Granted Patent B1

US 12,322,135 · App. 19/024,045 · Granted Jun 3, 2025

Inventors: Yi Huang (Nanjing, CN); Yong Zhao (Nanjing, CN); Chunmei Wang (Nanjing, CN)

Assignees: Nanjing Kingyoung Intelligent Science & Technology Co., Ltd.; Shanghai Weitong Vision Technology CO., Ltd.

G06T7/74B61L27/57G06T7/0002G06T7/80G06T2207/10028

Abstract

Claims (35)

1. A method for gauge detection of a rail vehicle based on three-dimensional point cloud data, comprising following steps:

S 1 , building: building a gauge detection gate at a predetermined position wherein the rail vehicle leaves a garage, and installing a profilometer on the gauge detection gate, wherein the profilometer comprises a plurality of laser camera modules, a proximity switch and a speed measuring unit are further installed on the gauge detection gate or at the predetermined position wherein the rail vehicle leaves the garage;

verifying whether a factory calibration result of each of the laser camera modules is within a calibration expected range, and if the factory calibration result conforms to the calibration expected range, directly executing a process of multi-module calibration; and

if the factory calibration result does not conform to the calibration expected range, firstly, using calibration reference to perform single-module calibration on each of the laser camera modules in turn, and then performing the process of the multi-module calibration after all the laser camera modules complete the single-module calibration, wherein the calibration reference comprises a sawtooth calibration block vertically set on a horizontal plane;

a process of the single-module calibration comprises:

correcting calibration: adjusting a position of the sawtooth calibration block, so the sawtooth calibration block and a line laser emitted by each of the laser camera modules are located on a same vertical axis;

collecting images: adjusting a distance between each of the laser camera modules and the sawtooth calibration block, and collecting N images at positions with N different distances, wherein N≥1;

starting calibration: respectively calculating positions of feature points of peaks and valleys of the N images in an image coordinate system according to straight line fitting and a formula of straight line intersection calculation, then calculating a transformation relationship from a pixel coordinate system of each of the laser camera modules to a sawtooth calibration block coordinate system according to known actual sizes between the peaks and the valleys, and saving data of the transformation relationship as a single-module calibration file;

S 2 , calibrating: performing the multi-module calibration on all the laser camera modules built by using a calibration structure, wherein the calibration structure is a frame formed by a plurality of sawtooth calibration blocks, and an upper part and both sides of the calibration structure are the sawtooth calibration blocks connected end to end;

recording current calibration parameters of all the laser camera modules, and using the current calibration parameters as a point cloud stitching basis for subsequent real rail vehicle images;

by performing the multi-module calibration on all the built laser camera modules, transforming a laser camera module coordinate system into a calibration structure coordinate system;

S 2 - 1 , loading 3D profiles collected by all the laser camera modules;

S 2 - 2 , extracting bevel data of a sawtooth, fitting extracted bevel data into a straight line, and then calculating to obtain an intersection of two oblique lines, thus obtaining 3D coordinates of a sawtooth vertex in the laser camera module coordinate system;

S 2 - 3 , obtaining 3D coordinates of a sawtooth point in the calibration structure coordinate system according to a physical size of the calibration structure;

S 2 - 4 , calculating a rotation and translation transformation matrix, calculating to obtain a transformation relationship from the laser camera module coordinate system to the calibration structure coordinate system, and transforming a sawtooth line from the laser camera module coordinate system to the calibration structure coordinate system, so the sawtooth line and the sawtooth point coincide in a predetermined range; and

S 2 - 5 , saving the current calibration parameters as a multi-module calibration file for subsequent multi-camera point cloud stitching;

S 3 , detecting: when the rail vehicle undergoing gauge detection passes through the gauge detection gate, sensing a vehicle speed of a current rail vehicle by the speed measuring unit in real time and reporting to a processing unit; and

performing a full section scan of an outer contour of the rail vehicle by the profilometer to generate a three-dimensional point cloud map of the current rail vehicle;

S 4 , comparing results: comparing the three-dimensional point cloud map of the current rail vehicle generated in the S 3 with built-in standard gauge contour data to judge whether the current rail vehicle is out-of-gauge; and

S 5 , outputting results: according to comparison results of the S 4 , if the current rail vehicle is not out-of-gauge, sending a notice of not out-of-gauge; and

if the current rail vehicle is out-of-gauge, sending a notice of out-of-gauge, and providing out-of-gauge parameters of the current rail vehicle.

2. The method for the gauge detection of the rail vehicle according to claim 1 , wherein step S 3 further comprises:

S 3 - 1 , loading a 3D model of predetermined images and the multi-module calibration file saved in step S 2 - 5 locally, and processing actually collected point cloud data to make a profile of a fixed position out-of-gauge, simulating an out-of-gauge state, and verifying whether a gauge detection model is capable of detecting an out-of-gauge position:

if the gauge detection model currently constructed successfully detects the out-of-gauge position and an out-of-gauge amount, saving the current gauge detection model as a basis for subsequent out-of-gauge detection; and

if the gauge detection model currently constructed is not capable of detecting the out-of-gauge position or a deviation value of detected out-of-gauge amount is greater than an expected value, returning to step S 2 to re-execute a calibration procedure for the laser camera modules;

S 3 - 2 , when the rail vehicle undergoing the gauge detection passes through the gauge detection gate, sensing the vehicle speed of the current rail vehicle by the speed measuring unit in real time and reporting to the processing unit; and

S 3 - 3 , using the gauge detection model saved in the S 3 - 1 to generate the three-dimensional point cloud map of the current rail vehicle, and displaying body contour data of the current rail vehicle.

3. The method for the gauge detection of the rail vehicle according to claim 2 , wherein step S 4 further comprises:

S 4 - 1 , based on the three-dimensional point cloud map drawn in step S 3 - 3 , calculating coordinates of a maximum Z value corresponding to each contour point on an X axis to obtain maximum contour points; and

S 4 - 2 , judging whether all the maximum contour points are out-of-gauge:

if all the maximum contour points are not out-of-gauge, all contours are not out-of-gauge; and

if any at least one of the maximum contour points is out-of-gauge, checking whether all profiles are out-of-gauge through a range of a current maximum contour point out-of-gauge on the X axis.

4. The method for the gauge detection of the rail vehicle according to claim 3 , wherein an odd number of intersections between a horizontal ray of the point and an area is used to judge being in a closed area;

if a horizontal line drawn to right has even intersections with edges of the closed area, a current point is outside the closed area; and

if the horizontal line drawn to the right has odd intersections with the edges of the closed area, the current point is in the closed area.

Assignments (1)

Priority Claims (1)

Continuity (1)

References Cited (35)