Track non-contact intelligent detection system and method based on multi-source data fusion

Abstract

The invention discloses a track non-contact intelligent detection system and method based on multi-source data fusion, relates to the technical field of track detection, and aims at efficiently completing track detection with low cost. The system comprises a frame body, a plurality of line structured light 3D cameras, a synchronous controller, a processing computer and a speedometer; the linear structured light 3D cameras are arranged on the two sides of the frame body, and the optical axes of the linear structured light 3D cameras are arranged to be perpendicular to the axis direction of the steel rail to be measured; the synchronous controller, the processing computer and the speedometer are integrated at the bottom of the frame body, and the synchronous controller and the speedometer are respectively connected with the structured light 3D camera through cables.

Description

Technical Field

[0001] This invention relates to the field of track inspection technology, and in particular to a non-contact intelligent track inspection system and method based on multi-source data fusion. Background Technology

[0002] As a core infrastructure of railway transportation, the technical condition of railway tracks directly affects the safety, economy, and comfort of train operation. With the increase in railway mileage and train speeds, the requirements for the accuracy and efficiency of track inspection are constantly rising. Current track inspection technologies have the following limitations: Traditional contact-type track inspection instruments rely on mechanical contact measurement, with a detection speed of less than 8km / h, making it difficult to complete large-area inspections within the track maintenance window; the equipment has a high wear rate, requiring replacement of the contact wheel every 100km of inspection, increasing maintenance costs; they cannot pass through complex areas such as turnouts and small-radius curves, resulting in blind spots; and they are affected by the environment, leading to large errors in track gauge measurement and insufficient accuracy.

[0003] Non-contact large track inspection vehicle: The cost of a single unit is high, making it difficult to popularize at the grassroots level; in addition, the non-contact large track inspection vehicle integrates nearly a hundred sensors, requiring professional personnel to operate it, and the training cycle is long; it is only suitable for periodic inspections on main lines, and cannot meet the daily fine inspection needs of stations, branch lines and other areas; the sampling frequency is low during high-speed inspection, making it difficult to capture local defects within 3mm.

[0004] Existing technologies suffer from a trade-off between efficiency, cost, and accuracy, and there is an urgent need for a detection technology that balances high efficiency, low cost, high accuracy, and adaptability to complex environments. Summary of the Invention

[0005] In view of this, this application provides a non-contact intelligent track detection system and method based on multi-source data fusion, which can efficiently and cost-effectively complete the detection of tracks.

[0006] The first aspect of this application provides a non-contact intelligent rail inspection system based on multi-source data fusion. The system includes: a vehicle frame body, multiple line-structured light 3D cameras, a synchronization controller, a processing computer, and an odometer; the line-structured light 3D cameras are disposed on both sides of the vehicle frame body, and the optical axes of the line-structured light 3D cameras are set perpendicular to the axis of the rail to be measured; the synchronization controller, processing computer, and odometer are integrated at the bottom of the vehicle frame body, and the synchronization controller and odometer are respectively connected to the line-structured light 3D cameras via cables; wherein... The synchronization controller is used to output a single trigger pulse to the line structured light 3D camera when the coded mileage output by the odometer first reaches or exceeds the next untriggered mileage point in the predetermined mileage sequence; the mileage points are generated at fixed spatial intervals. The line structured light 3D camera is used to synchronously acquire the inner profile data of the rail to be measured; The processing computer is used to preprocess the inner profile data and extract specific three-dimensional point cloud data of the corresponding rail waist, rail bottom and rail head regions through clustering and segmentation. The processing computer is also used to match different weighting coefficients to the specific three-dimensional point cloud data corresponding to the rail waist, the specific three-dimensional point cloud data corresponding to the rail bottom, and the specific three-dimensional point cloud data corresponding to the rail head, so as to obtain the profile data after weight matching. The processing computer is also used to calculate the vertical wear and lateral wear of the rail to be measured based on the profile data after matching the weighting coefficients.

[0007] A second aspect of this application provides a non-contact intelligent detection method for tracks based on multi-source data fusion, the method comprising: The synchronization controller outputs a single trigger pulse to the line structured light 3D camera when the coded mileage output by the odometer first reaches or exceeds the next untriggered mileage point in the predetermined mileage sequence; the mileage points are generated at fixed spatial intervals. The inner profile data of the rail to be measured is simultaneously acquired using the line structured light 3D camera. The processing computer preprocesses the inner profile data and extracts specific three-dimensional point cloud data of the corresponding rail waist, rail bottom and rail head regions through clustering and segmentation. The processing computer matches different weighting coefficients to the specific three-dimensional point cloud data corresponding to the rail waist, the specific three-dimensional point cloud data corresponding to the rail bottom, and the specific three-dimensional point cloud data corresponding to the rail head, and obtains the profile data after weight matching. The computer uses the processing to calculate the vertical and lateral wear of the rail under test based on the profile data after matching the weighting coefficients.

[0008] Compared with existing technologies, this application provides a non-contact intelligent track detection system and method based on multi-source data fusion. Its advantages are as follows: The non-contact intelligent track detection system based on multi-source data fusion in this embodiment includes a chassis body, which is a lightweight mobile trolley that can be positioned in the middle of the rail to be measured. Multiple line structured light 3D cameras, a synchronous controller, a processing computer, and an odometer collect multi-source data. The line structured light 3D cameras collect the inner profile data of the rail to be measured, the laser inertial navigation unit collects the motion attitude of the chassis body, the odometer collects the relative mileage data of the chassis body compared to the previous moment, and the GNSS positioning module collects the real-time positioning of the chassis body. The data collected by the above devices are converted to a unified coordinate system, and the multi-source data collected by each device is aligned with the GPS timestamp of the GNSS positioning module as a reference. By correcting the profile data, the accuracy of the profile data is increased, thereby accurately calculating wear. A Kalman filter algorithm is used for integrated navigation to generate the motion trajectory of the vehicle frame in the geodetic coordinate system. This trajectory is projected onto the profile data to obtain the trajectory's representation in the profile data, i.e., the trajectory representing the vertex transformation of the rail. Track data is then extracted from the trajectory calculation. This process solves the problems of low efficiency, rapid wear, and narrow applicability of traditional contact-type equipment, as well as the high cost and complex operation of large-scale non-contact equipment. It has advantages such as high detection accuracy, high speed, low cost, and adaptability to complex environments, and can meet the refined inspection needs of main lines, stations, and branch lines. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0010] Figure 1 This is a schematic diagram of the device relationships in the non-contact intelligent track detection system based on multi-source data fusion proposed in the embodiments of this application; Figure 2 This is a schematic diagram of the chassis of a track non-contact intelligent detection system based on multi-source data fusion; Figure 3 This is an information flow diagram of the track non-contact intelligent detection system based on multi-source data fusion proposed in the embodiments of this application during operation; Figure 4 This is a schematic diagram of the track orientation in one example of this application; Figure 5 This is a schematic diagram of the rail track cross-section in one example of this application. Detailed Implementation

[0011] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0012] In this application, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0013] Example 1 In view of the problems described in the technical background, this application provides a non-contact intelligent detection method for tracks based on multi-source data fusion, which is applied to a non-contact intelligent detection system for tracks based on multi-source data fusion, so as to complete the detection of track parameters efficiently and at low cost.

[0014] Figure 1 This is a schematic diagram of the equipment relationships in the non-contact intelligent track detection system based on multi-source data fusion proposed in the embodiments of this application. Figure 1 The diagram shows the connection relationships of the devices in the track non-contact intelligent detection system, which are independent of the device's installation location. Figure 2 This is a schematic diagram of the chassis of a track non-contact intelligent detection system based on multi-source data fusion. (Reference) Figure 1 and Figure 2 The multi-source data fusion-based non-contact intelligent track inspection system includes: a vehicle frame body, multiple line structured light 3D cameras 1, a laser inertial navigation unit 2, and a GNSS positioning module 3; the wheels of the vehicle frame body coincide with the rails, and it also includes a synchronization controller, a processing computer, and an odometer; (See reference...) Figure 2 The line structured light 3D camera is set on both sides of the vehicle frame; in one example, the optical axis of the line structured light 3D camera is set to be perpendicular to the axis of the rail to be measured.

[0015] The system also includes a laser inertial navigation unit 2 and a GNSS positioning module 3; the laser inertial navigation unit and the GNSS positioning module are located in the middle of the vehicle frame body; the synchronization controller, the processing computer and the odometer are integrated at the bottom of the vehicle frame body 4, and the synchronization controller and the odometer are connected to the line structured light 3D camera, the laser inertial navigation unit, the GNSS positioning module and the processing computer respectively via cables.

[0016] Figure 3 This is an information flow diagram of the non-contact intelligent track detection system based on multi-source data fusion proposed in this application, referring to... Figure 3 This describes the process by which a multi-source data fusion-based non-contact intelligent detection system for tracks executes a non-contact intelligent detection method for tracks: The process of track detection, in which the aforementioned multiple line structured light 3D cameras, synchronization controllers, processing computers, odometers, laser inertial navigation units, and GNSS positioning modules cooperate with each other through data transmission, includes: S1: When the coded mileage output by the odometer first reaches or exceeds the next untriggered mileage point in the predetermined mileage sequence, the synchronization controller outputs a single trigger pulse to the line structured light 3D camera; the mileage points are generated at fixed spatial intervals.

[0017] For example, assuming a fixed spatial interval of ΔL, the synchronous controller generates mileage points when the odometer outputs the code 0, ΔL, 2ΔL, and 3ΔL respectively. Mileage points are generated when the odometer first reaches 0, and then when it exceeds ΔL, and so on.

[0018] In one example, the synchronization controller is equipped with a mechanism to prevent repeated triggering, ensuring that each mileage point is triggered only once.

[0019] Specifically, when the coded mileage output by the odometer reaches the current element in the predetermined mileage sequence, the synchronization controller outputs a trigger pulse to the line structured light 3D camera, the laser inertial navigation unit, and the GNSS positioning module, so that the line structured light 3D camera, the laser inertial navigation unit, and the GNSS positioning module can synchronously acquire data.

[0020] For example, the odometer outputs a trigger pulse every time the coded mileage reaches 0.1 mm of travel.

[0021] S2: Data is collected synchronously via a line structured light 3D camera, a laser inertial navigation unit, and a GNSS positioning module, including: S21: The inner profile data of the rail to be measured is synchronously acquired by the line structured light 3D camera.

[0022] S22: The motion attitude of the vehicle frame body is acquired through the laser inertial navigation unit.

[0023] The motion posture of the frame body is the motion state of the frame body on the track.

[0024] S23: Collect the relative mileage data of the vehicle frame body at the current moment and the previous moment through the odometer.

[0025] The laser inertial navigation unit outputs data at a frequency of 200Hz relative to the previous moment's mileage. The odometer outputs a pulse for every 1mm traveled by the vehicle frame, and the controller inside the odometer accumulates the pulses to obtain the mileage.

[0026] S24: The real-time positioning of the vehicle frame body is collected through the GNSS positioning module.

[0027] S3: Based on the inherent three-dimensional geometry of the rail profile calibration plate, the processing computer aligns the camera coordinate systems of the first line structured light 3D camera and the second line structured light 3D camera using an iterative nearest-point method; the first line structured light 3D camera is placed on a specific side of the vehicle frame body, and the second line structured light 3D camera is placed on the opposite side of the specific side; a transformation relationship between the camera coordinate system and the inertial navigation coordinate system is established, and the data collected by each line structured light 3D camera is transformed into the inertial navigation coordinate system.

[0028] For a specific rail profile calibration plate, based on its mechanical design, the three-dimensional geometry of the rail profile calibration plate is known and fixed.

[0029] This unifies the coordinate systems of all devices, providing a foundation for subsequent calculations. S3, based on the rail profile calibration plate, calculates the positional relationship between the same feature point on the rail and the carrier, thereby obtaining the transformation relationship between the line structured light 3D camera coordinate system and the inertial navigation coordinate system.

[0030] S4: The GNSS positioning module outputs a GPS timestamp to the laser inertial navigation unit and the line structured light 3D camera. After the laser inertial navigation unit collects data, it marks the collected data with a GPS timestamp.

[0031] Both the laser inertial navigation unit and the line structured light 3D camera use the timestamp of the GNSS positioning module to unify the time of data acquisition by each device.

[0032] S5: The processing computer uses the timestamp of the data collected by the GNSS positioning module as the reference time, and interpolates the reference time with the data corresponding to the GPS timestamp to obtain the data collected by the laser inertial navigation unit at the reference time, and aligns the data collected by the laser inertial navigation unit with the data collected by the GNSS positioning module.

[0033] By using the above data interpolation method, minor deviations in the data collection time of different devices can be corrected.

[0034] Steps S1 and S5 synchronize the data collected by each device in time and space. An odometer pulse signal, with each pulse corresponding to a 0.1mm travel distance, serves as the trigger source. The synchronization controller drives each device to sample, with time deviation controlled within ±50μs. Data is aligned using GPS time stamps. Minor deviations are corrected using linear interpolation.

[0035] For example, the process of performing S5 to correct small deviations using linear interpolation includes: Suppose the laser inertial navigation unit is at two time points , The data are as follows: , The interpolated data value for the laser inertial navigation unit at time t is calculated as follows: (2) Since different devices collect data at different frequencies, the above method is used to correct the slight deviations between the data collected by different devices due to the different collection frequencies.

[0036] S6: The processing computer preprocesses the inner profile data and extracts specific three-dimensional point cloud data of the corresponding rail waist, rail bottom and rail head regions through clustering and segmentation.

[0037] Clustering yields specific 3D point cloud data corresponding to the rail waist and rail head. The data corresponding to the rail waist, rail bottom, and rail head are distinguished in the 3D point cloud data, and specific 3D point cloud data corresponding to the rail bottom is extracted to provide data for subsequent fitting of the optimal circular arc model.

[0038] Because there may be debris (such as gravel, weeds, leaves, etc.) at the bottom of the track, the preprocessing process also uses a clustering algorithm to distinguish the bottom area of ​​the track. The bottom area of ​​the track is then smoothed or the adjacent unobstructed contours are used to remove the debris interference. For the missing obstructed areas, adjacent frame data is used to fill in the gaps.

[0039] S7: The processing computer matches different weighting coefficients to the specific three-dimensional point cloud data corresponding to the rail waist, the specific three-dimensional point cloud data corresponding to the rail bottom, and the specific three-dimensional point cloud data corresponding to the rail head, to obtain the profile data after weight matching.

[0040] The S7 process matches the processed profile with a standard template, matching the weights of the profile data corresponding to different locations of the rail to be measured. This process includes: S71: Using the RANSAC algorithm combined with the least squares method, a circular arc is fitted to the specific three-dimensional point cloud data corresponding to the track bottom, the optimal circular arc model is fitted, and the center coordinates of the circular arc model are extracted.

[0041] S72: Calculate the serpentine running angle of the chassis body based on the line connecting the center coordinates of the optimal circular arc model and the initial angle; obtain the serpentine trajectory by integrating the mileage information obtained from the odometer count; scan the profile with two line structured light 3D cameras set on one side of the chassis body, and extract the center of the circular arc at the bottom of the profile to obtain the preset center. The initial angle is the installation angle of the line structured light 3D cameras during mechanical design.

[0042] S73: Compensate for the spatial displacement caused by the serpentine motion of the chassis body in the profile data; at the same time, construct a standard reference plane aligned with the track design datum, project the compensated profile point cloud onto this plane, and align it with the standard rail design profile to obtain the geometrically corrected profile data.

[0043] The processed profile is matched with a standard template. First, the profile is initially matched with the template based on the center coordinates of the circle with a radius of 20mm at the rail base and the installation angle. Then, a fine match is performed using an improved ICP algorithm with a matching threshold of 0.05mm. The improved ICP algorithm matches the rail head and rail base by setting different matching weights (0.1 for rail head and 0.9 for rail base). To avoid the rail head wear affecting the matching results, a lower weight is set.

[0044] In one example, the process of calculating the angle and trajectory of the frame body's serpentine movement, generating a standard projection plane, and projecting the compensated profile data back to the standard profile dimensions to obtain the corrected profile data includes: The lateral sway angle of the vehicle is detected by two parallel 3D cameras (500mm apart), which dynamically compensate for the deviation of the profile data and correct the measurement error caused by the serpentine motion (amplitude ±5mm).

[0045] refer to Figure 2 Assume the distance L = 500mm between the line structured light 3D camera 11 and line structured light 3D camera 13 in front of and behind the crossbeam where the laser inertial navigation unit and GNSS positioning module are located, and the coordinates of the detection point of the left-side line structured light 3D camera 11 are... The coordinates of the right-side structured light 3D camera detection point 12 are: Then the lateral deviation Lateral swing angle The calculation is as follows: (3) (4) (5) in, To correct the coordinates of the point, The original point coordinates, To correct the rotation matrix. For example, to... Substitute into the above formulas (3), (4), and (5). = After being corrected .

[0046] By rotation matrix Profile data can be corrected.

[0047] S8: The processing computer calculates the vertical wear and side wear of the rail to be measured based on the profile data after matching the weighting coefficients.

[0048] Vertical and lateral wear can be calculated based on the matching results. Vertical wear is measured at the vertical centerline of the rail, and lateral wear is measured 14mm below the rail top. The total wear is calculated as vertical wear + 1 / 2 × lateral wear. The coordinates of the rail top midpoint P1 and the gauge measurement point P2 are also extracted.

[0049] The trajectory of the vehicle frame is acquired by a laser inertial navigation unit. The relative positional relationship between the vehicle frame and the rails is obtained by calculating the rail profile calibration plate using a line structured light camera. The rail trajectory is the trajectory curve of the rail top midpoint. The curve formed by the coordinates of the rail top midpoint P1 and the gauge measurement point P2 is mapped onto the inertial navigation coordinate system to obtain the true trajectory of the rail.

[0050] Specifically, the processing computer calculates the vertical wear and lateral wear of the rail to be measured based on the corrected profile data after matching the weighting coefficients.

[0051] This embodiment matches the processed profile data with a standard template, and calculates the vertical and lateral wear based on the matching results. Before wear calculation, alignment with the standard rail profile is required. This embodiment uses the following ICP algorithm to align the profile data with the standard rail profile: Create a collection containing profile data The set of data points obtained by the line structured light 3D camera scanning the rail profile calibration plate is obtained. The following formula is used to iteratively calculate the nearest point of the profile data, aligning the profile data with the rail profile calibration plate; P is the set of sampling points in the rail profile calibration plate. It is the first in the rail profile calibration plate During calibration, Q is the set of scan points obtained by the line structured light 3D camera scanning the rail profile calibration plate. During measurement, Q is the set of scan points obtained by the line structured light 3D camera scanning the rail, and N is the number of sampling points or scan points in the set. The first part of the set obtained after scanning the steel rail with a line structured light 3D camera. One scan point; (1); in, For rotation matrix, It is a translation vector. It is the weight of data point i.

[0052] The above calculations will set and set The three-dimensional points are mapped one-to-one, and the output is calculated when the result of formula (1) is the minimum value. and The transformation relationship between the camera and the inertial navigation coordinate system is: rotation matrix R, translation vector T.

[0053] S9: The processing computer combines the relative mileage data, the real-time positioning, and the motion posture using a Kalman filter algorithm to generate the motion trajectory of the vehicle body in the geodetic coordinate system.

[0054] The process of executing S9 includes: System state vector Defined as: The system state vector is the input to the Kalman filter and represents the current state of the chassis.

[0055] Among them, the position of the vehicle frame body in the geodetic coordinate system , It refers to the value of the horizontal axis of the vehicle frame in the east-west direction in the geodetic coordinate system. It refers to the value of the horizontal axis of the vehicle frame in the north-south direction of the geodetic coordinate system. It refers to the value of the vertical axis of the chassis body in the geodetic coordinate system, and the speed of the chassis body. , It is the velocity component of the vehicle frame body in the east-west horizontal axis direction of the geodetic coordinate system. It is the velocity component of the vehicle frame body in the horizontal axis direction (north-south) of the geodetic coordinate system. It refers to the velocity component of the chassis body along the vertical axis in the Earth coordinate system, and the attitude of the chassis body. It is the rolling Euler angle, It is pitch Euler angle, Orientation to Euler angle, zero gyro deviation of the chassis body The accelerometer of the chassis body is zero biased. Equations of state for: in, Here is the state transition matrix. For noise driving matrix, This is process noise, which occurs during actual data acquisition. and It is coupled into the acquired data.

[0056] The Kalman filter iterative formula set is as follows: Step 1: Predict the current system state based on the system state at the previous moment. in, The prior state estimate includes the current position, velocity, attitude angle, and sensor bias (gyroscope, accelerometer). The prior covariance matrix reflects the uncertainty of the predicted state; It is the noise covariance matrix of the Kalman filter iteration process, corresponding to random disturbances in the physical processes not precisely described by the system model. Its core function is to quantify the uncertainty of the model prediction. Indicates the number of iterations.

[0057] Step 2: Observation (acquiring sensor data and building an observation model) in, For the observation matrix, To observe noise; The observation vector is the GNSS position. / speed Relative mileage to odometer Fusion value: Step 3: Update (used to correct the predicted trajectory based on the observed values) in, Kalman gain, representing the predicted value. Compared with observed values Feasibility. This is the posterior state estimate, i.e., the corrected trajectory parameters. The posterior covariance matrix reflects the uncertainty of the corrected state. It is the measurement noise covariance matrix.

[0058] After obtaining the motion trajectory of the vehicle frame in the geodetic coordinate system, the trajectory is mapped to the inertial navigation coordinate system. S10: The computer extracts data from the motion trajectory and calculates the track parameters of the rail to be measured.

[0059] The process of executing S10, in which the computer extracts data from the motion trajectory and calculates the track parameters of the rail to be measured, includes: The motion trajectory is mapped to the actual trajectories of the left and right rail apexes of the rail to be measured by the rotation matrix and translation vector of the line structured light 3D camera. The actual trajectory of the right rail apex of the rail to be measured includes the trajectory of the left rail and the trajectory of the right rail. Figure 4 This is a schematic diagram of the trajectory orientation of one example of the trajectory in this application, such as... Figure 4 As shown, put Figure 4 Replace the coordinates in () ), ( ), ( This is an example of a trajectory elevation diagram. The process of calculating the trajectory direction and elevation using the midpoint chord method includes: extracting the top coordinates of the actual trajectory and mapping them to the inertial navigation coordinate system to obtain p1( p2( p3 p2 is the midpoint between p1 and p3 along the y-axis. (Track direction) The calculation formula is: High and low The calculation formula is: Calculate the horizontal and superelevation H using the elevation difference between the left and right rail tops: in, The standard rail top midpoint spacing is 1500mm, and θ is the rail surface inclination angle, i.e., the trolley roll attitude angle.

[0060] Calculation of the triangular pit T based on the algebraic difference of the horizontal difference within the base length: in, Figure 5 This is a schematic diagram of the rail cross-section in one example of this application, such as... Figure 5 As shown, This is the horizontal (superelevation) measurement value at section 1 of the rail track; The measurement value of the horizontal (superelevation) at point 2 of the rail section.

[0061] Tangents are drawn at the vertices of the left and right rail tracks of the rail to be measured, respectively, to obtain the first and second tangents; Shift the first and second tangents downward by 16mm under the same cross section, and calculate the coordinates of the intersection point of the left rail profile data and the right rail profile data of the rail to be measured. The Euclidean distance is calculated based on the coordinates of the intersection of the left and right rail profile data of the rail to be measured, and the rail gauge value is obtained. Shift the line connecting the vertices of the left and right rails (tangent) downwards by 16mm, calculate the Euclidean distance from the intersection point of the left and right profiles, which is the rail gauge value. The coordinates of the structured light 3D camera on the left side in the inertial navigation coordinate system are: The coordinates of the right-side structured light 3D camera in the inertial navigation coordinate system are: Track gauge value for: This invention provides a non-contact intelligent track inspection system based on multi-source data fusion. The system includes a chassis body, a lightweight mobile trolley that can be positioned in the middle of the rail to be measured. Multiple line-structured light 3D cameras, a synchronous controller, a processing computer, and an odometer collect multi-source data. The line-structured light 3D cameras collect the inner profile data of the rail to be measured. A laser inertial navigation unit collects the motion attitude of the chassis body. The odometer collects the relative mileage data of the chassis body compared to the previous moment. A GNSS positioning module collects the real-time positioning of the chassis body. The data collected by each device is converted to a unified coordinate system, and the multi-source data collected by each device is aligned using the GPS timestamp of the GNSS positioning module as a reference. By correcting the profile data, the accuracy of the profile data is increased, thereby accurately calculating wear. A Kalman filter algorithm is used for combined navigation to generate the motion trajectory of the chassis body in the geodetic coordinate system. The motion trajectory is projected onto the profile data to obtain the trajectory's representation in the profile data, i.e., the trajectory representing the vertex transformation of the rail. Track data is extracted from the trajectory calculation. The above process solves the problems of low efficiency, rapid wear and tear, and narrow applicability of traditional contact equipment, as well as high cost and complex operation of non-contact large equipment. It has advantages such as high detection accuracy, fast speed, low cost and adaptability to complex environments, and can meet the refined inspection needs of trunk lines, stations, branch lines and other scenarios.

[0062] Example 2 Based on the multi-source data fusion-based non-contact intelligent track detection method provided in Embodiment 1 of this application, correspondingly, Embodiment 2 of this application also provides a multi-source data fusion-based non-contact intelligent track detection system, which includes: The system includes: a chassis body, multiple line structured light 3D cameras, a synchronization controller, a processing computer, and an odometer; the line structured light 3D cameras are located on both sides of the chassis body, and the optical axes of the line structured light 3D cameras are set perpendicular to the axis of the rail to be measured; the synchronization controller, processing computer, and odometer are integrated at the bottom of the chassis body, and the synchronization controller and odometer are connected to the line structured light 3D cameras respectively via cables; wherein, The synchronization controller is used to output a single trigger pulse to the line structured light 3D camera when the coded mileage output by the odometer first reaches or exceeds the next untriggered mileage point in the predetermined mileage sequence; the mileage points are generated at fixed spatial intervals. The line structured light 3D camera is used to synchronously acquire the inner profile data of the rail to be measured; The processing computer is used to preprocess the inner profile data and extract specific three-dimensional point cloud data of the corresponding rail waist, rail bottom and rail head regions through clustering and segmentation. The processing computer is also used to match different weighting coefficients to the specific three-dimensional point cloud data corresponding to the rail waist, the specific three-dimensional point cloud data corresponding to the rail bottom, and the specific three-dimensional point cloud data corresponding to the rail head, so as to obtain the profile data after weight matching. The processing computer is also used to calculate the vertical wear and lateral wear of the rail to be measured based on the profile data after matching the weighting coefficients.

[0063] Optionally, the system further includes a laser inertial navigation unit and a GNSS positioning module; the laser inertial navigation unit and the GNSS positioning module are located in the middle of the vehicle frame body; a synchronization controller and an odometer are integrated at the bottom of the vehicle frame body, and the synchronization controller and the odometer are connected to a line structured light 3D camera, the laser inertial navigation unit, the GNSS positioning module, and a processing computer via cables; wherein, The synchronization controller is specifically used to output a trigger pulse to the line structured light 3D camera, laser inertial navigation unit, and GNSS positioning module when the coded mileage output by the odometer reaches the current element in the predetermined mileage sequence. The laser inertial navigation unit is used to collect the motion attitude of the vehicle frame body; The odometer is used to collect the relative mileage data of the vehicle frame body and the previous moment; The GNSS positioning module is used to collect the real-time positioning of the vehicle frame body; The processing computer is used to combine the relative mileage data, the real-time positioning, and the motion attitude through a Kalman filter algorithm to generate the motion trajectory of the vehicle body in the geodetic coordinate system. The computer is used to extract data from the motion trajectory and calculate the track parameters of the rail to be measured.

[0064] Optionally, the processing computer is also used for Based on the inherent three-dimensional geometry of the rail profile calibration plate, the camera coordinate systems of the first line structured light 3D camera and the second line structured light 3D camera are aligned by iterating the nearest point; the first line structured light 3D camera is placed on a specific side of the vehicle frame body, and the second line structured light 3D camera is set on the opposite side of the specific side; Establish the transformation relationship between the camera coordinate system and the inertial navigation coordinate system, and transform the data acquired by each line structured light 3D camera to the inertial navigation coordinate system.

[0065] Optionally, the processing computer is also used for: Using RANSAC least squares, the specific 3D point cloud data corresponding to the bottom of the profile data is fitted into circular arc data, and the center coordinates of the circular arc data are extracted. The angle of the serpentine movement of the chassis body is calculated based on the angle between the line connecting the center coordinates and the initial angle, and the serpentine movement trajectory is obtained by integrating the mileage information obtained from the odometer count. The spatial displacement caused by the serpentine motion of the chassis body is compensated in the profile data; at the same time, a standard reference plane aligned with the track design benchmark is constructed, the compensated profile point cloud is projected onto the standard reference plane, and aligned with the profile data of the rail profile calibration plate to obtain the geometrically corrected profile data. The processing computer is also used to calculate the vertical wear and lateral wear of the rail to be measured based on the profile data after matching the weighting coefficients, including: The processing computer is also used to calculate the vertical wear and lateral wear of the rail to be measured based on the profile data after matching and correcting the weighting coefficients.

[0066] Optionally, the GNSS positioning module is also used to output a GPS timestamp to the laser inertial navigation unit and the synchronization controller; After acquiring data, the laser inertial navigation unit marks the acquired data with GPS timestamps and mileage pulse counts. The processing computer is further configured to use the timestamp of the data collected by the GNSS positioning module as the reference time, and to perform data interpolation on the reference time using the data corresponding to the GPS timestamp to obtain the data collected by the laser inertial navigation unit at the reference time, and to align the data collected by the laser inertial navigation unit with the data collected by the GNSS positioning module.

[0067] Optionally, the processing computer extracts data from the motion trajectory and attitude sequence to calculate the track parameters of the rail to be measured, including: The motion trajectory is mapped to the actual trajectories of the left and right rail apexes of the rail to be measured by the rotation matrix and translation vector of the line structured light 3D camera. The actual trajectory of the right rail apex of the rail to be measured includes the trajectory of the left rail and the trajectory of the right rail. The midpoint of the actual trajectory is extracted, and the trajectory direction and elevation are calculated using the midpoint chord method.

[0068] Optionally, the computer processing the data extracted from the motion trajectory and attitude sequence to calculate the track parameters of the rail to be measured also performs the following process: The horizontal and superelevation are calculated by the difference in elevation between the left and right rail tops, and the triangular pit is calculated based on the algebraic difference of the horizontal difference within the base length.

[0069] Optionally, the computer processing the data extracted from the motion trajectory and attitude sequence to calculate the track parameters of the rail to be measured also performs the following process: Tangents are drawn at the vertices of the left and right rail tracks of the rail to be measured, respectively, to obtain the first and second tangents; Shift the first and second tangents downward by 16mm under the same cross section, and calculate the coordinates of the intersection point of the left rail profile data and the right rail profile data of the rail to be measured. The Euclidean distance is calculated based on the coordinates of the intersection of the left and right rail profile data of the rail to be measured, thus obtaining the track gauge value.

[0070] Optionally, the processing computer is further configured to calculate the vertical wear and lateral wear of the rail to be measured based on the profile data after matching weighting coefficients, including the following process: Create a collection containing profile data The set of data points obtained by the line structured light 3D camera scanning the rail profile calibration plate is obtained. The following formula is used to iteratively calculate the nearest point of the profile data; P is the set of sampling points in the rail profile calibration plate. It is the first in the rail profile calibration plate During calibration, Q is the set of scan points obtained by the line structured light 3D camera scanning the rail profile calibration plate. During measurement, Q is the set of scan points obtained by the line structured light 3D camera scanning the rail, and N is the number of sampling points or scan points in the set. The first part of the set obtained after scanning the steel rail with a line structured light 3D camera. One scan point; (1); in, Let be a rotation matrix. It is a translation vector. It is the weight of data point i.

[0071] The specific principles and execution processes of each unit in the track non-contact intelligent detection system based on multi-source data fusion disclosed in Embodiment 2 of this application are the same as those in the track non-contact intelligent detection method based on multi-source data fusion disclosed in Embodiment 1 of this application. Please refer to the corresponding parts in the track non-contact intelligent detection method based on multi-source data fusion disclosed in Embodiment 1 of this application. They will not be repeated here.

[0072] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computing software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0073] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0074] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A non-contact intelligent track detection system based on multi-source data fusion, characterized in that, The system includes: a chassis body, multiple line structured light 3D cameras, a synchronization controller, a processing computer, and an odometer; the line structured light 3D cameras are located on both sides of the chassis body, and the optical axes of the line structured light 3D cameras are set perpendicular to the axis of the rail to be measured; the synchronization controller, processing computer, and odometer are integrated at the bottom of the chassis body, and the synchronization controller and odometer are connected to the line structured light 3D cameras respectively via cables; wherein, The synchronization controller is used to output a single trigger pulse to the line structured light 3D camera when the coded mileage output by the odometer first reaches or exceeds the next untriggered mileage point in the predetermined mileage sequence; the mileage points are generated at fixed spatial intervals. The line structured light 3D camera is used to synchronously acquire the inner profile data of the rail to be measured; The processing computer is used to preprocess the inner profile data and extract specific three-dimensional point cloud data of the corresponding rail waist, rail bottom and rail head regions through clustering and segmentation. The processing computer is also used to match different weighting coefficients to the specific three-dimensional point cloud data corresponding to the rail waist, the specific three-dimensional point cloud data corresponding to the rail bottom, and the specific three-dimensional point cloud data corresponding to the rail head, so as to obtain the profile data after weight matching. The processing computer is also used to calculate the vertical wear and lateral wear of the rail to be measured based on the profile data after matching the weighting coefficients.

2. The track non-contact intelligent detection system based on multi-source data fusion according to claim 1, characterized in that, The system also includes a laser inertial navigation unit and a GNSS positioning module; the laser inertial navigation unit and the GNSS positioning module are located in the middle of the vehicle frame body; the synchronization controller and the odometer are integrated at the bottom of the vehicle frame body, and the synchronization controller and the odometer are connected to the structured light 3D camera, the laser inertial navigation unit, the GNSS positioning module, and the processing computer via cables respectively; wherein, The synchronization controller is specifically used to output a trigger pulse to the line structured light 3D camera, laser inertial navigation unit, and GNSS positioning module when the coded mileage output by the odometer reaches the current element in the predetermined mileage sequence. The laser inertial navigation unit is used to collect the motion attitude of the vehicle frame body; The odometer is used to collect the relative mileage data of the vehicle frame body at the current moment and the previous moment; The GNSS positioning module is used to collect the real-time positioning of the vehicle frame body; The processing computer is used to combine the relative mileage data, the real-time positioning, and the motion attitude through a Kalman filter algorithm to generate the motion trajectory of the vehicle body in the geodetic coordinate system. The computer is used to extract data from the motion trajectory and calculate the track parameters of the rail to be measured.

3. The track non-contact intelligent detection system based on multi-source data fusion according to claim 1, characterized in that, The processing computer is also used for, Based on the inherent three-dimensional geometry of the rail profile calibration plate, the camera coordinate systems of the first line structured light 3D camera and the second line structured light 3D camera are aligned using an iterative nearest point method; the first line structured light 3D camera is placed on a specific side of the vehicle frame body, and the second line structured light 3D camera is set on the opposite side of the specific side; Establish the transformation relationship between the camera coordinate system and the inertial navigation coordinate system, and transform the data acquired by each line structured light 3D camera to the inertial navigation coordinate system.

4. The track non-contact intelligent detection system based on multi-source data fusion according to claim 1, characterized in that, Computers are also used for: The RANSAC algorithm combined with the least squares method is used to fit the circular arc to the specific three-dimensional point cloud data corresponding to the track bottom, to fit the optimal circular arc model, and to extract the center coordinates of the optimal circular arc model. The angle of the serpentine movement of the chassis body is calculated based on the angle between the line connecting the center coordinates of the optimal circular arc model and the initial angle, and the serpentine movement trajectory is obtained by integrating the mileage information obtained from the odometer count. The spatial displacement caused by the serpentine motion of the chassis body is compensated in the profile data; at the same time, a standard reference plane aligned with the track design benchmark is constructed, the compensated profile point cloud is projected onto the standard reference plane, and aligned with the profile data of the rail profile calibration plate to obtain the geometrically corrected profile data. The processing computer is also used to calculate the vertical wear and lateral wear of the rail to be measured based on the profile data after matching the weighting coefficients, including: The processing computer is also used to calculate the vertical wear and lateral wear of the rail to be measured based on the profile data after matching and correcting the weighting coefficients.

5. The non-contact intelligent track detection system based on multi-source data fusion according to claim 2, characterized in that, The GNSS positioning module is also used to output GPS timestamps to the laser inertial navigation unit and the synchronization controller; After acquiring data, the laser inertial navigation unit marks the acquired data with GPS timestamps and mileage pulse counts. The processing computer is further configured to use the timestamp of the data collected by the GNSS positioning module as the reference time, and to perform data interpolation on the reference time using the data corresponding to the GPS timestamp to obtain the data collected by the laser inertial navigation unit at the reference time, and to align the data collected by the laser inertial navigation unit with the data collected by the GNSS positioning module.

6. The track non-contact intelligent detection system based on multi-source data fusion according to claim 2, characterized in that, The computer processes data extracted from the motion trajectory and attitude sequence to calculate the track parameters of the rail to be measured, including: The motion trajectory is mapped to the actual trajectories of the left and right rail apexes of the rail to be measured by the rotation matrix and translation vector of the line structured light 3D camera. The actual trajectory of the right rail apex of the rail to be measured includes the trajectory of the left rail and the trajectory of the right rail. The midpoint of the actual trajectory is extracted, and the trajectory direction and elevation are calculated using the midpoint chord method.

7. The track non-contact intelligent detection system based on multi-source data fusion according to claim 6, characterized in that, The computer, used to extract data from the motion trajectory and attitude sequence to calculate the track parameters of the rail to be measured, also performs the following process: The horizontal and superelevation are calculated by the difference in elevation between the left and right rail tops, and the triangular pit is calculated based on the algebraic difference of the horizontal difference within the base length.

8. The track non-contact intelligent detection system based on multi-source data fusion according to claim 6, characterized in that, The computer, used to extract data from the motion trajectory and attitude sequence to calculate the track parameters of the rail to be measured, also performs the following process: Tangents are drawn at the vertices of the left and right rail tracks of the rail to be measured, respectively, to obtain the first and second tangents; Shift the first and second tangents downward by 16mm under the same cross section, and calculate the coordinates of the intersection point of the left rail profile data and the right rail profile data of the rail to be measured. The Euclidean distance is calculated based on the coordinates of the intersection of the left and right rail profile data of the rail to be measured, thus obtaining the track gauge value.

9. The non-contact intelligent track detection system based on multi-source data fusion according to claim 3, characterized in that, The processing computer is also used to calculate the vertical wear and lateral wear of the rail to be measured based on the profile data after matching the weighting coefficients. The process includes: Create a collection containing profile data The set of data points obtained by the line structured light 3D camera scanning the rail profile calibration plate is obtained. The following formula is used to iteratively calculate the nearest point of the profile data; （1）； in, Let be a rotation matrix. It is a translation vector. Here, P is the weight of data point i, and P is the set of sampling points in the rail profile calibration plate. It is the first in the rail profile calibration plate During calibration, Q is the set of scan points obtained by the line structured light 3D camera scanning the rail profile calibration plate. During measurement, Q is the set of scan points obtained by the line structured light 3D camera scanning the rail, and N is the number of sampling points or scan points in the set. The first part of the set obtained after scanning the steel rail with a line structured light 3D camera. One scan point.

10. A non-contact intelligent detection method for tracks based on multi-source data fusion, characterized in that, An application is made to a non-contact intelligent rail inspection system based on multi-source data fusion. The system includes: a vehicle frame, multiple line structured light 3D cameras, a synchronization controller, a processing computer, and an odometer; the line structured light 3D cameras are positioned on both sides of the vehicle frame, with their optical axes perpendicular to the axis of the rail to be measured; the synchronization controller, processing computer, and odometer are integrated at the bottom of the vehicle frame, and the synchronization controller and odometer are connected to the line structured light 3D cameras via cables; the method includes: The synchronization controller outputs a single trigger pulse to the line structured light 3D camera when the coded mileage output by the odometer first reaches or exceeds the next untriggered mileage point in the predetermined mileage sequence; the mileage points are generated at fixed spatial intervals. The inner profile data of the rail to be measured is simultaneously acquired using the line structured light 3D camera. The processing computer preprocesses the inner profile data and extracts specific three-dimensional point cloud data of the corresponding rail waist, rail bottom and rail head regions through clustering and segmentation. The processing computer matches different weighting coefficients to the specific three-dimensional point cloud data corresponding to the rail waist, the specific three-dimensional point cloud data corresponding to the rail bottom, and the specific three-dimensional point cloud data corresponding to the rail head, and obtains the profile data after weight matching. The computer uses the processing to calculate the vertical and lateral wear of the rail under test based on the profile data after matching the weighting coefficients.

Images