud
line
rail
point
catenary
Prior art date
Application number
PCT/CN2021/124110
Other languages
English (en)
French (fr)
Inventor
许磊
豆孝磊
张冠军
牟春霖
巩健
谢春喜
刘成
杨元维
梁永
石德斌
董延喜
赵梦杰
Original Assignee
中国铁路设计集团有限公司
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 中国铁路设计集团有限公司 filed Critical 中国铁路设计集团有限公司
Publication of WO2023019709A1 publication Critical patent/WO2023019709A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00Image enhancement or restoration
    • G06T5/70Denoising; Smoothing
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/21Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G06F18/2135Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/10Segmentation; Edge detection
    • G06T7/11Region-based segmentation
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/20Analysis of motion
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/30Determination of transform parameters for the alignment of images, i.e. image registration
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/10Image acquisition modality
    • G06T2207/10028Range image; Depth image; 3D point clouds
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/30Subject of image; Context of image processing
    • G06T2207/30241Trajectory

Definitions

  • the invention relates to the field of rapid comprehensive detection of rail transit, in particular to an automatic detection method for catenary height and pull-out value based on vehicle-mounted mobile laser point cloud.
  • the railway is an important national infrastructure, the main artery of the national economy, and plays a key role in the comprehensive transportation system.
  • 1975 the entire Baocheng Railway was electrified and became the first electrified railway in my country. Since then, my country's electrified railways have developed rapidly. As of the end of 2020, the total mileage of my country's railway operations has exceeded 146,300 kilometers, of which high-speed railways have reached 38,000 kilometers, 106,000 kilometers of electrified railways, and an electrification rate of 72.8%.
  • catenary plays an important role in ensuring the safety of railway transportation, improving transportation efficiency and reducing transportation energy consumption.
  • the catenary is placed in the open environment, which is prone to aging and geometric changes in harsh environments. During the operation of the train, due to irregular operation, abnormal relationship between pantograph and catenary, electrical corrosion and other reasons, the catenary will be deformed or even disconnected. When the catenary fails, the operation of the entire railway line will be affected because there is no backup line, resulting in huge economic losses and adverse social impacts.
  • the geometric parameters of the catenary mainly include the guide height and pull-out value of the contact wire.
  • the guide height refers to the vertical distance from the bottom surface of the contact wire to the rail surface
  • the pull-out value refers to the offset of the contact wire at the positioning point to the center line of the pantograph.
  • Non-contact detection methods mainly include image detection and laser detection.
  • Image detection is to install a CCD camera on the roof of the detection vehicle to obtain images, and use image recognition technology to detect the guide height and pull-out value of the catenary. Due to the different measurement standards, the image detection method usually needs to compensate the vibration of the vehicle body, and the calculation process is complicated. At the same time, image detection methods are easily affected by factors such as weather and light, resulting in unstable detection quality, difficult target recognition, and low accuracy.
  • the traditional portable laser measuring equipment is generally used in the work areas of the power supply section and the maintenance section.
  • the static point measurement of the conduction height and pull-out value of the contact line is carried out at the skylight point, and the laser range finder is used at each location of the contact line.
  • Manually aiming at and capturing the contact line for parameter sampling measurement is labor-intensive, low in efficiency, low in safety factor, and has a single detection function. Therefore, it is urgent to develop a new technology for detection of catenary height and pull-out value with high detection efficiency, high detection accuracy, high degree of automation, and contactless, so as to ensure the normal and stable operation of the catenary system.
  • Vehicle-mounted mobile laser scanning technology refers to the integration of global satellite navigation system (Global Navigation Satellite System, GNSS), inertial measurement unit (Inertial Measure Unit, IMU), laser scanner, digital camera, digital video camera and other sensors on the mobile carrier.
  • GNSS Global Navigation Satellite System
  • IMU Inertial Measure Unit
  • Various types of sensors automatically collect various positions, attitudes, images and laser scanning data while moving, and realize non-contact spatial geographic information collection, processing and storage through unified geographic reference and data collection synchronization technology.
  • the integrated 3D laser scanning system is mounted on the rail car, and through the movement of the carrier, it quickly collects massive point cloud and image data within tens to hundreds of meters on both sides of the rail transit.
  • vehicle-mounted mobile scanning has an efficient and flexible data acquisition method, and is more and more used in engineering practice.
  • Vehicle-mounted mobile scanning technology is one of the current research hotspots. In recent years, more and more researchers have used the vehicle-mounted mobile laser point cloud to realize the research of catenary conduction height and pull-out value detection.
  • Zhou Jingsong of Southwest Jiaotong University discussed the improved random sampling consistent RANSAC algorithm based on the characteristics of catenary point cloud in his master's thesis "Research on Catenary Geometric Parameter Detection Based on 3D Point Cloud Data", and extracted the contact line point cloud data; and then rotated by space
  • the translation information obtains the transformation matrix from the camera coordinate system to the world coordinate system, and thus calculates the conduction height and pull-out value of the contact line.
  • This method needs to transform the measurement reference from the camera center to the track plane, and the measurement results are easily affected by factors such as vehicle body vibration and wheel-rail travel.
  • Cida Patent Publication No. CN103852011A discloses a lidar-based geometric parameter analysis method for railway catenary. This method uses information such as the minimum conductance value of the contact line point cloud and the extreme value of the pull-out value at the column of the catenary. Detection of catenary geometric parameters. However, this method has the following disadvantages: (1) This method fails when the pull-out value at the catenary column is not an extreme value; (2) This method does not detect the guide height and pull-out value at the suspension point of the hanging string.
  • the present invention provides an automatic detection method for catenary conduction height and pull-out value based on vehicle-mounted mobile laser point cloud.
  • the method has the characteristics of high efficiency, high accuracy and high degree of automation.
  • An automatic detection method for catenary guide height and pull-out value based on vehicle-mounted mobile laser point cloud comprising the following steps:
  • the vehicle-mounted mobile laser scanning system to quickly collect point cloud data of infrastructure and surrounding environments along the rail transit; based on the laser point cloud collected along the line, by setting the range value of the laser scanning emission angle corresponding to the track and catenary point cloud, adopt The angle threshold segmentation algorithm is used to segment the laser point cloud including the track and catenary from the full-section laser point cloud of rail transit; for the noise points in the segmented track and catenary laser point cloud, the statistical filtering algorithm is used to eliminate them;
  • the improved least squares fitting algorithm of the space straight line it is possible to perform piecewise linear fitting on the trajectory line to obtain equally spaced trajectory line segments, based on the relatively stable spatial position between the piecewise fitted trajectory line segment and the corresponding orbit relationship, using the cuboid segmentation algorithm to extract the rail point cloud from the track laser point cloud preprocessed by S1; generating the standard rail model point cloud according to the rail standard cross-sectional view, using the principal component analysis algorithm and the generalized iterative closest point algorithm to realize the rail point cloud and Accurate registration of the point cloud of the rail model, so as to obtain the spatial geometric parameters of the rail, extract the center line of the rail top, calculate the midpoint of the line, and then fit the center line of the line; use the extracted center line of the rail top to update the relationship between the track line and the point cloud of the rail Spatial position parameters, for the extraction and registration of the next rail point cloud; iteratively repeat the above steps to complete the extraction and registration of all rail point clouds;
  • Contact line and suspension point extraction According to the relatively stable spatial position relationship between the trajectory line and the catenary structure, based on the equally spaced trajectory line segments obtained by piecewise linear fitting in S2, the catenary laser obtained from S1 using the cylinder segmentation algorithm Extract the point cloud containing contact lines and suspension points from the point cloud; use the straight line random sampling consensus algorithm to fit the 3D line model to the point cloud containing contact lines and suspension points, and set random Sampling the parameters of the consistency algorithm to extract the point cloud segment of the contact wire and automatically fit the 3D straight line model; in the area containing only the contact line, the dimensional characteristics of the point cloud are linear; at the suspension point, due to the suspension string and positioning The device is connected with the contact line, and the dimensional characteristics of the point cloud are nonlinear. Using the dimensional analysis method, analyze the dimensional characteristics of the neighborhood of the target point, and set the linear threshold to extract the suspension point of the contact line;
  • the contact line guide height and pull-out value are automatically detected.
  • step S1 the point cloud data collection and preprocessing include the following steps:
  • the laser scanner scans in a circle perpendicular to the direction of the line or 45 degrees to the direction of the line, and obtains the laser point cloud within the length of the measured rail transit and a certain distance on both sides of the track;
  • Use the track line of the vehicle-mounted mobile laser scanning system to solve the scanned data obtain the laser point cloud under the unified geographical space coordinate system of the whole line, and export the laser point cloud under the unified geographical space coordinate system of the whole line into a general Data Format;
  • the laser point cloud of a certain angle range is segmented; according to the installation of the scanner Height, installation angle, and geometric scale of the segmented area, through spatial geometric analysis, use trigonometric function formulas to calculate the angle range of the segmented area, or directly select the boundary point of the track or catenary structure on the laser point cloud, and view the boundary point Scanning angle value; then set a floating value ⁇ of the scanning angle value, according to the formula (1), traverse all laser point clouds, and segment the track and catenary point cloud:
  • Segment RailTrack Segment Catenary
  • Segment Other track point cloud, catenary point cloud, and other point clouds in turn;
  • ⁇ min , ⁇ max the minimum scanning angle and the maximum scanning angle corresponding to the orbit, respectively, in degrees;
  • ⁇ min , ⁇ max respectively the minimum scanning angle and maximum scanning angle corresponding to the catenary structure, the unit is degree;
  • the angle division floating value ⁇ is 3 degrees
  • Statistical filtering algorithm is used to remove a small amount of noise points in the point cloud data acquired by the 3D laser scanning system.
  • the general data format is LAS format
  • the rotation angle of the scanner ranges from -180° to 180°
  • the scanning start direction is directly below the scanner
  • 180° is rotated to directly above the scanner
  • the trajectory The angle on the left side of the direction of the line is negative, and the angle on the right side is positive.
  • the certain distance range on both sides of the track is the width range of 100 meters on the left and right sides of the line.
  • step (2) if the vehicle-mounted mobile laser scanning system contains multiple scanners, the laser point cloud is divided according to the angle of each scanner, and then the laser point clouds divided by multiple scanners are merged.
  • step S2 the segmentation and registration steps of the rail point cloud are as follows:
  • Segmented linear fitting of the trajectory line firstly establish a multi-dimensional spatial index structure of the rail transit laser point cloud; set a fixed step (Fixed Depth), and segmentally linearly fit the trajectory line to obtain equidistant polylines;
  • a gridded three-dimensional standard rail model is established and uniformly sampled to form a point cloud of the standard rail model; the classic ICP (Iterative Closest Point) registration algorithm and its improved algorithm are both right
  • ICP Intelligent Closest Point
  • the requirements for the initial registration conditions are relatively strict, and the overlapping degree of the point clouds to be registered needs to be high, otherwise it is easy to fall into the local optimal trap, so it is necessary to use the coarse registration algorithm to align the point clouds to be registered.
  • the two sets of point clouds have basically coincided, which provides a good initial condition for accurate registration.
  • the GICP algorithm is used to realize the fine registration of rail point clouds.
  • This algorithm combines "point-to-point” ICP and "point-to-surface” ICP into the probability framework model, and uses the covariance matrix of the point cloud to construct the registration cost function. Improved speed and robustness of classical ICP registration.
  • the rough registration algorithm is used to roughly align the rail point cloud to be registered with the standard rail model point cloud, and then the GICP algorithm is used to achieve fine registration of the rail point cloud.
  • the finely aligned rail model point cloud to calculate the spatial geometric parameters of the rail, and then calculate the midpoint of the line based on the reference track, and connect the midpoints extracted by these sections in sequence to form the centerline of the line;
  • Steps 1) to 4) are performed in a loop, and the new geometric parameters are used for the extraction and registration of the point cloud of the next section of rail until the extraction and registration of all rail point clouds are completed.
  • the multi-dimensional spatial index structure of the rail transit laser point cloud is a KD tree.
  • the three-dimensional geometric model of the rail after the registration is used to calculate the geometric parameters of the rail to obtain gauge and direction information.
  • the reference rail is determined according to the definition of the reference rail, based on the reference rail, Horizontally offset 1/2 standard gauge to another rail (the standard gauge defined in China is 1435mm), calculate the midpoint of the line in sections, and then connect the midpoints of the lines extracted by these sections in sequence to form the centerline of the line.
  • the formed line center line is then smoothed to obtain a relatively smooth line center line.
  • step S3 the steps of automatic extraction of contact line and suspension point cloud are as follows:
  • the contact line is laid out in a "zigzag" shape along the line direction, and the pull-out value is the maximum value at the locator.
  • TG/GD 124-2015 “Rules for Operation and Maintenance of High-speed Railway Catenary”
  • TG/GD116-2017 “Rules for Operation and Maintenance of Normal-speed Railway Catenary”
  • the limit value of the pull-out value of normal-speed railway and high-speed railway contact line is 450mm.
  • the radius r of the cylinder can be calculated according to formula (2):
  • L is the distance between catenary columns, generally 50m;
  • the radius r is 0.1m
  • PCA Principal component analysis
  • the direction can be judged according to the eigenvector corresponding to ⁇ 1 , the point perpendicular to the ground is the hanging string suspension point, and the approximately parallel point is the positioning suspension point.
  • the linear eigenvalue threshold L ⁇ is 0.9.
  • step S4 the detection method of catenary guide height and pull-out value is as follows:
  • m the horizontal distance between the contact line of the positioning point and the center line of the line
  • the method for finding the rail surface at the corresponding position is a distance discrimination method
  • the outer rail superelevation is calculated according to the rail space geometric parameters obtained in step S2 in claim 1 .
  • the invention provides an automatic detection method for catenary height and pull-out value based on railway three-dimensional laser point cloud data acquired by a light-duty vehicle-mounted mobile measurement system.
  • the method of the invention is based on a laser scanning mode, is not affected by factors such as light, has strong environmental adaptability, and has stable and reliable detection results.
  • the laser scanner is mounted on the roof of the rail car, and only the point cloud of the contact line is obtained.
  • the measurement reference is converted from the center of the scanner to the track plane, and the calculation The accuracy is easily affected by factors such as car body vibration and wheel-rail travel.
  • the present invention obtains the geometric parameters of the track, the contact line and the suspension point position based on the full-section point cloud of the railway line, and directly calculates the guide height and pull-out value of the contact line without converting the measurement reference, thereby effectively improving the measurement accuracy.
  • the existing rail point cloud segmentation algorithm uses the characteristics of the maximum local elevation value of the rail surface or the direction change continuity of the rail surface point cloud, and uses the statistical calculation method of the elevation histogram or the Kalman filter method to classify the rail point cloud , the existing method only extracts the rail point cloud, does not obtain the precise rail space geometric parameters of the rail and the rail point cloud extraction process is complicated.
  • the present invention is based on the stable positional relationship between the trajectory line of the mobile scan and the rail, uses the cuboid segmentation algorithm to realize the point cloud extraction, uses PCA and GICP to realize the precise registration of the rail point cloud and the rail model point cloud, and obtains the precise position of the rail. Space geometry parameters.
  • the existing contact line extraction method extracts the contact line laser point cloud based on all the laser point clouds of the catenary.
  • the invention utilizes a cylinder segmentation algorithm to segment and obtain contact line point cloud and suspension point cloud, and then perform subsequent contact line extraction and suspension point extraction, thereby effectively improving calculation efficiency.
  • the existing catenary conduction height and pull-out value detection method based on laser point cloud does not locate the suspension point position, and the conduction height and pull-out value of the suspension point position are the concerns of the electrical department.
  • the present invention uses a dimensional analysis method to determine the position of the suspension point of the contact line, and identifies the type of the suspension point according to the Z value component of the eigenvector corresponding to the first feature of the covariance matrix of the neighborhood point cloud.
  • the present invention not only improves the detection accuracy, but also can quickly realize the detection of contact line conduction height and pull-out value.
  • the vehicle-mounted mobile laser scanning method is used to avoid the inspection personnel from going online and reduce the risk of safe operation.
  • the invention can be used as a new measuring method for catenary height and pull-out value, and can serve for the static acceptance of newly-built railway lines and the normalized detection of existing railway lines.
  • Fig. 1 is the flow chart of catenary conduction height and pull-out value detection method of the present invention
  • Fig. 2 is the track and catenary point cloud after the method preprocessing of the present invention
  • Fig. 3 is a schematic diagram of rail point cloud segmentation in the present invention.
  • Fig. 4 is the segmentation result of rail point cloud among the present invention.
  • Fig. 5 is a registration flow chart in the present invention.
  • Figure 6a is a standard rail model
  • Figure 6b is the standard rail point cloud
  • Fig. 8 is track detection result
  • Fig. 9 is a schematic diagram of contact line extraction
  • Figure 10 is a schematic diagram of the dimensional analysis method
  • Fig. 11 is the detection result of contact line and suspension point
  • Fig. 12 is a state diagram of acquiring point cloud data using the light railway mobile measurement system
  • Figure 13 is a schematic diagram of the high-speed rail line from Yancheng to Nantong;
  • Figure 14 is the laser point cloud of Yantong high-speed rail obtained in the present invention.
  • Fig. 15 is a comparison chart of the detection results of the guide height and pull-out value and the detection data of the DJJ-8 measuring instrument in the present invention.
  • the catenary guide height and pull-out value automatic detection method of the present invention includes: point cloud data collection and preprocessing; rail point cloud segmentation and registration; contact line and suspension point point cloud extraction and guide height and pull-out value calculation. As shown in Figure 1, the above steps are as follows:
  • data acquisition and preprocessing use the vehicle-mounted mobile laser scanning system to quickly collect data on the infrastructure and surrounding environment along the rail transit; based on the laser point cloud collected along the line, set the laser corresponding to the track point cloud and catenary point cloud Scan the range value of the launch angle, and use the angle threshold segmentation algorithm to segment the track and catenary laser point clouds from all the laser point clouds along the rail transit line, and then use the statistical filtering algorithm to remove the point cloud noise.
  • the specific method is as follows:
  • the laser scanner of the vehicle-mounted mobile laser scanning system adopts a mechanical rotating reflective prism method to realize 360°circumferential section scanning, and then moves on the track through the scanner-mounted platform to realize the scanning of the rail transit range and a certain distance range on both sides.
  • the scanner will record the rotation angle information of the scanning head during the data collection process.
  • the angle value of each step rotation of the scanning head is fixed, and the starting point of the angle of each circular scan is also fixed, generally starting from the bottom of the scanner. And end at this point, complete the circular scan. Since the calculation of catenary height and pull-out value only involves the point cloud of the track and catenary range, in order to reduce the influence of the background point cloud and other structural point clouds, improve the speed and accuracy of point cloud segmentation calculation, firstly, according to the track and catenary The scanning angle corresponding to the structure is segmented to obtain orbital and catenary point clouds.
  • the moving scanning trajectory line uses the moving scanning trajectory line to solve the scanning data, and obtain the laser point cloud under the unified coordinate system of the whole line.
  • the range of scanner rotation angle recorded by laser point cloud data in LAS format is -180° ⁇ 180°, that is, the scanner is directly below the scanner. When it is rotated to the top of the scanner, the scanning angle is 180°, the left angle of the trajectory line is a negative value, and the right angle is a positive value.
  • the laser point cloud of a certain angle range is segmented; according to the scanner's Parameters such as installation height, installation angle, and geometric scale of the segmented area can be calculated using trigonometric function formulas to obtain the angle range of the segmented area through spatial geometric relations, or directly select the boundary point of the track or catenary structure on the laser point cloud to view Select the scanning angle value of the point; consider the change of the scanner emission angle caused by the shaking of the vehicle during the vehicle scanning process, set a floating value ⁇ of a certain angle value (such as an angle of 3 degrees), and traverse all laser points according to the formula (1) Cloud, segment out orbital and catenary point cloud data:
  • Segment RailTrack Segment Catenary
  • Segment Other track point cloud, catenary point cloud, and other point clouds in turn;
  • ⁇ min , ⁇ max the minimum scanner angle and the maximum scanner angle corresponding to the orbit, respectively, in degrees;
  • ⁇ min , ⁇ max respectively the minimum scanner angle and the maximum scanner angle corresponding to the catenary structure, the unit is degree
  • the vehicle-mounted mobile laser scanning system contains multiple scanners, it is necessary to segment the laser point cloud according to the angle of each scanner, and then merge the laser point clouds segmented by multiple scanners.
  • the collected point cloud will contain a small amount of noise points due to various factors such as instrument vibration, rough surface of the measured object, mirror reflection, and obstruction. These few noise points far away from the main structure will affect the extraction of rail point cloud and contact line point cloud, and statistical filtering algorithm is used to remove these noise points.
  • the laser point cloud segmentation is carried out through the angle segmentation threshold in the above step (2), and the background point cloud data and other structural point clouds are eliminated, and only the track and catenary point clouds are retained.
  • the statistical filtering algorithm in the above step (3) is used to remove the point cloud noise, and the preprocessing results are shown in Figure 2.
  • the three-dimensional space structure of the railway and the scanning operation of the vehicle-mounted mobile laser scanning system have their own characteristics: (1) In the direction along the line, the two rails are basically parallel, and the distance between the inner sides of the two rails is generally the same as the standard gauge (1435mm) Very close; (2) The wheel of the vehicle-mounted mobile laser scanning system is stuck on the track and runs along the track, the relative position between the scanner and the two rails changes very little, and the scanning track line shape is consistent with the line shape of the railway line; (3 ) The rail is fixed on the sleeper or the ballastless track slab, and the rail is higher than the plane of the sleeper, ballast and track slab.
  • the mobile laser scanning trajectory file contains the spatial position and attitude data of the center of the laser scanning system.
  • the spatial position data of the trajectory line is the spatial position point of the scanner under the reference coordinate system, and these spatial position points are connected in turn to form a mobile scanning trajectory line.
  • the spatial position of the track line relative to the left and right rails will be fixed (straight line) or continuously change within a small range (curved line).
  • the invention utilizes the characteristics that the track line and the spatial position of the rail are relatively fixed and the entire line changes continuously, and combines the inherent structural information of the rail to quickly realize the segmentation of the laser point cloud of the rail.
  • the present invention first establishes a standard rail model point cloud (such as 60kg/m rail, which is relatively Commonly used rail types), and then use the combination of coarse registration and fine registration to accurately register the segmented rail point cloud and rail model point cloud, and obtain the spatial geometric parameters of the rail according to the registered rail model point cloud .
  • a standard rail model point cloud such as 60kg/m rail, which is relatively Commonly used rail types
  • a segmented method is used to extract the rail point cloud.
  • the least squares fitting algorithm of space straight line is improved, and the trajectory line is fitted with a fixed step length to obtain a polyline with a distance of D.
  • the principle of the improved least squares fitting algorithm for straight lines in space is as follows:
  • the trajectory data set has a total of N points
  • the coordinates of the starting point are P i (X i , Y i , Z i )
  • the coordinates of the jth point are P j (X j , Y j , Z j )
  • two The Euclidean distance between points is d ij .
  • the space straight line can be regarded as the intersection of two planes, and the two plane equations are fitted, and the space straight line can be obtained by further fitting.
  • the sum of squares of the residuals of the two planes is
  • the elevation of the top surface of the cuboid bounding box is determined to be H max + ⁇ , and the elevation of the bottom surface is H min -H rail - ⁇ . Cycle through the track point cloud and segment the rail point cloud in the cuboid bounding box. The result of rail point cloud segmentation is shown in Figure 4.
  • the present invention uses PCA and GICP registration to accurately register the rail point cloud and the rail model point cloud, thereby obtaining the precise spatial geometric parameters of the rail.
  • the registration flow chart is shown in Figure 5, specifically as follows:
  • a gridded three-dimensional rail model is established and uniformly sampled to form a point cloud of the standard rail model, as shown in Figures 6a and 6b.
  • the classic ICP (Iterative Closest Point) registration algorithm and its improved algorithm have relatively strict requirements on the initial registration conditions, requiring a high degree of overlap of the point clouds to be registered, otherwise it is easy to fall into the trap of local optimum, so it is necessary to use rough matching first.
  • the quasi-algorithm roughly aligns the point clouds to be registered. Since the rail length of the railway rail is much greater than the height and width of the rail, the rail point cloud space presents a linear distribution.
  • the present invention applies the principal component analysis (PCA) algorithm to the rough registration of the rail point cloud.
  • PCA principal component analysis
  • the eigenvalues and eigenvectors of the covariance matrices COV S and COV G were calculated by the singular value decomposition algorithm (SVD).
  • the rigid transformation matrix Trans of PCA registration is composed of rotation matrix R P and translation matrix T P.
  • U G and U S represent the matrix formed by the eigenvectors of the covariance matrix.
  • the present invention uses the GICP algorithm for fine registration, which combines "point-to-point” ICP and "point-to-surface” ICP into the probability frame model, uses the covariance matrix of the point cloud to construct the registration cost function, and improves the The speed and robustness of classic ICP registration, the principle is as follows:
  • Two sets of point clouds A and B can be generated by Gaussian distribution:
  • C i is the covariance matrix of each point.
  • the transformation matrix calculated by PCA and GICP is used to realize the precise registration of the rail point cloud and the rail model point cloud.
  • the registration result is shown in Figure 7.
  • move to another rail Horizontally offset 1/2 standard gauge the standard gauge defined by China is 1435mm
  • calculate the midpoint of the line in sections and then connect the midpoints of the lines extracted by these sections in sequence to form the centerline of the line.
  • the The formed line center line is then smoothed to obtain a relatively smooth line center line, as shown in FIG. 8 .
  • Steps 1) to 4) are performed in a loop, and the new geometric parameters are used for the extraction and registration of the point cloud of the next section of rail until the extraction and registration of all rail point clouds are completed.
  • the contact wires of China's electrified railways are laid out in a "zigzag" shape, and the pull-out value of the contact wires changes continuously along the line and the extreme value is at the locator of the catenary column.
  • the cylindrical segmentation algorithm is used to segment the contact line and suspension point cloud.
  • the segmentation principle is shown in Figure 9, and the random sampling consistency of the straight line is used to extract the contact line points Cloud and dimensional analysis method to extract suspension points, the specific process is as follows:
  • the contact line point C selects the contact line point C to obtain its three-dimensional coordinates; secondly, use the distance discrimination method to obtain the trajectory line segment MN corresponding to point C, and translate the trajectory line segment MN to C, The line segment after translation is M'N'; finally, a cylinder is established with M'N' as the axis and radius r, and the point cloud of catenary structure is traversed, and the point cloud of contact line and suspension point is obtained by segmentation.
  • the contact line is laid out in a "zigzag" shape along the line direction, and the pull-out value is the maximum value at the locator.
  • TG/GD 124-2015 “Rules for Operation and Maintenance of High-speed Railway Catenary”
  • TG/GD116-2017 “Rules for Operation and Maintenance of Normal-speed Railway Catenary”
  • the limit value of the pull-out value of normal-speed railway and high-speed railway contact line is 450mm.
  • the radius r of the cylinder can be calculated according to formula (25).
  • L is the distance between catenary columns, generally 50m.
  • PCA Principal component analysis
  • the radius r is 0.1 m
  • the linear eigenvalue threshold L ⁇ is 0.9.
  • the measurement of guide height and pull-out value can be realized by using the detection results of track, contact line and suspension point.
  • the specific method is as follows:
  • m the horizontal distance between the contact line of the positioning point and the center line of the line
  • Step 1 point cloud data acquisition and preprocessing:
  • the light railway mobile measurement system is used to acquire point cloud data, see Figure 12.
  • the light railway mobile measurement system consists of a light assembled rail car, a positioning system (including a GNSS receiver, an inertial measurement unit), a laser scanner and a panoramic camera.
  • the maximum measurement speed of the light railway mobile measurement system is 30km/h
  • the laser emission frequency is 1000kHz
  • the scanning line is 200 lines per second
  • the maximum measurement range is 119m
  • the relative accuracy of measurement is better than 1mm.
  • Yantong high-speed railway is a high-speed railway connecting Yancheng City and Nantong City, Jiangsu province. It is the "Coastal Corridor", one of the main passages of "eight vertical and eight horizontal" high-speed railways in the "Medium and Long-term Railway Network Planning” (2016 revision). The important part is from Yancheng Station to Nantong West Station.
  • the main line has a total length of 157.098 kilometers, with 6 stations and a maximum design speed of 350 km/h. The location of the line is shown in Figure 13.
  • the three-dimensional laser point cloud along the rail transit is obtained.
  • the angle threshold of the track and the catenary is set.
  • the angle segmentation algorithm is used for the segmented rail and catenary point cloud.
  • the statistical filtering algorithm is used to remove the point cloud noise.
  • the preprocessed track and catenary laser point cloud are more conducive to subsequent processing.
  • Step 2 Rail point cloud segmentation and registration:
  • the track line is fitted with a straight line at 1m intervals, and the fitting line segment is used as the reference line, and the relatively fixed spatial position relationship between the moving scanning track line and the track is used, and the cuboid segmentation algorithm is used to realize the laser point cloud of the rail segment segmentation.
  • PCA coarse registration and GICP fine registration are used to realize the precise registration of rail point cloud and rail model point cloud, and based on the parameters of the registered rail model point cloud and the transformation relationship of registration, the rail point cloud space is realized The precise measurement of geometric parameters further fits the line centerline.
  • Step 3 Extraction of contact lines and suspension points:
  • the cylinder segmentation algorithm is used to segment the contact line and the point cloud of the suspension point.
  • the straight line random sampling consensus algorithm is used to extract the contact line point cloud
  • the dimensional analysis method is used to extract the suspension point point cloud.
  • Step 4 Calculation of guide height and pull-out value:
  • the precise measurement of guide height and pull-out value can be realized by using the detection results of track, contact line and suspension point.
  • the measurement accuracy of the instrument is ⁇ 3mm, and the measurement accuracy of the pull-out value is ⁇ 4mm.
  • the measurement results of the detection method of the present invention are compared with the detection data of the DJJ-8 measuring instrument, and the comparison results are shown in Figure 15. According to statistics, the medium error of the pull-out value difference is ⁇ 6.6mm, and the maximum difference is 9.9mm; the medium error of the guide height difference is ⁇ 6.3mm, and the maximum difference is 9.6mm.
  • the measurement accuracy meets the "High-speed Railway Power Supply Safety Inspection Monitoring system (6C system) general technical specification "rements.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Multimedia (AREA)
  • Length Measuring Devices By Optical Means (AREA)

Abstract

本发明公开了一种基于车载移动激光点云的接触网导高与拉出值自动检测方法,首先,根据轨道结构和接触网结构对应的扫描角度,设置角度分割阈值,分割得到轨道点云和接触网点云并使用统计滤波算法剔除了点云噪声;其次,基于移动扫描的轨迹线,采用长方体分割算法分割钢轨点云、圆柱分割算法分割接触线和定位点点云;再次,使用PCA和GICP配准钢轨点云、RANSAC提取接触线点云、维度分析法提取悬挂点点云;最后,根据导高、拉出值定义计算接触线的导高、拉出值。该方法具有精度高、效率高、自动化程度高等特点,可作为接触网导高与拉出值常态化检测方式的有效补充,服务于新建铁路静态验收和既有铁路线路运营检测。

Description

基于车载移动激光点云的接触网导高与拉出值自动检测方法 技术领域
本发明涉及轨道交通快速综合检测领域,特别是涉及一种基于车载移动激光点云的接触网导高与拉出值自动检测方法。
背景技术
铁路是国家重要的基础设施,国民经济的大动脉,在综合交通运输体系中处于骨干地位。1975年宝成线全线完成电气化改造,成为了我国第一条电气化铁路。自此以后,我国电气化铁路发展迅速,截至2020年底,我国铁路运营总里程超过14.63万公里,其中高速铁路已达到3.8万公里,电气化里程10.6万公里,电化率72.8%。接触网作为电气化铁路的核心部分,在保障铁路运输安全、提高运输效率、降低运输能耗等方面发挥着重要作用。
接触网放置在露天环境中,容易在恶劣的环境中老化和发生几何形位的改变。列车运行中,由于运行不规律、弓网间关系异常、电气腐蚀等方面的原因,接触网会发生变形甚至断线的情况。当接触网发生故障时,由于无备用线路,整条铁路线路的运营都将会受到影响,从而造成巨大的经济损失和不良的社会影响。接触网几何参数主要包括接触线导高和拉出值,导高是指接触线底面至轨面的垂直距离,拉出值是指接触线在定位点对受电弓中心线的偏移量。对于新建接触网工程,应在验收初期及时发现并消除由于安装误差引起的导高和拉出值超限,确保后续动态提速测试与试运营安全顺利进行。在运营阶段,为保证电力机车受电弓与接触线之间有良好接触,减少磨耗,避免发生刮弓或钻弓事故,接触线导高和拉出值需保持在一定范围内,因此有必要对接触线导高、拉出值进行常态化检测。
接触网导高与拉出值自动检测通常采用接触式检测和非接触式检测。相较于接触式检测,非接触式检测方式更加精确、安全和智能。非接触式检测方法主要包括图像检测和激光检测。图像检测是在检测车顶安装CCD像机获取图像,利用图像识别技术检测接触网的导高和拉出值。由于测量基准不同,图像检测方法通常需要对车体振动进行补偿,计算过程复杂。同时,图像检测方法易受到天气和光照等因素的影响,导致检测质量不稳定,目标识别难度较大,准确度较低等问题。目前,供电段、维管段各工区现场一般采用传统的便携式激光测量设备,在天窗点内对接触线导高与拉出值进行静态点式测量,在接触线每个定位处利用激光测距仪人工瞄准捕获接触线进行参数的抽样测量,劳动强度大、作业效率低、安全系数低,检测功能单一。因此,急需研发一种检测效率高、检测精度高、自动化程度高以及无接触式的接触网导高与拉出值检测新技术,保障接触网系统正常稳定工作。
车载移动激光扫描技术是指在移动载体上集成全球卫星导航系统(Global Navigation Satellite System,GNSS)、惯性测量单元(Inertial Measure Unit,IMU)、激光扫描仪、数码相机、数码摄像机等多种传感器的综合测量检测技术。各类型传感器在移动状态下自动采集 各种位置、姿态、影像和激光扫描数据,通过统一的地理参考和数据采集同步技术,实现无接触式的空间地理信息采集、处理与入库。在作业过程中,将集成的三维激光扫描系统搭载在轨道车上,通过载体的移动,快速采集轨道交通两侧几十至几百米范围内海量点云和影像数据。通过地面GNSS基站、移动GNSS接收机、地面控制点、IMU和激光扫描仪数据的联合解算,得到高精度三维激光点云数据。相对于地面激光扫描和机载激光扫描,车载移动扫描具有高效灵活的数据采集方式,被越来越多的应用于工程实践中,车载移动扫描技术是目前研究的热点之一。近年来,越来越多的学者利用车载移动激光点云,实现接触网导高与拉出值检测的研究。
西南交通大学的周靖松在其硕士论文《基于三维点云数据的接触网几何参数检测研究》中论述了根据接触网点云特征改进随机抽样一致RANSAC算法,提取了接触线点云数据;再由空间旋转平移信息得出相机坐标系到世界坐标系的转换矩阵,并由此计算出接触线的导高和拉出值。该方法需要将测量基准由相机中心转换到轨道平面,测量结果易受车体振动和轮轨游间等因素影响。
中国发明专利公开号CN103852011A公开了一种基于激光雷达的铁路接触网几何参数分析方法,该方法利用接触线点云的导高值最小,拉出值极值位于接触网立柱处等信息,实现了接触网几何参数的检测。但是该方法存在以下缺点:(1)在接触网立柱处拉出值不是极值的情况下,此方法失效;(2)此方法没有检测吊弦悬挂点位置处的导高和拉出值。
发明内容
为解决现有技术存在的问题,本发明提供一种基于车载移动激光点云的接触网导高与拉出值自动检测方法,该方法具有效率高、准确性高和自动化程度高等特点。
为此,本发明的技术方案如下:
一种基于车载移动激光点云的接触网导高与拉出值自动检测方法,包括以下步骤:
S1,点云数据采集与预处理:
利用车载移动激光扫描系统对轨道交通沿线的基础设施及周边环境进行点云数据快速采集;基于沿线采集的激光点云,通过设定轨道和接触网点云对应的激光扫描发射角度的范围值,采用角度阈值分割算法,从轨道交通全断面激光点云中分割出包含轨道和接触网的激光点云;针对分割得到的轨道和接触网激光点云中的噪声点,采用统计滤波算法进行剔除;
S2,钢轨点云分割与配准:
利用改进的空间直线的最小二乘拟合算法使之能对轨迹线进行分段线性拟合,得到等间隔的轨迹线段,基于分段拟合的轨迹线段和对应轨道之间相对稳定的空间位置关系,采用长方体分割算法从S1预处理后的轨道激光点云中提取钢轨点云;根据钢轨标准断面图生成标准钢轨模型点云,使用主成分分析算法和广义迭代最近点算法实现钢轨点云与钢轨模型点云的精确配准,从而获取钢轨空间几何参数,提取轨顶中线,计算线路中点,进而拟合出线路中心线;利用提取的轨顶中线更新轨迹线和钢轨点云之间的空间位置参数,进行下一段钢 轨点云的提取与配准;迭代重复以上步骤,完成全部钢轨点云的提取和配准;
S3,接触线和悬挂点点云自动提取:
接触线和悬挂点提取:根据轨迹线和接触网结构之间相对稳定的空间位置关系,基于S2中分段线性拟合得到的等间隔轨迹线段,采用圆柱分割算法从S1中得到的接触网激光点云中提取包含接触线和悬挂点的点云;采用直线的随机采样一致性算法对包含接触线和悬挂点的点云进行三维直线模型拟合,根据标准接触导线的横断面设计图设置随机采样一致性算法的参数,进行接触导线点云分段提取与三维直线模型的自动拟合;在只包含接触线的区域,点云的维度特征呈线性;在悬挂点位置,由于吊弦和定位器与接触线相连,点云的维度特征表现为非线性。采用维度分析法,分析目标点邻域的维度特征,设置线性阈值提取接触线悬挂点;
S4,接触网导高、拉出值自动检测:
根据导高和拉出值的定义及测量方法,利用S2中钢轨几何参数的检测结果、S3中接触线和悬挂点的检测结果,对接触线导高和拉出值进行自动检测。
上述的步骤S1中,所述点云数据采集与预处理包括以下步骤:
(1)点云数据采集:
随着车载移动激光扫描系统的移动,激光扫描仪在垂直于线路方向或与线路方向呈45度方向进行圆周扫描,得到被测轨道交通长度范围内及轨道两侧一定距离范围的激光点云;利用所述车载移动激光扫描系统的轨迹线对所扫描的数据进行解算,获得全线统一地理空间坐标系下的激光点云,将所述全线统一地理空间坐标系下的激光点云导出成通用数据格式;
(2)激光点云角度阈值分割:
根据扫描仪在采集数据过程中记录的扫描头旋转的角度信息、扫描头每次步进旋转的角度值以及扫描仪圆周扫描的角度起点分割出一定角度范围的激光点云;根据扫描仪的安装高度、安装角度和分割区域的几何尺度,通过空间几何分析,利用三角函数公式计算得到分割区域所在的角度范围,或者直接在激光点云上选择轨道或者接触网结构的边界点,查看边界点的扫描角度值;再设定一个扫描角度值的浮动值δ,根据公式(1),遍历所有激光点云,分割出轨道和接触网点云:
Figure PCTCN2021124110-appb-000001
式中:
P i:第i个激光点;
Segment RailTrack、Segment Catenary、Segment Other:依次为轨道点云、接触网点云、其它点云;
α min、α max:分别为轨道对应的最小扫描角度和最大扫描角度,单位为度;
β min、β max:分别为接触网结构对应的最小扫描角度和最大扫描角度,单位为度;
δ:角度分割浮动值;
Figure PCTCN2021124110-appb-000002
是轨道点云中第i个激光点的扫描角度,单位为度;
Figure PCTCN2021124110-appb-000003
是接触网点云中第i个激光点的扫描角度,单位为度;
优选的是,所述角度分割浮动值δ为3度;
(3)点云噪声过滤:
使用统计滤波算法剔除三维激光扫描系统获取的点云数据中的少量的噪声点。
步骤(1)中,所述通用数据格式为LAS格式,扫描仪的旋转角度范围为-180°~180°,扫描仪正下方为扫描起始方向,旋转至扫描仪正上方为180°,轨迹线前进方向左侧角度为负值,右侧角度为正值。所述轨道两侧一定距离范围为线路左右两侧各100米的宽度范围。
步骤(2)中,如果车载移动激光扫描系统含有多个扫描仪,则分别按照每个扫描仪角度分割激光点云,再将多个扫描仪分割的激光点云进行合并。
具体地,步骤S2中,钢轨点云的分割与配准步骤如下:
(1)轨迹线分段线性拟合:首先建立轨道交通激光点云的多维空间索引结构;设定固定步长(Fixed Depth),分段线性拟合轨迹线,获得等间距的多段线;
(2)钢轨点云分割与配准,首先,右侧钢轨的点云分割与配准包括以下步骤:
1)轨迹线平移:
选择检测起始点,利用距离判别法找到起始点对应的轨迹线段AB,将该轨迹线段AB平移到起始点的右侧钢轨顶面点R处,平移后的线段为A'B';
2)长方体分割:
利用钢轨几何尺寸以及平移后的轨迹线段为A'B',建立长方体包围盒,分割钢轨点云;以直线A'B'为轴,在A'B'长度范围内两侧各s宽度,建立竖直面作为长方体包围盒的左右侧面,所述s可取钢轨底面宽度的一半;设A'B'的最大高程为H max、最小高程为H min、钢轨的高度为H rail;运营线路的钢轨顶面光带会导致钢轨顶面点云的离散度较大,因此设置一个高度浮动值ε,长方体包围盒顶面高程确定为H max+ε,底面高程为H min-H rail-ε,循环遍历轨道点云,分割长方体包围盒内的钢轨点云;优选的是,所述高度浮动值ε为0.01m;
3)点云配准:
根据检测线路使用钢轨的标准断面图,建立网格化三维标准钢轨模型并对其进行均匀采样,形成标准钢轨模型点云;经典的ICP(Iterative Closest Point)配准算法和其改进的算法都对初始配准条件要求相对严格,需要待配准点云的重叠度高,否则容易陷入局部最优陷阱,所以需要先利用粗配准算法使待配准点云对齐。经过PCA粗配准后,两组点云已基本重合, 这为精确配准提供良好的初始条件。采用GICP算法实现钢轨点云的精配准,该算法将“点到点”ICP和“点到面”ICP结合到概率框架模型上,利用点云的协方差矩阵来构建配准的成本函数,提高了经典ICP配准的速度和鲁棒性。首先使用粗配准算法使待配准的钢轨点云和标准钢轨模型点云大致对齐,再采用GICP算法实现钢轨点云的精配准。利用精配准后的钢轨模型点云进行钢轨的空间几何参数的计算,再以基准轨道为基础,分段计算线路中点,将这些分段提取的线路中点依次连接,形成线路中心线;
4)几何参数更新:
设经过配准后的钢轨模型点云顶面中线为A”B”,将B”替代R点作为新的检测起始点,重新计算轨迹线相对钢轨的几何关系;
5)循环迭代:
循环进行步骤1)~4),将新的几何参数用于下一段钢轨点云的提取和配准,直至完成所有钢轨点云的提取与配准。
按上述步骤1)-5)对左侧钢轨的点云进行分割与配准。
优选的是,步骤(1)中,所述轨道交通激光点云的多维空间索引结构为KD树。
上述的步骤3)中,利用所述配准后的钢轨三维几何模型进行钢轨几何参数的计算,得到轨距、轨向信息,根据基准轨道的定义确定基准轨道后,再以基准轨道为基础,向另外一根钢轨水平偏移1/2标准轨距(中国定义的标准轨距为1435mm),分段计算线路中点,再将这些分段提取的线路中点依次连接,形成线路中心线,将所形成的线路中心线再进行平滑处理,得到相对平滑的线路中心线。
步骤S3中,接触线和悬挂点点云自动提取的步骤如下:
(1)接触线和悬挂点点云分割:
对于角度分割后的接触网结构,使用圆柱分割算法分割接触线和悬挂点点云,首先,选择接触线点C,获得其三维坐标;其次,采用距离判别法,获取C点对应的轨迹线段MN,将轨迹线段MN平移到C处,平移后的线段为M'N';最后,以M'N'为轴,半径r建立圆柱,遍历接触网结构点云,分割得到接触线和悬挂点点云。接触线沿着线路方向呈“之”字形布设,拉出值在定位器处为极大值。TG/GD 124—2015《高速铁路接触网运行维修规则》和TG/GD116—2017《普速铁路接触网运行维修规则》规定了普速铁路和高速铁路接触线拉出值的限界值为450mm。按照拉出值的极限条件下考虑,圆柱半径r可按式(2)计算:
Figure PCTCN2021124110-appb-000004
式中:L为接触网立柱间距,一般为50m;
优选的是,所述半径r为0.1m;
(2)接触线提取:
因接触线截面积较小且距离测量系统较远,当扫描作业速度过快时,会导致接触线点云稀少,出现一个接触线断面只有一个点的情况,故采用随机采样一致性的直线模型。将接触 线的实际截面直径(高度)设置为直线采样的距离阈值,线性采样获取接触线点云和直线模型起终点M”N”。
(3)悬挂点提取:
采用主成分分析(PCA)解算出目标点邻域点云集合对应的特征向量和特征值,然后根据特征值λ 1、λ 2、λ 3123)判断点云维度特征。在只包含接触线的区域,点云呈线性特征;在悬挂点处,吊弦和定位器与接触线相连,点云的维度特征表现为非线性。遍历圆柱分割后的点云,计算目标点半径r邻域点云的线性特性值,设置线性特征值阈值L λ,提取悬挂点。对于悬挂点周围点云,可以根据与λ 1对应的特征向量判断方向,与地面垂直的点为吊弦悬挂点,近似平行的为定位悬挂点。优选的是,线性特征值阈值L λ为0.9。
(4)循环迭代:
将直线模型终点N”点替代C点,作为新的检测起始点,重复1~3步,直至完成所有接触线和悬挂点的提取。
步骤S4中,接触网导高与拉出值检测方法如下:
(1)导高计算:
根据悬挂点和接触线的分割结果,提取悬挂点处接触线底部高程,找到对应位置的轨面高程,求差后获取接触线导高值;
(2)拉出值计算:
对于直线地段,不存在外轨超高,直接计算定位悬挂点处接触线底部点至线路中心线的XY平面的偏移量;对于曲线地段,由于存在外轨超高,接触线的拉出值按式(3)和(4)计算:
a=m+c       (3)
Figure PCTCN2021124110-appb-000005
式中:a:接触网拉出值;
m:定位点接触线与线路中心线的水平距离;
c:定位点受电弓中心与线路中心线的水平距离;
h:外轨超高;
H:接触线高度;
L:轨距;
优选的是,所述对应位置的轨面的寻找方法为距离判别法;
所述外轨超高根据权利要求1中的步骤S2得到的钢轨空间几何参数计算。
本发明基于轻型车载移动测量系统获取的铁路三维激光点云数据,提供了一种接触网导高与拉出值的自动化检测方法。首先,根据轨道结构和接触网结构对应的扫描角度,设置角度分割阈值,分割得到轨道点云和接触网点云并使用统计滤波算法剔除了点云噪声;其次,基于移动扫描的轨迹线,采用长方体分割算法分割钢轨点云、圆柱分割算法分割接触线和定 位点点云;再次,使用PCA和GICP配准钢轨点云、RANSAC提取接触线点云、维度分析法提取悬挂点点云;最后,根据导高、拉出值定义计算接触线的导高、拉出值。
本发明具有以下有益效果:
(1)已有的检测方法多为图像检测的方式,图像检测方法易受到天气和光照等因素的影响,导致检测质量不稳定,目标识别难度较大,准确度较低等问题。本发明的方法基于激光扫描方式,不受光照等因素的影响,环境适应性强,检测结果稳定可靠。
(2)已有的激光检测方法将激光扫描仪搭载于轨道车顶,只获取接触线点云,还需根据轨道车与轨道的几何关系,将测量基准由扫描仪中心转换到轨道平面,计算精度易受车体振动和轮轨游间等因素影响。本发明基于铁路线路的全断面点云获取轨道的几何参数、接触线和悬挂点位置,直接进行接触线导高、拉出值的计算,无需进行测量基准的转换,有效提高了测量精度。
(3)已有的钢轨点云分割算法是利用钢轨表面局部高程值最大或者钢轨表面点云的方向变化连续性等特征,利用高程直方图统计计算的方法或者卡尔曼滤波方法进行钢轨点云分类,已有方法只提取钢轨点云,未获取钢轨的精确的钢轨空间几何参数且钢轨点云提取过程繁杂。本发明基于移动扫描的轨迹线和钢轨之间稳定的位置关系,利用长方体分割算法实现点云的提取,采用PCA和GICP实现钢轨点云和钢轨模型点云的精确配准,获取了钢轨精确的空间几何参数。
(4)已有的接触线提取方法基于接触网全部激光点云提取接触线激光点云。本发明利用圆柱分割算法,分割得到接触线点云和悬挂点点云,再进行后续的接触线提取和悬挂点提取,有效提高了计算效率。
(5)已有的基于激光点云的接触网导高与拉出值检测方法,没有定位悬挂点位置,悬挂点位置的导高和拉出值才是电务部门所关心的。本发明根据点云的邻域特征,采用维度分析法确定接触线悬挂点位置,并根据邻域点云协方差矩阵的第一特征对应的特征向量的Z值分量大小,识别悬挂点的类型。
(6)本发明的方法测量接触线导高、拉出值的精度为0~10mm,测量精度满足《高速铁路供电安全检测监测系统(6C系统)总体技术规范》要求。
(7)相对传统接触线导高与拉出值检测方法,本发明不仅提高了检测的精度,还可以快速实现接触网导高与拉出值的检测。采用车载移动激光扫描方式,避免了检测人员上线作业,降低了安全作业风险。本发明能作为新型接触网导高、拉出值测量方式,能服务于新建铁路线路静态验收和既有铁路线路常态化检测。
附图说明
图1为本发明的接触网导高与拉出值检测方法的流程图;
图2为为本发明的方法预处理后的轨道与接触网点云;
图3为本发明中钢轨点云分割原理图;
图4为本发明中钢轨点云的分割结果;
图5为本发明中配准流程图;
图6a为标准钢轨模型;
图6b为标准钢轨点云;
图7为配准结果;
图8为轨道检测结果;
图9为接触线提取示意图;
图10为维度分析法原理图;
图11为接触线和悬挂点检测结果;
图12为使用轻型铁路移动测量系统获取点云数据的状态图;
图13为盐城至南通的高铁线路走向示意图;
图14为本发明中得到的盐通高铁激光点云;
图15为本发明中导高、拉出值检测结果和DJJ-8测量仪检测数据的比较图。
具体实施方式
下面结合附图和具体实施例进一步说明本发明的技术方案。
本发明的接触网导高与拉出值自动检测方法包括:点云数据采集与预处理;钢轨点云分割与配准;接触线和悬挂点点云提取以及导高、拉出值计算。如图1所示,上述步骤具体如下:
S1,数据采集与预处理:利用车载移动激光扫描系统对轨道交通沿线的基础设施及周边环境进行快速数据采集;基于沿线采集的激光点云,通过设定轨道点云和接触网点云对应的激光扫描发射角度的范围值,采用角度阈值分割算法,从轨道交通沿线全部激光点云中分割出轨道和接触网激光点云,再采用统计滤波算法剔除点云噪声,具体方法如下:
(1)点云数据采集:
车载移动激光扫描系统激光扫描仪采用机械式旋转反射棱镜方式,实现360°圆周断面扫描,再通过扫描仪搭载平台在轨道上移动,实现轨道交通范围及两侧一定距离范围的扫描。
扫描仪在采集数据过程中会记录扫描头旋转的角度信息,扫描头每次步进旋转的角度值是固定的,每次圆周扫描的角度起点也是固定的,一般是从扫描仪正下方开始,并以这点结束,完成圆周扫描。由于接触网导高与拉出值得计算只涉及轨道和接触网范围的点云,为了减少背景点云和其它结构点云的影响、提高点云分割计算速度及准确性,首先按照轨道和接触网结构对应的扫描角度,分割得到轨道和接触网点云。
利用移动扫描轨迹线对扫描数据进行解算,获得全线统一坐标系下的激光点云。将预处理后的扫描数据导出成通用数据格式,例如LAS格式的数据,LAS格式的激光点云数据记录的扫描仪旋转角度范围为-180°~180°,即:扫描仪正下方为扫描起始方向,旋转至扫描仪正上方时扫描角为180°,轨迹线前进方向左侧角度为负值,右侧角度为正值。
(2)激光点云角度阈值分割:
根据扫描仪在采集数据过程中会记录的扫描头旋转的角度信息、扫描头每次步进旋转的角度值以及扫描仪圆周扫描的角度起点分割出一定角度范围的激光点云;根据扫描仪的安装高度、安装角度、分割区域的几何尺度等参数,通过空间几何关系,利用三角函数公式计算得到分割区域所在的角度范围,或者直接在激光点云上选择轨道或者接触网结构的边界点,查看选择点的扫描角度值;考虑车载扫描过程中,因车辆晃动引起的扫描仪发射角度变化,设定一定角度值的浮动值δ(如3度角),根据公式(1),遍历所有激光点云,分割出轨道和接触网点云数据:
Figure PCTCN2021124110-appb-000006
式中:
P i:第i个激光点;
Segment RailTrack、Segment Catenary、Segment Other:依次为轨道点云、接触网点云、其它点云;
α min、α max:分别为轨道对应的最小扫描仪角度和最大扫描仪角度,单位为度;
β min、β max:分别为接触网结构对应的最小扫描仪角度和最大扫描仪角度,单位为度;
Figure PCTCN2021124110-appb-000007
是轨道点云中第i个激光点的扫描角度,单位为度;
Figure PCTCN2021124110-appb-000008
是接触网点云中第i个激光点的扫描角度,单位为度。
如果车载移动激光扫描系统含有多个扫描仪,需要分别按照每个扫描仪角度分割激光点云,再将多个扫描仪分割的激光点云进行合并。
(3)点云去噪:
三维激光扫描系统获取点云数据时,由于仪器的震荡、被测物表面粗糙不平、镜面反射、遮挡物遮挡等各种因素会使采集到的点云含有少量的噪声点。这些少量远离主体结构的噪声点将会影响钢轨点云和接触线点云的提取,使用统计滤波算法,剔除这些噪声点。
S2,钢轨点云分割与配准:
利用车载移动激光扫描系统的轨迹线和钢轨之间相对固定的空间位置关系,采用数据驱动的方法,实现左右钢轨激光点云的自动分割。
通过上述步骤(2)中的角度分割阈值进行激光点云分割,将背景点云数据和其它结构点云进行剔除,仅保留轨道和接触网点云。对于分割后的轨道和接触网点云通过上述步骤(3)中的统计滤波算法,剔除点云噪声,预处理结果如图2所示。
铁路三维空间结构和车载移动激光扫描系统扫描作业都有其自身的特点:(1)在沿线路 方向上,两根钢轨基本平行,且两根钢轨内侧面的距离一般与标准轨距(1435mm)非常接近;(2)车载移动激光扫描系统的车轮卡在轨道上沿着轨道行驶,扫描仪与两轨的相对位置变化很小,扫描轨迹线线形与铁路线路的线形是保持一致的;(3)钢轨固定在轨枕上或无砟轨道板上,钢轨高于轨枕、道砟和轨道板平面。
移动激光扫描轨迹文件包含激光扫描系统中心的空间位置及姿态数据。轨迹线的空间位置数据是扫描仪在参考坐标系统下的空间位置点,依次将这些空间位置点连接,形成移动扫描的轨迹线。当整套扫描系统安装至移动平台上进行扫描时,轨迹线相对左右钢轨的空间位置将固定(直线线路)或者在很小的范围内连续变化(曲线线路)。本发明利用轨迹线和钢轨空间位置相对固定和全线连续变化的特点,结合钢轨固有的结构信息,快速实现钢轨激光点云的分割。由于分割得到的钢轨点云的空间几何参数和几何信息是未知的,因此本发明首先根据铁路线路所使用钢轨的标准断面图建立标准的钢轨模型点云(如60kg/m钢轨,这是中国比较常用的钢轨类型),再采用粗配准和精配准相结合的方式,使分割的钢轨点云和钢轨模型点云精确配准,根据配准后的钢轨模型点云获取钢轨的空间几何参数。
为了分割得到钢轨的空间几何参数,分割原理图见图3,共需要进行以下两步工作:
(1)轨迹线分段拟合:
为提高钢轨点云的提取效率和提取精度,采用分段的方法提取钢轨点云。改进空间直线的最小二乘拟合算法,对轨迹线进行固定步长的线形拟合,获得间距为D的多段线。改进的空间直线的最小二乘拟合算法原理如下:
1)设轨迹线数据集共有N个点,起始点坐标为P i(X i,Y i,Z i),第j个点的坐标为P j(X j,Y j,Z j),两点之间的欧式距离为d ij
2)如果d ij≤D,则将j点置于集合A中。采用空间直线的最小二乘拟合算法对集合A中的点进行线性拟合,原理如下:
空间直线标准方程式为
Figure PCTCN2021124110-appb-000009
对(2)式进行等价转换,得到两平面方程为
Figure PCTCN2021124110-appb-000010
Figure PCTCN2021124110-appb-000011
式中,
Figure PCTCN2021124110-appb-000012
因此,空间直线可视为两平面相交得到的,对这两个平面方程进行拟合,进一步可拟合得到空间直线。利用残差平方和最小理论得出最优直线参数,两平面的残差平方和为
Q 1=∑(x i-k 1z i-b 1) 2        (5)
Q 2=∑(y i-k 2z i-b 2) 2        (6)
最小化残差平方和,分别对k 1和k 2求导得
Figure PCTCN2021124110-appb-000013
联立(7)中各式,解得平面参数为
Figure PCTCN2021124110-appb-000014
Figure PCTCN2021124110-appb-000015
Figure PCTCN2021124110-appb-000016
Figure PCTCN2021124110-appb-000017
3)过P i作拟合直线的垂线,垂足为P N,计算沿拟合直线距离垂足为D的点P M(沿扫描方向),将拟合直线端点P M作为新的P i点。
4)循环1)、2)、3)步,直至完成整条轨迹线的分段直线拟合。
(2)钢轨点云分割与配准,包括以下步骤:
以下以右侧钢轨为例进行说明,左侧钢轨的处理过程和右侧相同。
1)轨迹线段平移:
选择钢轨的右侧顶面中心点R和左侧顶面中心点L,采用距离判别法,得到R和L对应的分段拟合的轨迹线段AB。将轨迹线段AB平移到R处,平移后的轨迹线段为A'B'。
2)长方体分割:
利用钢轨几何尺寸以及平移后的轨迹线段A'B'建立长方体包围盒,分割钢轨点云。具体为:以直线A'B'为轴,在A'B'长度范围内两侧各s(s可取钢轨底面宽度的一半)宽度,建立竖直面作为长方体包围盒的左右侧面。设A'B'的最大高程为H max、最小高程为H min、钢轨 的高度为H rail。运营线路钢轨光带会导致钢轨顶面点云的离散度较大,因此设置一个高度浮动值ε,长方体包围盒顶面高程确定为H max+ε,底面高程为H min-H rail-ε。循环遍历轨道点云,分割长方体包围盒内的钢轨点云,钢轨点云分割结果见图4。
3)钢轨点云配准:
在完成钢轨点云的分割后,本发明采用PCA和GICP配准的方式使钢轨点云与钢轨模型点云精确配准,从而获取钢轨的精确空间几何参数,配准流程图见图5,具体如下:
a)标准钢轨模型点云生成:
根据检测线路所使用的钢轨类型的标准断面图,建立网格化三维钢轨模型并对其进行均匀采样,形成标准钢轨模型点云,如图6a、6b所示。
b)基于PCA的点云粗配准
经典的ICP(Iterative Closest Point)配准算法和其改进的算法都对初始配准条件要求相对严格,需要待配准点云的重叠度高,否则容易陷入局部最优陷阱,所以需要先利用粗配准算法使待配准点云大致对齐。由于铁路钢轨的钢轨长度远大于钢轨高度和宽度时,因此钢轨点云空间呈现线形分布,本发明将主成分分析算法(PCA)算法运用到钢轨点云的粗配准。PCA进行粗配准的具体流程如下:
假设提取的钢轨点云为S i(x i,y i,z i),i=1,2,3...,m;钢轨模型点云为G i(x i,y i,z i),i=1,2,3...,n。计算两组点云的均值
Figure PCTCN2021124110-appb-000018
Figure PCTCN2021124110-appb-000019
Figure PCTCN2021124110-appb-000020
Figure PCTCN2021124110-appb-000021
分别构造两组点云的协方差矩阵COV S和COV G
Figure PCTCN2021124110-appb-000022
Figure PCTCN2021124110-appb-000023
通过奇异值分解算法(SVD)计算协方差矩阵COV S和COV G的特征值和特征向量。
PCA配准的刚性变换矩阵Trans由旋转矩阵R P和平移矩阵T P构成。
Figure PCTCN2021124110-appb-000024
Figure PCTCN2021124110-appb-000025
Figure PCTCN2021124110-appb-000026
式中,U G和U S代表由协方差矩阵特征向量构成的矩阵。
经过PCA计算的刚性变换矩阵Trans转换后,两组点云已基本重合,这为精确配准提供良好的初始条件。本发明采用GICP算法进行精配准,该算法将“点到点”ICP和“点到面”ICP结合到概率框架模型上,利用点云的协方差矩阵来构建配准的成本函数,提高了经典ICP配准的速度和鲁棒性,原理如下:
假设通过最近点查找,找到两组点云A和B:
Figure PCTCN2021124110-appb-000027
式中,a i和b i相互匹配。
在概率模型中,我们假设有一组潜在的点
Figure PCTCN2021124110-appb-000028
Figure PCTCN2021124110-appb-000029
(乱码):
Figure PCTCN2021124110-appb-000030
两组点云A和B可通过高斯分布来生成:
Figure PCTCN2021124110-appb-000031
式中:C i是每个点的协方差矩阵。
定义刚体平移旋转矩阵为T,每一对对应点的配准误差d i则为:
d i=b i-Ta i        (22)
因为a i和b i相互独立且都服从高斯分布,因此d也服从高斯分布:
Figure PCTCN2021124110-appb-000032
利用极大似然估计,计算刚体平移旋转矩阵T的成本函数为:
Figure PCTCN2021124110-appb-000033
利用PCA和GICP计算得到的变换矩阵实现钢轨点云和钢轨模型点云精确配准,配准结果如图7所示。利用配准后的钢轨模型点云,进行轨道几何参数的计算,得到实际轨距、轨向等信息,根据基准轨道的定义,确定基准轨道后,再以基准轨道为基础,向另外一根钢轨水平偏移1/2标准轨距(中国定义的标准轨距为1435mm),分段计算线路中点,再将这些分段提取的线路中点依次连接,形成线路中心线,优选的是,将所形成的线路中心线再进行平滑处理,得到相对平滑的线路中心线,如图8所示。
4)几何参数更新:
设经过配准后的钢轨模型点云顶面中线为A”B”,将B”替代R点作为新的检测起始点, 重新计算轨迹线相对钢轨的几何关系。
5)循环迭代:
循环进行步骤1)~4),将新的几何参数用于下一段钢轨点云的提取和配准,直至完成所有钢轨点云的提取与配准。
S3,接触线和悬挂点点云的提取:
中国电气化铁路的接触线呈“之”字形布设,接触线的拉出值沿线连续变化且极值在接触网立柱的定位器处。利用车载移动扫描的轨迹线和接触线之间局部相对固定的空间位置关系,使用圆柱分割算法分割接触线和悬挂点点云,分割原理图见图9,采用直线的随机采样一致性提取接触线点云、维度分析法提取悬挂点,具体流程如下:
(1)接触线和悬挂点点云的分割:
对于角度分割和统计滤波去噪后的接触网点云,选择接触线点C,获得其三维坐标;其次,采用距离判别法,获取C点对应的轨迹线段MN,将轨迹线段MN平移到C处,平移后的线段为M'N';最后,以M'N'为轴,半径r建立圆柱,遍历接触网结构点云,分割得到接触线和悬挂点点云。接触线沿着线路方向呈“之”字形布设,拉出值在定位器处为极大值。TG/GD 124—2015《高速铁路接触网运行维修规则》和TG/GD116—2017《普速铁路接触网运行维修规则》规定了普速铁路和高速铁路接触线拉出值的限界值为450mm。按照拉出值的极限条件下考虑,圆柱半径r可按式(25)计算。
Figure PCTCN2021124110-appb-000034
式中:L为接触网立柱间距,一般为50m。
(2)接触线提取:
因接触线截面积较小且距离扫描仪较远,当扫描作业速度过快时,会导致接触线点云稀少,出现一个接触线断面只有一个点的情况,故采用随机采样一致性的直线模型。将接触线的实际截面直径(高度)设置为直线采样的距离阈值,线性采样获取接触线点云和直线模型起终点M”N”。
(3)悬挂点提取:
采用主成分分析(PCA)解算出目标点邻域点云集合对应的特征向量和特征值,然后根据特征值λ 1、λ 2、λ 3123)判断点云维数特征,维度分析法原理见图10。在只包含接触线的区域,点云呈线性特征;在悬挂点位置,由于吊弦或定位器和接触线连接,此处邻域点云呈非线性。遍历圆柱分割后的点云,计算目标点半径r邻域点云的线性特性值L λ,设置线性特征值阈值,提取悬挂点。对于悬挂点位置点云,可以根据与λ 1对应的特征向量判断方向,与地面垂直的点为吊弦悬挂点,近似平行的为定位悬挂点。
优选的是,所述半径r为0.1m,线性特征值阈值L λ为0.9。
(4)循环迭代:
将直线模型终点N”点替代C点,作为新的处理起始点,重复以上过程,直至完成所有 接触线和悬挂点的提取,提取结果见图11。
S4,接触网导高、拉出值自动检测:
根据导高、拉出值的相关定义,利用轨道、接触线和悬挂点的检测结果可实现导高和拉出值的测量,具体方法如下:
(1)导高计算:
根据悬挂点和接触线的分割结果,提取悬挂点处接触线底部高程,找到对应位置的轨面高程,求差后获取接触线导高值。
(2)拉出值计算:
对于直线地段,不存在外轨超高,直接计算悬挂点处接触线底部点至线路中心线的XY平面的偏移量。对于曲线地段,由于存在外轨超高,接触线的拉出值按式(26)和(27)计算。
a=m+c       (26)
Figure PCTCN2021124110-appb-000035
式中:a:接触网拉出值;
m:定位点接触线与线路中心线的水平距离;
c:定位点受电弓中心与线路中心线的水平距离;
h:外轨超高;
H:接触线高度;
L:轨距。
实施例1
步骤1,点云数据采集与预处理:为了验证本发明方法的正确性,使用轻型铁路移动测量系统获取点云数据,见图12。轻型铁路移动测量系统由轻型组装式轨道车、定位系统(包括GNSS接收机、惯性测量单元)、激光扫描仪和全景相机组成,具有上线方便、作业方式灵活、检测精度和检测效率高等特点,可以获取铁路三维激光点云和高清全景影像数据。轻型铁路移动测量系统的最高测量速度为30km/h,激光发射频率为1000kHz,扫描线为200线/秒,最大测程为119m,测量的相对精度优于1mm。
为了评价该方法的可靠性和精度,选择在新建盐通高铁上进行扫描实验。盐通高铁是一条连接江苏省盐城市与南通市的高速铁路,它是《中长期铁路网规划》(2016年修订版)中“八纵八横”高速铁路主通道之一“沿海通道”的重要组成部分,由盐城站至南通西站,正线全长157.098千米,设6座车站,最高设计速度350千米/小时,该线路位置见图13。利用轻型铁路移动测量系统以20km/h的速度在盐通高铁全线进行扫描作业,采集盐通高铁线路高精度、高密度三维点云数据,其中轨道平均点密度为9000点/m 2,接触线平均点密度为60点/m,采集得到的盐通高铁点云数据如图14所示。
通过轻型铁路移动测量系统,获得轨道交通沿线三维激光点云,通过设定轨道和接触网的角度阈值,采用角度分割算法,实现轨道和接触网激光点云分割。对于分割得到的钢轨与接触网点云,再使用统计滤波算法剔除点云噪声。相对全部三维激光点云数据,预处理后的轨道和接触网激光点云更有利于后续的处理工作。
步骤2:钢轨点云分割与配准:
沿铁路线路方向,1m间隔直线拟合轨迹线,以拟合直线段为参考线,利用移动扫描轨迹线和轨道之间相对固定的空间位置关系,分段采用长方体分割算法,实现钢轨激光点云的分段分割。采用PCA粗配准和GICP精配准的方法,实现钢轨点云与钢轨模型点云的精确配准,基于配准后的钢轨模型点云的参数和配准的转换关系,实现钢轨点云空间几何参数的精确测量,进一步拟合出线路中心线。
步骤3:接触线和悬挂点的提取:
利用步骤2中分段拟合的轨迹线和接触网结构相对稳定的位置关系,采用圆柱分割算法分割得到接触线和悬挂点点云。对于分割后的接触线和悬挂点点云采用直线的随机采样一致性算法提取接触线点云,使用维度分析法提取悬挂点点云。
步骤4:导高、拉出值计算:
根据导高、拉出值的定义及测量方法,利用轨道、接触线和悬挂点的检测结果实现导高和拉出值的精确测量。
实验精度分析:
山东省科学院激光研究所济南蓝动激光技术有限公司研制的“DJJ-8激光接触网检测仪”在国内接触网导高与拉出值静态检测中得到广泛使用和认可,该仪器导高测量精度±3mm,拉出值测量精度为±4mm,将本发明检测方法的测量结果和DJJ-8测量仪检测数据进行比较,比较结果如图15所示。经统计,拉出值差值的中误差为±6.6mm,最大差值为9.9mm;导高差值的中误差为±6.3mm,最大差值9.6mm,测量精度满足《高速铁路供电安全检测监测系统(6C系统)总体技术规范》要求。

Claims (10)

  1. 一种基于车载移动激光点云的接触网导高与拉出值自动检测方法,包括以下步骤:
    S1,点云数据采集与预处理:
    利用车载移动激光扫描系统对轨道交通沿线的基础设施及周边环境进行点云数据快速采集;基于沿线采集的激光点云,通过设定轨道和接触网点云对应的激光扫描发射角度的范围值,采用角度阈值分割算法,从轨道交通全断面激光点云中分割出包含轨道和接触网的激光点云;针对分割得到的轨道和接触网激光点云中的噪声点,采用统计滤波算法进行剔除;
    S2,钢轨点云分割与配准:
    利用改进的空间直线的最小二乘拟合算法使之能对轨迹线进行分段线性拟合,得到等间隔的轨迹线段,基于分段拟合的轨迹线段和对应轨道之间相对稳定的空间位置关系,采用长方体分割算法从S1预处理得到的轨道激光点云中提取钢轨点云;根据钢轨标准断面图生成标准钢轨模型点云,使用主成分分析算法和广义迭代最近点算法实现钢轨点云与钢轨模型点云的精确配准,从而获取钢轨空间几何参数,提取轨顶中线,计算线路中点,进而拟合出线路中心线;利用提取的轨顶中线更新轨迹线和钢轨点云之间的空间位置参数,进行下一段钢轨点云的提取与配准;迭代重复以上步骤,完成全部钢轨点云的提取和配准;
    S3,接触线和悬挂点点云自动提取:
    接触线和悬挂点提取:根据轨迹线和接触网结构之间相对稳定的空间位置关系,基于S2中分段线性拟合得到的等间隔轨迹线段,采用圆柱分割算法从S1中得到的接触网激光点云中提取包含接触线和悬挂点的点云;采用直线的随机采样一致性算法对包含接触线和悬挂点的点云进行三维直线模型拟合,根据标准接触导线的横断面设计图设置随机采样一致性算法的参数,进行接触导线点云分段提取与三维直线模型的自动拟合;采用维度分析法,分析目标点邻域的维度特征,设置线性阈值提取接触线悬挂点;
    S4,接触网导高、拉出值自动检测:
    根据导高和拉出值的定义及测量方法,利用S2中钢轨几何参数的检测结果、S3中接触线和悬挂点的检测结果,对接触线导高和拉出值进行自动检测。
  2. 根据权利要求1所述的接触网导高与拉出值自动检测方法,其特征在于,步骤S1中所述点云数据采集与预处理包括以下步骤:
    (1)点云数据采集:
    随着车载移动激光扫描系统的移动,激光扫描仪在垂直于线路方向或与线路方向呈45度方向进行圆周扫描,得到被测轨道交通长度范围内及轨道两侧一定距离范围的激光点云;利用所述车载移动激光扫描系统的轨迹线对所扫描的数据进行解算,获得全线统一地理空间坐标系下的激光点云,将所述全线统一地理空间坐标系下的激光点云导出成通用数据格式;
    (2)激光点云角度阈值分割:
    根据扫描仪在采集数据过程中记录的扫描头旋转的角度信息、扫描头每次步进旋转的角度值以及扫描仪圆周扫描的角度起点分割出一定角度范围的激光点云;根据扫描仪的安装高度、安装角度和分割区域的几何尺度,通过空间几何分析,利用三角函数公式计算得到分割区域所在的角度范围,或者直接在激光点云上选择轨道或者接触网结构的边界点,查看边界点的扫描角度值;再设定一个扫描角度值的浮动值δ,根据公式(1),遍历所有激光点云,分割出轨道和接触网点云:
    Figure PCTCN2021124110-appb-100001
    式中:
    P i:第i个激光点;
    Segment RailTrack、Segment Catenary、Segment Other:依次为轨道点云、接触网点云、其它点云;
    α min、α max:分别为轨道对应的最小扫描角度和最大扫描角度,单位为度;
    β min、β max:分别为接触网结构对应的最小扫描角度和最大扫描角度,单位为度;
    δ:角度分割浮动值;
    Figure PCTCN2021124110-appb-100002
    是轨道点云中第i个激光点的扫描角度,单位为度;
    Figure PCTCN2021124110-appb-100003
    是接触网点云中第i个激光点的扫描角度,单位为度;
    优选的是,所述角度分割浮动值δ为3度;
    (4)点云噪声过滤:
    使用统计滤波算法剔除三维激光扫描系统获取的点云数据中的少量的噪声点。
  3. 根据权利要求2所述的接触网导高与拉出值自动检测方法,其特征在于:步骤(1)中,所述通用数据格式为LAS格式,扫描仪的旋转角度范围为-180°~180°,扫描仪正下方为扫描起始方向,旋转至扫描仪正上方为180°,轨迹线前进方向左侧角度为负值,右侧角度为正值。
  4. 根据权利要求2所述的接触线导高与拉出值自动检测方法,其特征在于:步骤(2)中,如果车载移动激光扫描系统含有多个扫描仪,则分别按照每个扫描仪角度分割激光点云,再将多个扫描仪分割的激光点云进行合并。
  5. 根据权利要求2所述的接触网导高与拉出值自动检测方法,其特征在于:步骤(1)中,所述轨道两侧一定距离范围为线路左右两侧各100米的宽度范围;
  6. 根据权利要求1所述的接触网导高与拉出值自动检测方法,其特征在于:步骤S2中钢轨点云的分割与配准步骤如下:
    (1)轨迹线分段线性拟合:首先建立轨道交通激光点云的多维空间索引结构;设定固定步长(Fixed Depth),分段线性拟合轨迹线,获得等间距的多段线;
    (2)钢轨点云分割与配准,首先,右侧钢轨的点云分割与配准包括以下步骤:
    1)轨迹线平移:
    选择检测起始点,利用距离判别法找到起始点对应的轨迹线段AB,将该轨迹线段AB平移到起始点的右侧钢轨顶面点R处,平移后的线段为A'B',
    2)长方体分割:
    利用钢轨几何尺寸以及平移后的轨迹线段为A'B',建立长方体包围盒,分割钢轨点云,以直线A'B'为轴,在A'B'长度范围内两侧各s宽度,建立竖直面作为长方体包围盒的左右侧面,所述s可取钢轨底面宽度的一半;设A'B'的最大高程为H max、最小高程为H min、钢轨的高度为H rail;设置一个高度浮动值ε,长方体包围盒顶面高程确定为H max+ε,底面高程为H min-H rail-ε,循环遍历轨道点云,分割长方体包围盒内的钢轨点云;优选的是,所述高度浮动值ε为0.01m;
    3)点云配准:
    根据检测线路使用钢轨的标准断面图,建立网格化三维标准钢轨模型并对其进行均匀采样,形成标准钢轨模型点云;首先使用粗配准算法使待配准的钢轨点云和标准钢轨模型点云大致对齐,再采用GICP算法实现钢轨点云的精配准,利用精配准后的钢轨模型点云进行钢轨的空间几何参数的计算,再以基准轨道为基础,分段计算线路中点,将这些分段提取的线路中点依次连接,形成线路中心线;
    4)几何参数更新:
    设经过配准后的钢轨模型点云顶面中线为A”B”,将B”替代R点作为新的检测起始点,重新计算轨迹线相对钢轨的几何关系;
    5)循环迭代:
    循环进行步骤1)~4),将新的几何参数用于下一段钢轨点云的提取和配准,直至完成所有钢轨点云的提取与配准;
    按上述步骤1)-5)对左侧钢轨的点云进行分割与配准。
  7. 根据权利要求6所述的接触网导高与拉出值自动检测方法,其特征在于:步骤(1)中,所述轨道交通激光点云的多维空间索引结构为KD树。
  8. 根据权利要求6所述的接触网导高与拉出值自动检测方法,其特征在于:步骤3)中,利用所述配准后的钢轨三维几何模型进行钢轨几何参数的计算,得到轨距、轨向信息,根据基准轨道的定义确定基准轨道后,再以基准轨道为基础,向另外一根钢轨水平偏移1/2标准轨距,分段计算线路中点,再将这些分段提取的线路中点依次连接,形成线路中心线,将所形成的线路中心线再进行平滑处理,得到相对平滑的线路中心线。
  9. 根据权利要求1所述的接触网导高与拉出值自动检测方法,其特征在于:步骤S3中接触线和悬挂点点云自动提取的步骤如下:
    (1)接触线和悬挂点点云分割:
    对于角度分割后的接触网结构,使用圆柱分割算法分割接触线和悬挂点点云,首先,选择接触线点C,获得其三维坐标;其次,采用距离判别法,获取C点对应的轨迹线段MN,将轨迹线段MN平移到C处,平移后的线段为M'N';最后,以M'N'为轴,半径r建立圆柱,遍历接触网结构点云,分割得到接触线和悬挂点点云;接触线沿着线路方向呈“之”字形布设,拉出值在定位器处为极大值;普速铁路和高速铁路接触线拉出值的限界值为450mm,按照拉出值的极限条件下考虑,圆柱半径r可按式(2)计算:
    Figure PCTCN2021124110-appb-100004
    式中:L为接触网立柱间距,一般为50m;
    优选的是,所述半径r为0.1m;
    (2)接触线提取:
    因接触线截面积较小且距离测量系统较远,当扫描作业速度过快时,会导致接触线点云稀少,出现一个接触线断面只有一个点的情况,故采用随机采样一致性的直线模型;将接触线的实际截面直径(高度)设置为直线采样的距离阈值,线性采样获取接触线点云和直线模型起终点M”N”;
    (3)悬挂点提取:
    采用主成分分析(PCA)解算出目标点邻域点云集合对应的特征向量和特征值,然后根据特征值λ 1、λ 2、λ 3123)判断点云维度特征;在只包含接触线的区域,点云呈线性特征;在悬挂点处,吊弦和定位器与接触线相连,点云的维度特征表现为非线性;遍历圆柱分割后的点云,计算目标点半径r邻域点云的线性特性值,设置线性特征值阈值L λ,提取悬挂点;对于悬挂点周围点云,可以根据与λ 1对应的特征向量判断方向,与地面垂直的点为吊弦悬挂点,近似平行的为定位悬挂点,优选的是,所述线性特征值阈值L λ为0.9;
    (4)循环迭代:
    将直线模型终点N”点替代C点,作为新的检测起始点,重复(1)~(3)步,直至完成所有接触线和悬挂点的提取。
  10. 根据权利要求1所述的接触网导高与拉出值自动检测方法,其特征在于:步骤S4中接触网导高与拉出值检测方法如下:
    (1)导高计算:
    根据悬挂点和接触线的分割结果,提取悬挂点处接触线底部高程,找到对应位置的轨面高程,求差后获取接触线导高值;
    (2)拉出值计算:
    对于直线地段,不存在外轨超高,直接计算定位悬挂点处接触线底部点至线路中心线的 XY平面的偏移量;对于曲线地段,由于存在外轨超高,接触线的拉出值按式(3)和(4)计算:
    a=m+c  (3)
    Figure PCTCN2021124110-appb-100005
    式中:a:接触网拉出值;
    m:定位点接触线与线路中心线的水平距离;
    c:定位点受电弓中心与线路中心线的水平距离;
    h:外轨超高;
    H:接触线高度;
    L:轨距,
    优选的是,所述对应位置的轨面的寻找方法为距离判别法;
    所述外轨超高步骤S2得到的钢轨空间几何参数计算。
PCT/CN2021/124110 2021-08-19 2021-10-15 基于车载移动激光点云的接触网导高与拉出值自动检测方法 WO2023019709A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110955517.6A CN113870123B (zh) 2021-08-19 2021-08-19 基于车载移动激光点云的接触网导高与拉出值自动检测方法
CN202110955517.6 2021-08-19

Publications (1)

Publication Number Publication Date
WO2023019709A1 true WO2023019709A1 (zh) 2023-02-23

Family

ID=78990708

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/124110 WO2023019709A1 (zh) 2021-08-19 2021-10-15 基于车载移动激光点云的接触网导高与拉出值自动检测方法

Country Status (2)

Country Link
CN (1) CN113870123B (zh)
WO (1) WO2023019709A1 (zh)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116147525A (zh) * 2023-04-17 2023-05-23 南京理工大学 一种基于改进icp算法的受电弓轮廓检测方法及系统
CN116385684A (zh) * 2023-04-10 2023-07-04 长江水利委员会水文局 一种基于库岸点云的河道断面自动提取方法
CN116452467A (zh) * 2023-06-16 2023-07-18 山东曙岳车辆有限公司 基于激光数据的集装箱实时定位方法
CN116543037A (zh) * 2023-07-05 2023-08-04 中国铁路设计集团有限公司 Crtsiii型无砟轨道板承轨台中心提取方法
CN116892912A (zh) * 2023-09-11 2023-10-17 中国铁路设计集团有限公司 Crtsiii型无砟轨道板承轨台快速精测精调方法
CN117162098A (zh) * 2023-10-07 2023-12-05 合肥市普适数孪科技有限公司 一种狭小空间机器人姿态自主规划系统及方法
CN117576144A (zh) * 2024-01-15 2024-02-20 湖北工业大学 一种激光点云电力线提取方法、装置和电子设备

Families Citing this family (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114440767A (zh) * 2022-01-26 2022-05-06 株洲时代电子技术有限公司 铁路综合巡检方法
CN114782925B (zh) * 2022-06-17 2022-09-02 四川省公路规划勘察设计研究院有限公司 一种基于车载lidar数据的高速公路护栏矢量化方法及设备
CN115546266B (zh) * 2022-11-24 2023-03-17 中国铁路设计集团有限公司 基于局部法向相关的多条带机载激光点云配准方法
CN115657067B (zh) * 2022-12-26 2023-03-14 中国铁路设计集团有限公司 多尺度阵列式激光雷达测量系统及获取轨道边界条件方法
CN116128835B (zh) * 2023-02-02 2023-09-15 南方电网数字电网研究院有限公司 基于点云分析的输电导线弧垂测量方法、装置和设备
CN115824070B (zh) * 2023-02-23 2023-05-30 湖南睿图智能科技有限公司 一种钢轨件尺寸测量方法及装置
CN116486012B (zh) * 2023-04-27 2024-01-23 中国民用航空总局第二研究所 一种航空器三维模型构建方法、存储介质及电子设备
CN116862976B (zh) * 2023-06-08 2024-04-02 中铁第四勘察设计院集团有限公司 一种基于无人机激光点云的铁轨中心线提取方法及系统

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103852011A (zh) * 2014-03-20 2014-06-11 北京天格高通科技有限公司 基于激光雷达的铁路接触网几何参数分析方法
WO2016118672A2 (en) * 2015-01-20 2016-07-28 Solfice Research, Inc. Real time machine vision and point-cloud analysis for remote sensing and vehicle control
CN108564575A (zh) * 2018-04-11 2018-09-21 西南交通大学 一种基于三维点云数据的非接触式接触网参数检测方法
CN110647798A (zh) * 2019-08-05 2020-01-03 中国铁路设计集团有限公司 基于车载移动激光点云的轨道中线自动检测方法
CN112414309A (zh) * 2020-11-25 2021-02-26 北京交通大学 基于机载激光雷达的高铁接触线导高及拉出值巡检方法
US20210146970A1 (en) * 2018-06-01 2021-05-20 Plasser & Theurer Export Von Bahnbaumaschinen Gmbh Method for determining an actual position of rails of a track

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103489218B (zh) * 2013-09-17 2016-06-29 中国科学院深圳先进技术研究院 点云数据质量自动优化方法及系统
EP4079597A1 (en) * 2019-12-17 2022-10-26 Telefonos, Lineas y Centrales, S.A. Method for in-situ and real-time collection and processing of geometric parameters of railway lines

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103852011A (zh) * 2014-03-20 2014-06-11 北京天格高通科技有限公司 基于激光雷达的铁路接触网几何参数分析方法
WO2016118672A2 (en) * 2015-01-20 2016-07-28 Solfice Research, Inc. Real time machine vision and point-cloud analysis for remote sensing and vehicle control
CN108564575A (zh) * 2018-04-11 2018-09-21 西南交通大学 一种基于三维点云数据的非接触式接触网参数检测方法
US20210146970A1 (en) * 2018-06-01 2021-05-20 Plasser & Theurer Export Von Bahnbaumaschinen Gmbh Method for determining an actual position of rails of a track
CN110647798A (zh) * 2019-08-05 2020-01-03 中国铁路设计集团有限公司 基于车载移动激光点云的轨道中线自动检测方法
CN112414309A (zh) * 2020-11-25 2021-02-26 北京交通大学 基于机载激光雷达的高铁接触线导高及拉出值巡检方法

Cited By (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116385684A (zh) * 2023-04-10 2023-07-04 长江水利委员会水文局 一种基于库岸点云的河道断面自动提取方法
CN116385684B (zh) * 2023-04-10 2023-11-03 长江水利委员会水文局 一种基于库岸点云的河道断面自动提取方法
CN116147525B (zh) * 2023-04-17 2023-07-04 南京理工大学 一种基于改进icp算法的受电弓轮廓检测方法及系统
CN116147525A (zh) * 2023-04-17 2023-05-23 南京理工大学 一种基于改进icp算法的受电弓轮廓检测方法及系统
CN116452467A (zh) * 2023-06-16 2023-07-18 山东曙岳车辆有限公司 基于激光数据的集装箱实时定位方法
CN116452467B (zh) * 2023-06-16 2023-09-22 山东曙岳车辆有限公司 基于激光数据的集装箱实时定位方法
CN116543037B (zh) * 2023-07-05 2024-03-26 中国铁路设计集团有限公司 Crtsiii型无砟轨道板承轨台中心提取方法
CN116543037A (zh) * 2023-07-05 2023-08-04 中国铁路设计集团有限公司 Crtsiii型无砟轨道板承轨台中心提取方法
CN116892912A (zh) * 2023-09-11 2023-10-17 中国铁路设计集团有限公司 Crtsiii型无砟轨道板承轨台快速精测精调方法
CN116892912B (zh) * 2023-09-11 2023-12-22 中国铁路设计集团有限公司 Crtsiii型无砟轨道板承轨台快速精测精调方法
CN117162098A (zh) * 2023-10-07 2023-12-05 合肥市普适数孪科技有限公司 一种狭小空间机器人姿态自主规划系统及方法
CN117162098B (zh) * 2023-10-07 2024-05-03 合肥市普适数孪科技有限公司 一种狭小空间机器人姿态自主规划系统及方法
CN117576144A (zh) * 2024-01-15 2024-02-20 湖北工业大学 一种激光点云电力线提取方法、装置和电子设备
CN117576144B (zh) * 2024-01-15 2024-03-29 湖北工业大学 一种激光点云电力线提取方法、装置和电子设备

Also Published As

Publication number Publication date
CN113870123B (zh) 2023-01-03
CN113870123A (zh) 2021-12-31

Similar Documents

Publication Publication Date Title
WO2023019709A1 (zh) 基于车载移动激光点云的接触网导高与拉出值自动检测方法
CN110647798B (zh) 基于车载移动激光点云的轨道中线自动检测方法
Liu et al. A review of applications of visual inspection technology based on image processing in the railway industry
CN111830528A (zh) 基于激光点云的杆塔特征点自动识别及倾斜参数自动测量方法
CN109855562B (zh) 钢轨磨耗测量方法及装置
CN111547084B (zh) 基于自移动式轨道交通移动三维扫描系统的数据处理方法
CN108362308B (zh) 一种利用隧道环缝的移动激光测量系统里程校正方法
CN110986878A (zh) 基于移动测量系统自动化提取铁轨断面的方法
Geng et al. UAV-LiDAR-based measuring framework for height and stagger of high-speed railway contact wire
CN102914290A (zh) 地铁限界检测系统及其检测方法
CN114719884A (zh) 一种惯导系统姿态测量精度评估方法及应用
Du et al. Dislocation detection of shield tunnel based on dense cross-sectional point clouds
CN110986877A (zh) 基于高精度车载激光移动测量系统铁路工程限界检测方法
CN112964195B (zh) 基于激光三角法的供电轨几何参数综合检测方法及系统
Neubert et al. Extraction of railroad objects from very high resolution helicopter-borne LiDAR and ortho-image data
CN113673011A (zh) 基于点云数据智能识别运营期隧道侵界的方法
CN114046747A (zh) 一种基于移动点云数据的智能分析盾构隧道环间错台量方法
CN114935308B (zh) 一种基于曲线配准算法的列车受电弓磨耗检测方法
CN112581521B (zh) 一种磁浮轨道中心线提取方法
CN114897777A (zh) 顾及空间关系的接触网支撑设施激光点云全自动提取方法
Jeon et al. Automated reconstruction of railroad rail using helicopter-borne light detection and ranging in a train station
CN117848238B (zh) 一种基于激光点云的轨道交通站台实际限界测量方法
Han et al. Automated Extraction of Rail Point Clouds by Multi-Scale Dimensional Features From MLS Data
Wu et al. Track gauge measurement based on model matching using UAV image
Guo et al. Frog and blade based branching direction detection in LiDAR data

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21953945

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE