CN-121982023-A - Three-dimensional imaging and state detection method and system for high-speed moving train

CN121982023ACN 121982023 ACN121982023 ACN 121982023ACN-121982023-A

Abstract

The invention discloses a three-dimensional imaging and state detection method and system for a high-speed mobile train, which are applied to the technical field of train state detection, and eliminate the motion distortion of the train under high-speed running and restore high-precision static point cloud data through multi-source data fusion, motion compensation and elastic deformation correction; the method comprises the steps of adopting feature constraint downsampling and poisson curved surface reconstruction, generating a smooth and complete three-dimensional model while retaining edge features, generating RGB-D multi-modal data by aligning the three-dimensional model with image data, introducing a fast-RCNN network of a channel attention mechanism to realize weight self-adaption of textures and geometric features, remarkably improving the accuracy of all-weather defect detection, realizing dynamic trend monitoring and predictive maintenance by correlation and contrast analysis of cross-site defect results, and greatly improving the automation level and efficiency of train state detection.

Inventors

Dai Daowei
WANG HAILIN
LI YAN
BAI SHIYU
ZANG CHUANZHI
TIAN JIYUAN
YU LONGBO
ZHOU HAIJUN
PENG XIAOLU
Liao Xihang
LI QIANG
XU YANG
LI SHIDONG

Assignees

辽宁鼎汉奇辉电子系统工程有限公司

Dates

Publication Date: 20260505
Application Date: 20260402

Claims (10)

1. The three-dimensional imaging and state detection method for the high-speed moving train is characterized by comprising the following steps of: Respectively acquiring multi-source data with a time stamp when the same train passes through a plurality of train stations, and constructing space-time indexes of the trains at different positions based on the multi-source data, wherein the multi-source data comprises point cloud data, image data, motion sensing data and train information; Combining the motion sensing data of each station, and calculating rigid transformation and elastic deformation of the train based on the constructed train model so as to perform deformation correction and motion compensation on the point cloud data to obtain corrected static point cloud data; Respectively carrying out self-adaptive weighted fusion on static point cloud data of the same site based on the space-time index to generate fusion point clouds corresponding to each carriage; performing voxel grid division on the fusion point cloud, performing feature constraint downsampling and filtering treatment, and then reconstructing a curved surface to obtain a three-dimensional model corresponding to each carriage, wherein a sampling function in the feature constraint downsampling is constructed based on normal vectors and curvatures of each point set in the voxel grid; Respectively aligning the three-dimensional model of each station with corresponding image data to generate RGB-D multi-mode data, carrying out full-carriage state detection on the RGB-D multi-mode data by adopting an improved fast-RCNN network, and outputting defect detection results under each station; And correlating the defect detection results of the same train at different stations, and performing comparison analysis to obtain the state detection results of the train.
2. The method for three-dimensional imaging and status detection of a high-speed mobile train according to claim 1, wherein the train model comprises a rigid body motion model and an elastic deformation model; The rigid motion model adopts extended Kalman filtering to fuse the motion sensing data, and generates a rigid state matrix when the train runs at each moment; The elastic deformation model is used for describing the elastic deformation of the vehicle body by using a mode superposition method and generating a deformation compensation matrix.
3. The three-dimensional imaging and status detection method of a high-speed mobile train according to claim 2, wherein the method for describing the elastic deformation of the train body by using a modal superposition method to generate a deformation compensation matrix comprises the following steps: Obtaining the vibration mode vector of each train body node of the train in each order mode through finite element analysis; The method comprises the steps of utilizing collected vibration data and identifying modal parameters in real time through a random subspace method, and updating elastic deformation, wherein the modal parameters comprise natural frequencies and damping ratios; estimating modal coordinates from the vibration data by kalman filtering based on the mode shape vector and the modal parameters; And calculating local deformation displacement of the vehicle body position corresponding to each point cloud by using a mode superposition method based on the mode coordinates and the vibration mode vector, and obtaining a deformation compensation matrix based on the local deformation displacement.
4. A method of three-dimensional imaging and condition detection of a high speed mobile train according to claim 3, wherein the modal coordinates are estimated from the vibration data by kalman filtering, see the following formula: ; Wherein, the (T) represents the j-th order modal coordinates, Indicating the natural frequency of the jth order, The damping ratio is indicated by the expression, Representing the general force of the force in the broad sense, The speed is indicated by the velocity of the light, Representing acceleration; Local deformation displacement of each point cloud at t moment See the following formula: ; Wherein, the The mode shape vector representing the j-th order mode, M representing the total number of modes, and i representing the spatial index.
5. The three-dimensional imaging and state detection method for a high-speed mobile train according to claim 4, wherein the self-adaptive weighted fusion is performed on static point cloud data of the same station based on space-time indexes to generate fusion point clouds corresponding to each carriage, and the method comprises the following steps: Extracting a fast point characteristic histogram of each static point cloud data, and obtaining an initial transformation matrix through sampling consistency initial registration to realize coarse alignment; Based on the initial transformation matrix, and adopting an iterative nearest point algorithm based on elastic deformation for fine alignment, the objective function of the algorithm can be expressed as the following formula: ; Wherein, the 、 Representing static point cloud data pairs corresponding to different acquisition nodes of the same carriage after coarse alignment, R represents a rotation matrix, The translation vector is represented as a function of the translation vector, The regularization coefficient is represented as a function of the regularization coefficient, Representing the weight of each static point cloud data pair, The square of the norm is represented, Regularization term for elastic deformation; And calculating the weight of each point in the aligned static point cloud data, and carrying out weighted average to obtain the fusion point cloud of each carriage.
6. The method for three-dimensional imaging and status detection of a high-speed mobile train according to claim 1, wherein the feature constrained downsampling comprises the steps of: Calculating the normal vector and curvature of each point set in the voxel grid; Based on the following sampling function Sampling: ; Wherein, the And Representing the average normal vector and the average curvature within the voxel grid respectively, 、 Are weight coefficients, and I represent norms, Representing the normal vector at the mth point set, Representing the curvature at the mth point set, Representing points in the voxel grid, m representing a point index within the voxel grid; Traversing all point sets within a voxel grid, selecting a sampling function The largest point set remains.
7. The three-dimensional imaging and state detection method of the high-speed mobile train according to claim 1, wherein the three-dimensional model corresponding to each carriage is obtained through curved surface reconstruction, and the method comprises the following steps: Converting the fusion point cloud after filtering into a continuous watertight triangular mesh model by using a Poisson equation, wherein the Poisson equation converts the reconstruction problem into a solution indicating function : ; Wherein, the The representation indicates the gradient field of the function, Representing a point cloud normal vector field; Based on the elastic deformation as a constraint, see the following formula: ; Wherein, the Indicating an indication function estimated from the elastic deformation, Representing the balance weights.
8. A three-dimensional imaging and status detection system for a high-speed mobile train, for implementing the three-dimensional imaging and status detection method for a high-speed mobile train according to any one of claims 1 to 7, characterized in that the system comprises: The system comprises an acquisition module, a data processing module and a data processing module, wherein the acquisition module is used for respectively acquiring multi-source data with time stamps when the same train passes through a plurality of train stations and constructing time-space indexes of the train at different positions based on the multi-source data, and the multi-source data comprises point cloud data, image data, motion sensing data and train information; the compensation module is connected with the acquisition module and used for combining the motion sensing data of each station, calculating rigid transformation and elastic deformation of the train based on the constructed train model, and carrying out deformation correction and motion compensation on the point cloud data to obtain corrected static point cloud data; The fusion module is connected with the compensation module and is used for carrying out self-adaptive weighted fusion on the static point cloud data of the same site based on the space-time index to generate fusion point clouds corresponding to each carriage; The reconstruction module is connected with the fusion module and is used for carrying out voxel grid division on the fusion point cloud, carrying out feature constraint downsampling and filtering treatment, and then carrying out curved surface reconstruction to obtain a three-dimensional model corresponding to each carriage; The detection module is connected with the reconstruction module and used for respectively aligning the three-dimensional model of each site with corresponding image data to generate RGB-D multi-mode data, carrying out full carriage state detection on the RGB-D multi-mode data by adopting an improved Faster-RCNN network and outputting a defect detection result under each site; and the output module is connected with the detection module and is used for correlating the defect detection results of the same train at different stations and carrying out comparison analysis to obtain the state detection results of the train.
9. An electronic device comprising a memory, a processor, and a computer program, wherein the computer program is stored in the memory and configured to be executed by the processor to implement the method of three-dimensional imaging and status detection of a high speed mobile train as claimed in any one of claims 1 to 7.
10. A computer-readable storage medium having stored thereon a computer program, the computer program being executable by a processor to implement a method of three-dimensional imaging and condition detection of a high speed moving train as claimed in any one of claims 1 to 7.

Description

Three-dimensional imaging and state detection method and system for high-speed moving train Technical Field The invention relates to the technical field of train state detection, in particular to a method and a system for three-dimensional imaging and state detection of a high-speed moving train. Background Along with the continuous rising of railway freight traffic, the railway freight car is easy to have potential safety hazards such as deformation, corrosion, component loss, foreign matter residue and the like in a carriage structure under long-term high-load operation. The current railway cargo inspection operation mainly depends on the following modes: 1. Manual train inspection The operation and maintenance personnel perform visual inspection or hammering inspection after the train is stopped. The mode is high in labor intensity, dangerous in working environment and low in efficiency, and is easily influenced by subjective factors to cause missed detection. 2. 2D vision detection system And shooting train images by using a high-definition linear array camera, and identifying faults by manpower or an algorithm. Although partial automation is realized, the 2D image can only acquire object surface texture information, lacks depth data, cannot quantify deformation degree, concave depth of a carriage, size of goods and space position of foreign matters, has natural defects for vehicle door bracket, overload and unbalanced load and foreign matter identification under laminated shielding, and is extremely easily influenced by light fluctuation, shadow interference and other environmental factors. 3. 3D detecting system Existing partial 3D detection attempts mostly employ static scanning or multi-site stitching techniques. The point cloud is acquired by using a laser scanner, but the point cloud is in multi-view registration, the process is complex and time-consuming, and the rapid detection requirement of the railway wagon in high-speed dynamic communication of 5-80km/h is difficult to adapt. In addition, in the existing 3D point cloud processing technology, in order to meet the requirement of real-time traditional voxel downsampling, key features such as carriage edges and angles are often lost, if full data are reserved, the calculated amount is huge, and the second-level processing requirement of continuous passing of a train cannot be met. In addition, the model generated by the traditional algorithm has many noise points and unsmooth surface and affects the subsequent intelligent recognition precision due to the motion blur caused by the outdoor complex illumination and high-speed movement, and the model generated by the traditional reconstruction algorithm is lack of effective connection with the fault recognition algorithm, so that the overall automation degree of the detection system is not high. Therefore, there is a strong need to develop a method and a system for three-dimensional imaging and state detection of a high-speed moving train to solve the above problems. Disclosure of Invention The invention provides a three-dimensional imaging and state detection method and system for a high-speed moving train, which solve the problems in the prior art. According to a first aspect of the invention, a three-dimensional imaging and state detection method for a high-speed moving train is provided. The method comprises the following steps: Respectively acquiring multi-source data with a time stamp when the same train passes through a plurality of train stations, and constructing space-time indexes of the trains at different positions based on the multi-source data, wherein the multi-source data comprises point cloud data, image data, motion sensing data and train information; Combining the motion sensing data of each station, and calculating rigid transformation and elastic deformation of the train based on the constructed train model so as to perform deformation correction and motion compensation on the point cloud data to obtain corrected static point cloud data; Respectively carrying out self-adaptive weighted fusion on static point cloud data of the same site based on the space-time index to generate fusion point clouds corresponding to each carriage; performing voxel grid division on the fusion point cloud, performing feature constraint downsampling and filtering treatment, and then reconstructing a curved surface to obtain a three-dimensional model corresponding to each carriage, wherein a sampling function in the feature constraint downsampling is constructed based on normal vectors and curvatures of each point set in the voxel grid; Respectively aligning the three-dimensional model of each station with corresponding image data to generate RGB-D multi-mode data, carrying out full-carriage state detection on the RGB-D multi-mode data by adopting an improved fast-RCNN network, and outputting defect detection results under each station; And correlating the defect detection results of the same train at