CN111932676B - Railway track gauge rapid measurement device and method - Google Patents

Abstract

The invention discloses a railway track gauge rapid measurement device and a railway track gauge rapid measurement method, wherein an aircraft is used as a carrier, a measurement device is arranged at the bottom of the aircraft, the aircraft flies in a space above a railway track and along the track, the measurement device collects original steel rail image data, a pre-built steel rail BIM model point cloud is used as a template, the inner side surface of the template point cloud is extracted and is used as a reference surface on the basis of point cloud matching facing the template, and therefore the inner side surface of the actually measured steel rail point cloud is extracted, and finally the rapid measurement of the track gauge is realized. The invention has the advantages of less resources occupation, short detection time, non-contact measurement and the like.

Description

A device and method for rapid measurement of railway track gauge

Technical field

The invention belongs to the field of optical detection technology, relates to railway track detection technology, and specifically relates to a rapid measurement device for railway track gauge and a rapid measurement method thereof.

Background technique

Railway transportation has become the preferred mode of transportation for today's public travel, and plays an important role in cargo transportation. As the track is the basis of railway operation, the issue of rapid detection of railway environmental infrastructure cannot be ignored. Changes in track gauge will cause various vibrations of the train and change the wheel-rail force. It is a controlling factor affecting the safety and stability of train operation on the track. It is also an important cause of damage and failure of track structural components. With the development of high-speed railways, With the improvement of operation speed and the continuous expansion of operation scale, regular track gauge inspection and timely grasp of track gauge status information have become one of the most important contents of the track inspection project.

Traditional contact gauge detection requires railway maintenance personnel to place detection equipment (such as mechanical rulers) on two rails and continuously advance along the railway direction. If the measured gauge does not meet the standard requirements, the rulers will get stuck. , this method not only requires a lot of human resources, but also the measurement accuracy of the ruler is affected by external conditions such as temperature and climate, so the accuracy of the measurement cannot be guaranteed.

The opposite of contact measurement is non-contact measurement, that is, by mounting physical scanning equipment, such as cameras, sonic meters, lasers, etc., on the track inspection vehicle (referred to as the track inspection vehicle) or the train body, the rail profile data is captured to calculate the track gauge. , the data acquisition equipment used in this type of measurement method does not directly contact the rail. This type of method uses sensors to obtain rail profile data and use it to calculate the track gauge. The carrier needs to move on the track to obtain the data, which requires occupying track resources and poses personnel safety risks. Moreover, the contact between the equipment carrier and the rail can easily lead to secondary wear of the rail.

By carrying camera equipment, drones can collect data without touching the rails or taking up track resources. In "Vision based rail track extraction and monitoring through drone imagery", Arun Kumar Singh and others used drones to obtain images of railway scenes, then separated the tracks from the surrounding environment based on HSV color, and used Canny edge detection and nearest neighbor algorithms to extract parallel rails It is also used to calculate the track gauge, which proves the feasibility and efficiency of using aircraft for track detection. However, this method does not take into account the three-dimensional I-shaped structure of the rail, and only studies the two-dimensional plan view, resulting in the inability to distinguish the extracted edge line. Whether it belongs to the inner working surface of the rail head or to the bottom of the rail rail will have a negative impact on the distance measurement accuracy, which in turn affects the practicality of the distance measurement equipment.

Contents of the invention

One of the purposes of the present invention is to propose a railway track gauge measuring device that can detect quickly without contact, does not occupy track resources, and is low in cost, so as to meet the needs of railway construction and maintenance departments for high-efficiency measurement of rail gauge.

The technical solution adopted by the present invention to solve the technical problem is: a railway track gauge rapid measurement device, which includes a three-dimensional camera installed at the bottom of the aircraft and a main control module and a data acquisition module respectively connected to the three-dimensional camera. The data acquisition module The module is connected to a data storage module, and the data storage module is connected to a data processing module. The data storage module and the data processing module are respectively connected to the main control module. The main control module controls the three-dimensional camera to take pictures and images of the railway track. The data acquisition module The image is sampled and sent to the data storage module. The main control module controls the data processing module to read the data from the data storage module, analyze and process the data, and then send it to the data storage module for storage. The main control module is also connected to a power supply.

The second object of the present invention is to propose a railway track gauge measurement method, which includes the following steps: Step 1, data preprocessing

Step 11: Create a standard BIM model of the rail based on the information on the rail head, rail waist, rail bottom, rail gauge, and the inner working surface under the rail head tread according to the national standards for rail design requirements, and discretize it into a three-dimensional point cloud. As a template point cloud for matching;

Step 12: At the same time, the measured rail point cloud is constructed based on the measured images collected by the aircraft, and then the measured rail point cloud is segmented from the scene point cloud to obtain the measured point cloud;

Step 2, template-oriented point cloud matching: Using the point cloud of the standard BIM model as a template, using the template-oriented point cloud registration algorithm, the template point cloud is registered to the measured rail point cloud in two steps, so that the two point clouds Achieve maximum overlap;

Step 21: Perform initial point cloud matching based on the SAC_IA (Sampling Consistency Initial Alignment) algorithm;

Step 22, determine whether the distance error sum function reaches the minimum, if not, repeat step 21, if so, stop the current iteration;

Step 23: Use the template point cloud and the measured point cloud obtained by the SAC_IA algorithm to perform precise registration based on ICP (IterativeClosest Point);

Step 24, determine whether the distance error is less than the threshold or whether the maximum number of iterations has been reached;

Step 25, if not, repeat step 23. If so, the template point cloud and the measured point cloud will overlap to the greatest extent, and the registered template point cloud and the measured point cloud will be obtained;

Step 3. Calculation of track gauge based on template reference surface

Step 31: Use the inner working surface in the template point cloud as the template reference surface, extract the inner points of the template point cloud, and calculate the template point cloud reference surface;

Step 32: Extract the inner points of the measured point cloud and fit them to obtain the inner working surface. Based on this, calculate the track gauge of the measured point cloud, and then output the track gauge.

In step 21 of the railway track gauge rapid measurement method, the fast point feature histogram (FPFH) of the source point cloud (i.e., template point cloud) and target point cloud (i.e., measured point cloud) is calculated separately, and the points are obtained Concentrate the feature description of each point, then iteratively obtain the sampling point pairs of the template point cloud and the measured point cloud, and calculate the transformation matrix from the source point cloud to the target point cloud based on the sampling point pair, and finally select the transformation with the smallest distance error as the final result of the match.

Further, the specific steps are:

Step 211, calculate the normal vector and fast point feature histogram feature of each point in the source point cloud P and target point cloud Q respectively;

Step 212, automatically select n sampling points p _k (k=1,2,...,n) in the source point cloud P, and ensure that any two sampling points p _ki and p _kj satisfy the following formula:

Among them, d _min is the specified distance threshold between points;

Step 213: Find the corresponding sampling point q k that has similar fast point feature histogram characteristics to the sampling point _p k (k=1,2,...,n) in the target point cloud _Q , and use it as the source point cloud P in the target point cloud One-to-one corresponding points in Q;

Step 214: Calculate the rigid body transformation matrix between the sampling point p _k and the corresponding sampling point q _k , and the distance difference l _i between the point p _k ′ obtained after the rigid body transformation of the sampling point p _k and the corresponding sampling point q _k , then we have l _i =‖p′ _k -q _k ‖ ₂ , where p _k ′(x _k ′,y _k ′,z _k ′)=Rp _k (x _k ,y _k ,z _k )+T, where R represents Rotation matrix, T represents translation matrix;

Step 215, a set of optimal transformations should finally be found such that the distance error and function has the smallest value, and the matrix transformation at this time is regarded as the final registration transformation matrix, where,

In the formula, m _l is the given distance threshold, l _i is the distance difference after transformation of the i-th group of sampling point pairs.

Step 22 of the described method for rapid measurement of railway track gauge is to iteratively select pairs of corresponding relationship points in the source point cloud and the target point cloud, and calculate the optimal rigid body transformation matrix between the point pairs until the convergence accuracy is met. The requirement is that the distance error sum is minimum. The difference is that when finding the corresponding point for each point in the source point cloud, the principle of random selection is no longer used, but the point closest to the point is selected as the corresponding point.

And the basic principle is to determine the corresponding point of the target point cloud in the nearest neighbor of the source point cloud, and then calculate the optimal rigid body transformation matrix between the point pairs until the convergence accuracy requirements are met, that is, the distance error sum is minimum.

Further, the specific steps are:

Step 221: A sampling point set of the source point cloud is P′. For each sampling point p′ _k (k=1,2,…,N) in P′, find its nearest corresponding point q in the target point cloud Q. _k :

Step 222: Calculate the rotation matrix R and the translation matrix T for the sampling point pair set of the point set P' and the target point cloud Q, so as to minimize the mean square error E(R,T) of the sampling point set:

Step 223, use the current best transformation matrices R and T for the point set P′ to obtain a new point set P″:

p″ _k (x″ _k , y″ _k , z″ _k )=Rp′ _k (x′ _k y′ _k z′ _k )+T

Among them, p″ _k represents the point after p′ _k transformation;

Step 224: Calculate the mean distance error d _ε and the current iteration number I. If the convergence conditions are met, the iteration is terminated. Otherwise, the point set P″ is used to replace the source point set P′, and step 221 is repeated until the convergence conditions are met:

I>I _max

In the formula, ε is the specified allowable error, and I _max is the specified maximum number of iterations.

The described method for rapid measurement of railway track gauge, step 3 is to use the inner working surface of the template point cloud after initial matching of the point cloud as a reference plane, and search and determine the fit most similar to the reference surface in the measured point cloud. Plane, the fitting plane represents the inner side of the measured rail.

According to the rapid measurement method of railway track gauge, the steps for extracting inner points of the template point cloud in step 31 are:

It is known that the template point cloud P consists of two parallel rail model point clouds P _{l and} P _r . The point cloud after registration with the measured point cloud is the source point cloud P′, where P _l and P _r after registration Corresponding to P _l ′, P _r ′ respectively;

For any two non-overlapping vertices p _i ′, p _j ′, i≠j in the source point cloud P′, the distance between the two points is:

d(p′ _i ,p′ _j )=||p′ _i -p′ _j || ₂ ;

For any point p _li ′ in the source point cloud P′, if the point is also a point in P _l ′, a point p _j ′ can be found in P _r ′ that satisfies: min d(p′ _li ,p′ _j )= d ₀ ,p′ _j ∈P′ _r , that is, if the shortest distance from a point on the left rail to the right rail is equal to the standard gauge, then the point is said to be the inner point of the left rail of the template point cloud;

In the same way, for any point p _ri ′ in the source point cloud P′, if the point is a point in P _r ′, a point p _k ′ can be found in P _l ′ that satisfies: min d(p _ri ′,p _k ′ )=d ₀ ,p _k ′∈P _l ′ is said to be the inner point of the rail on the right side of the template point cloud.

According to the railway track gauge measurement method, the fitting step of the inner side of the template point cloud in step 32 is: through inner point extraction, the inner point set of the template point cloud can be obtained:

U _pl ={p _l1 ′,p _l2 ′,p _l3 ′,…,p _lm ′}, U _pr ={p _r1 ′,p _r2 ′,p _r3 ′,…,p _rn ′}, through the minimum Square plane fitting can obtain the two inner planes of the template point cloud respectively;

Method for extracting the inner side of the measured point cloud: After registering the template point cloud and the measured point cloud, the shape and position of the measured point cloud are very similar to the template point cloud. The inner side of the template point cloud is known as the reference surface. By searching and determining the fitting plane most similar to the reference surface in the measured point cloud, it is the inner side of the measured point cloud.

According to the railway track gauge measurement method, the measured point cloud gauge calculation step in step 32 is as follows:

The plane fitting equations of the left and right rail inner point sets U _ql and U _qr of the measured point cloud Q obtained after extracting the inner surface are S _ql and S _qr respectively. For any point q in the space and a certain plane S, the point The distance between q and plane S is D(q,S), then the track distance d _q of the measured point cloud is calculated according to the following formula:

Among them, n _l and n _r represent the scale of the sets U _ql and U _qr respectively; at this point, the track gauge of the track is obtained.

Compared with the existing technology, the advantage of the present invention is that it uses an aircraft as a platform and takes advantage of the aircraft's fast moving speed, large field of view of the three-dimensional camera, and rapid imaging to achieve rapid track gauge measurement without occupying railway resources. ; Measure the rail gauge using a non-contact method, which will not cause wear to the sensor and rail.

Description of drawings

Figure 1 is a schematic diagram of the relative positions of the rail template point cloud and the measured point cloud;

Figure 2 is a partial enlargement of the rail in Figure 1;

Figure 3 is a structural principle block diagram of the measuring device of the present invention;

Figure 4 is an overall flow chart of the measurement method of the present invention;

Figure 5 is a schematic diagram of the aircraft measuring track gauge.

Each reference number is marked as: 1—rail head, 2—rail waist, 3—rail bottom, 4—rail gauge, 5—inside working surface, 6—inside point of template point cloud, 7—inside point of measured point cloud.

Detailed ways

The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings: it should be understood that the preferred embodiments are only for illustrating the present invention and are not intended to limit the scope of the present invention.

Example 1

Referring to Figure 3, the rapid measuring device of railway track gauge of the present invention includes an aircraft equipped with a measuring device. The measuring device includes a three-dimensional camera, a data acquisition module, a data storage module, a data processing module and a power supply.

The main control module is connected to the three-dimensional camera, data acquisition module, data storage module and power supply. The power supply supplies power to the main control module. The main control module controls the three-dimensional camera, data acquisition module and data storage module. The three-dimensional camera is connected to the data acquisition module. ; The data acquisition module is connected to the data storage module; the data storage module is connected to the data processing module.

The main control module controls the three-dimensional camera to take pictures and images of the railway track. After the image is sampled by the data acquisition module, it is sent to the data storage module. The main control module controls the data processing module to read the data from the data storage module. Analyze and process, and the processed data is sent to the data storage module for storage.

The following is one of the typical application scenarios:

The flight parameters of the aircraft: the flight speed is 5 meters/second, the flight altitude is 24.8 meters, the heading and side overlap rates are 90% and 60% respectively, and the main route angle is 111°.

Example 2

Referring to Figures 1 to 5, the present invention discloses a method for rapid measurement of railway track gauge. The measurement steps are as follows:

Stage 1, data preprocessing.

Based on the information of the rail head 1, rail waist 2, rail bottom 3, rail gauge 4 and the inner working surface 5 under the rail head tread, a standard BIM model of the rail is created according to the national standards for rail design requirements and discretized into three dimensions. Point cloud, used as template point cloud for matching. At the same time, based on the image set collected by the aircraft, a measured scene point cloud is constructed, and then the measured rail point cloud is segmented from the scene to obtain the measured point cloud.

Stage 2, template-oriented point cloud matching.

At this stage, the point cloud of the standard BIM model is used as a template, and the template point cloud registration algorithm is used to register the template point cloud to the measured rail point cloud in two steps, so that the two point clouds can overlap to the greatest extent;

Step 21: Perform initial point cloud matching based on the SAC_IA (Sampling Consistency Initial Alignment) algorithm;

Step 22, determine whether the distance error sum function reaches the minimum, if not, repeat step 21, if so, stop the current iteration;

Step 23: Use the template point cloud and the measured point cloud obtained by the SAC_IA algorithm to perform precise registration based on ICP (IterativeClosest Point);

Step 24, determine whether the distance error is less than the threshold or whether the maximum number of iterations has been reached;

Step 25. If not, repeat step 23. If so, the template point cloud and the measured point cloud will overlap to the greatest extent, and the registered template point cloud and the measured point cloud will be obtained.

The matching process at this stage is divided into two methods. The first method has its own set of iteration processes, which are described in steps 21 and 22: After the iteration of the first method is terminated, the result obtained is used as the result of the second method. input, and then restart a new, independent set of iterations using the second method, so steps 23 and 24 are solely around the second method, which is ICP-based registration.

This invention uses the point cloud of the standard BIM model as a template, and uses a template-oriented point cloud registration algorithm to register the template point cloud to the measured rail point cloud in two steps, so that the two point clouds can overlap to the greatest extent.

The point cloud corresponding to the standard BIM model of the rail (hereinafter referred to as the template point cloud) is used as the source point cloud, and the measured rail point cloud (hereinafter referred to as the measured point cloud) generated by the image set taken by the aircraft is used as the target point cloud, and a two-step template-oriented step is adopted. The registration algorithm achieves matching between the two. The two-step registration algorithm described is: first use the SAC_IA algorithm to achieve initial matching to ensure that the template point cloud and the measured point cloud have a good relative initial position, and then further achieve accurate matching based on the ICP algorithm.

The initial matching of point clouds based on the SAC_IA algorithm uses the SAC_IA algorithm to achieve initial matching to ensure that the template point cloud and the measured point cloud have a good relative initial position. The basic idea is: calculate the fast point feature histogram of the source point cloud and the target point cloud respectively. Figure (FPFH), obtain the feature description of each point in the point set, then iteratively obtain the sampling point pairs of the template point cloud and the measured point cloud, and calculate the transformation matrix from the source point cloud to the target point cloud based on the sampling point pair, and finally The distance error and the smallest transformation are selected as the final result of matching. Specific steps are as follows:

(1) Calculate the normal vector and FPFH features of each point in the source point cloud P and target point cloud Q respectively.

(2) Automatically select n sampling points p _k (k=1,2,...,n) in the source point cloud P, and ensure that any two sampling points p _ki and p _kj satisfy the following formula,

Among them, d _min is the specified distance threshold between points.

(3) Find the corresponding sampling point q _{k with similar FPFH characteristics to the sampling point p k} (k=1,2,...,n) in the target point cloud _Q , as a part of the source point cloud P in the target point cloud Q One corresponding point.

(4) Calculate the rigid body transformation matrix between the corresponding points p _k and q _k , and the distance difference l _i between the point p _k obtained after rigid body transformation (represented by p _k ′) and point q _k , then:

l _i =‖p′ _k -q _k ‖ ₂ ;

Among them, p _k ′ (x _k ′, y _k ′, z _k ′) = Rp _k (x _k , y _k , z _k ) + T, R represents the rotation matrix, and T represents the translation matrix.

(5) Finally, a set of optimal transformations should be found such that the distance error and function has the smallest value (as shown in the formula in the previous paragraph), and the matrix transformation at this time is regarded as the final registration transformation matrix (see the formula in this paragraph), where,

In the formula, m _l is the given distance threshold, l _i is the distance difference after transformation of the i-th group of sampling point pairs.

Precise point cloud matching based on ICP algorithm: The transformation matrix obtained based on the SAC-IA algorithm is not accurate, and it is difficult to ensure the matching effect between the template point cloud and the measured point cloud. For this reason, the ICP algorithm is used to achieve precise matching between the two. The basic idea of the algorithm is similar to the SAC_IA algorithm. It still iteratively selects corresponding point pairs in the source point cloud and the target point cloud, and calculates the optimal rigid body transformation matrix between the point pairs until it meets the convergence accuracy requirements, that is, the distance error sum is minimum. . The difference is that when finding the corresponding point for each point in the source point cloud, the principle of random selection is no longer used, but the point closest to the point is selected as the corresponding point. And the basic principle is to determine the corresponding point of the target point cloud in the nearest neighbor of the source point cloud, and then calculate the optimal rigid body transformation matrix between the point pairs until the convergence accuracy requirements are met, that is, the distance error sum is minimum. Specific steps are as follows:

(1) Step 221, a sampling point set of the source point cloud is P′. For each sampling point p′ _k (k=1, 2,...N) in P′, search for it in the target point cloud Q. The nearest corresponding point q _k :

(2) For the sampling point pair set of point set P′ and target point cloud Q, calculate the rotation matrix R and translation matrix T to minimize the mean square error E(R, T) of the sampling point set;

(3) Use the current best transformation matrices R and T for the point set P′ to obtain a new point set P″:

p″ _k (x″ _k , y″ _k , z″ _k )=Rp′ _k (x′ _k , y′ _k , z′ _k )+T

Among them, p″ _k represents the point after p′ _k transformation.

(4) Calculate the distance error mean d _ε and the current iteration number I. If formula 3.8 or 3.9 is satisfied, the iteration is terminated. Otherwise, use point set P″ to replace point set P′, and repeat step (3) until convergence is met. condition:

I＞ _Imax ;

In the formula, ε is the specified allowable error, and I _max is the specified maximum number of iterations.

Stage 3, track gauge calculation based on template reference surface.

Using the inner side of the template point cloud as the reference surface, extract the inner points of the measured point cloud and fit them to obtain the inner side. Based on this, the measured point cloud track gauge is calculated.

Without reconstructing the rail surface, it is very difficult to accurately extract the vertices 16mm below the rail head tread from the three-dimensional discrete point cloud based on the rail gauge definition (see Figure 1). However, after registering the template points with the point cloud The inner working surface of the cloud is used as a reference plane, and the fitting plane most similar to the reference surface is searched and determined in the measured point cloud. This plane represents the inner side of the measured rail. The inner point 6 of the template point cloud is on the inner working surface, and the inner point 7 of the measured point cloud is on the inner working surface. Therefore, after the two point clouds are registered, the calculation of the track gauge is transformed into extracting the reference surface in the template point cloud (i.e., the inner working surface of the template point cloud), extracting the inner surface based on the reference surface in the measured point cloud, and extracting the The process of calculating track gauge on the measured working surface.

The template point cloud reference surface extraction method includes the following steps:

(1) Extraction of inner points of template point cloud

It is known that the template point cloud P is composed of two parallel rail model point clouds P _l and P _r . The point cloud after registration with the measured point cloud is P′, where P _l and P _r correspond to each other after registration. P _l ′, P _r ′.

Definition 1: For any two non-overlapping vertices p _i ′, p _j ′, i≠j in the point cloud P′, the distance between the two points is: d(p′ _i ,p′ _j )=||p′ _i -p′ _j || ₂ .

Definition 2: For any point p _li ′ in the point cloud P′, if the point is also a point in P _l ′, a point p _j ′ can be found in P _r ′ that satisfies: min d(p′ _li ,p′ _j )=d ₀ ,p′ _j ∈P′ _r .

That is, if the shortest distance from a point on the left rail to the right rail is equal to the standard gauge, then the point is said to be the inner point of the left rail of the template point cloud.

In the same way, for any point p _ri ′ in the point cloud P′, if the point is a point in P _r ′, a point p _k ′ can be found in P _l ′ that satisfies: min d(p _ri ′,p _k ′) =d ₀ ,p _k ′∈P _l ′, this point is said to be the inner point of the rail on the right side of the template point cloud.

(2) Template point cloud inner side fitting

Through inner point extraction, the inner point set of the template point cloud can be obtained.

U _pl ={p _l1 ′,p _l2 ′,p _l3 ′,…,p _lm ′}, U _pr ={p _r1 ′,p _r2 ′,p _r3 ′,…,p _rn ′}, through the minimum Square plane fitting can obtain the two inner planes of the template point cloud respectively.

The steps of the method for extracting the inner side of the measured point cloud are: after registering the template point cloud and the measured point cloud, the shape and position of the measured point cloud are very similar to the template point cloud. The inner side of the template point cloud is known as the reference surface. By searching and determining the fitting plane most similar to the reference surface in the measured point cloud, it is the inner side of the measured point cloud.

The steps of the measured point cloud track gauge calculation method are: after extracting the inner surface, the plane fitting equations of the left and right rail inner point sets U _ql and U _qr of the measured point cloud Q are S _ql and S _qr respectively. For any point q in space and a certain plane S, the distance from point q to plane S is D(q,S), then the track distance d _q of the measured point cloud is calculated according to the following formula:

Among them, n _l and n _r represent the sizes of sets U _ql and U _qr respectively. At this point, the gauge of the track is obtained.

The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. In this way, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims (5)

1. A rapid measurement method of a rapid measurement device for railway track gauge is characterized in that,

based on a measuring device comprising a three-dimensional camera arranged at the bottom of an aircraft, a main control module and a data acquisition module, wherein the main control module and the data acquisition module are respectively connected with the three-dimensional camera, the data acquisition module is connected with a data storage module, the data storage module is connected with a data processing module, the data storage module and the data processing module are respectively connected with the main control module, the main control module controls the three-dimensional camera to photograph and image a railway track, the data acquisition module samples an image and then sends the image to the data storage module, and the main control module controls the data processing module to read data from the data storage module, analyze and process the data and then send the data to the data storage module for storage;

the method comprises the following steps:

step 1, data preprocessing

Step 11, creating a standard BIM model of the steel rail based on the information of the rail head (1), the rail web (2), the rail bottom (3), the rail gauge (4) and the inner side working surface (5) below the rail head tread, and discretizing the model into a three-dimensional point cloud serving as a template point cloud for matching;

step 12, constructing an actual measurement steel rail point cloud based on an image acquired by an aircraft, and dividing the actual measurement steel rail point cloud from a scene point cloud to obtain an actual measurement point cloud;

step 2, realizing point cloud matching for templates

Step 21, performing point cloud initial matching based on SAC_IA algorithm: respectively calculating fast point characteristic histograms of a source point cloud and a target point cloud, acquiring characteristic description of each point in a point set, then iteratively acquiring sampling point pairs of a template point cloud and a real point cloud, calculating a transformation matrix from the source point cloud to the target point cloud according to the sampling point pairs, and selecting a distance error and minimum transformation as a final matching result;

step 211, calculating normal vector and fast point characteristic histogram features of each point in the source point cloud P and the target point cloud Q respectively;

step 212, automatically selecting n sampling points P from the source point cloud P _k (k=1, 2, …, n) and ensures any two sampling points p _ki And p _kj Satisfies the following formula:

wherein d is _min A threshold value for a specified inter-point distance;

step 213, finding and sampling point p in target point cloud Q _k (k=1, 2, …, n) corresponding sample points q having similar fast point feature histogram features _k As a one-to-one correspondence point of the source point cloud P in the target point cloud Q;

step 214, calculating the sampling point p _k And corresponding sampling point q _k Rigid body transformation matrix between and sampling point p _k The point p obtained after rigid transformation _k ' and corresponding sample point q _k Distance difference l of (2) _i Then there is l _i ＝||p′ _k -q _k || ₂ Wherein p is _k ′(x _k ′,y _k ′,z _k ′)＝Rp _k (x _k ,y _k ,z _k ) +T, where R represents a rotation matrix and T represents a translation matrix;

step 215, eventually a set of optimal transformations should be found, such that the distance error and the functionThe matrix transformation at this time is considered to be the final registration transformation matrix, where,

m is in _l For a given distance threshold, l _i The distance difference after the i-th group sampling point pair is transformed;

step 22, judging whether the distance error and the function reach the minimum, if not, repeating step 21, and if so, stopping the current iteration;

selecting corresponding relation point pairs from the source point cloud and the target point cloud iteratively, and calculating an optimal rigid body transformation matrix between the point pairs until convergence accuracy requirements, namely distance errors and minimum, are met;

step 221, a sampling point set of the source point cloud is P ', for each sampling point P ' in P ' _k (k=1, 2, …, N), finding its nearest counterpart point Q in the target point cloud Q _k ：

Step 222, for the set of sampling point pairs of the point set P' and the target point cloud Q, calculating a rotation matrix R and a translation matrix T, so as to minimize a mean square error E (R, T) of the sampling point set:

step 223, using the current optimal transformation matrices R and T for the point set P', obtaining a new point set p″:

p" _k (x" _k ,y" _k ,z" _k )＝Rp' _k (x′ _k ,y' _k ，z′ _k )+T

wherein p' _k Represents p' _k Transformed points;

step 224, calculate the distance error mean d _ε And the current iteration number I, if the convergence condition is met, the iteration is terminated, otherwise, the point set P 'is used for replacing the source point set P', and the step 221 is repeated until the convergence condition is met:

I＞I _max

wherein ε is a specified allowable error, I _max For a specified maximum number of iterations;

step 23, carrying out ICP-based accurate registration on the template point cloud and the actual point cloud obtained by the SAC_IA algorithm;

step 24, judging whether the distance error is smaller than a threshold value or the maximum iteration number is reached;

step 25, if not, repeating the step 23, and if so, enabling the template point cloud and the real point cloud to coincide to the greatest extent, so as to obtain the registered template point cloud and real point cloud;

step 3, track gauge calculation based on template reference surface

Step 31, taking an inner working surface (5) in the template point cloud as a template reference surface, extracting inner points of the template point cloud, and calculating the template point cloud reference surface;

and step 32, extracting the inner points of the real-point cloud, fitting to obtain an inner working surface (5), calculating the track gauge of the real-point cloud, and then outputting the track gauge.

2. The method for rapidly measuring the railway track gauge according to claim 1, wherein the step 3 is to use an inner working surface (5) of a template point cloud after initial matching of the point cloud as a reference plane, search and determine a fitting plane most similar to the reference plane in a real-time point cloud, and represent the inner side surface of a measured steel rail by the fitting plane.

3. The method for rapidly measuring railway track gauge according to claim 2, wherein the step of extracting the inner points of the template point cloud in the step 31 is as follows:

the known template point cloud P is formed by two parallel steel rail model point clouds P _l ，P _r The point cloud registered with the real point cloud is a source point cloud P', wherein P is the point cloud P _l ，P _r Registered and respectively correspond to P _l ′，P _r ′；

For any two non-coincident vertexes P in the source point cloud P _i ′,p _j ' i not equal to j, the distance between two points is: d (p' _i ，p′ _j )＝||p′ _i -p′ _j || ₂ ；

For any point P in the source point cloud P _li ' if this point is also P _l ' one point in P _r One point p can be found in _j ' satisfy: min d (p' _li ，p′ _j )＝d ₀ ，p′ _j ∈P _r ' namely, the shortest distance from a certain point in the left steel rail to the right steel rail is equal to the standard track gauge, and the point is called as the inner side point of the left steel rail of the template point cloud;

similarly, for any point P in the source point cloud P _ri ' if the point is P _r ' one point in P _l One point p can be found in _k ' satisfy: mind (p) _ri ′,p _k ′)＝d ₀ ,p _k ′∈P _l The point is the inboard point of the right rail of the template point cloud.

4. The method for rapidly measuring railway track gauge according to claim 2, wherein the step of fitting the inner side of the template point cloud in the step 32 is as follows: extracting an inner side point set of the template point cloud through the inner side points: u (U) _pl ＝{p _l1 ′，p _l2 ′，p _l3 ′，……，p _lm ′}，U _pr ＝{p _r1 ′，p _r2 ′，p _r3 ′，……，p _rn ' two inner side planes of the template point cloud can be obtained respectively through least square plane fitting.

5. The method for rapidly measuring railway track gauge according to claim 2, wherein the actual point cloud computing step in the step 32 is as follows:

left and right rail inner side point set U of real side point cloud Q obtained through inner side face extraction _ql ，U _qr Plane fitting equations of S _ql ，S _qr For any point q and a certain plane S in the space, if the distance from the point q to the plane S is D (q, S), the track gauge D of the real point cloud is calculated according to the following formula _q ：

Wherein n is _l ，n _r Respectively represent the collection U _ql And U _qr Scale of (2); the track gauge of the track is thus obtained.