CN111882592B

CN111882592B - Steel rail contour robustness registration method based on constraint iteration closest point method

Info

Publication number: CN111882592B
Application number: CN202010711688.XA
Authority: CN
Inventors: 李艳福; 智英建; 钟晓芸; 蒋馥蔚; 刘宏立; 马子骥
Original assignee: Air Force Engineering University of PLA
Current assignee: Air Force Engineering University of PLA
Priority date: 2020-07-22
Filing date: 2020-07-22
Publication date: 2022-12-06
Anticipated expiration: 2040-07-22
Also published as: CN111882592A

Abstract

The invention discloses a steel rail contour robustness registration method based on a constraint iteration closest point method, which comprises the steps of firstly, moving a measured contour to an initial position close enough to a standard contour through a contour rough registration method, and thus ensuring the convergence of a CICP algorithm; then, the rail web is subjected to self-adaptive and global segmentation through the Ramer algorithm, a local pollution-free area is accurately extracted, the pollution-free performance of a CICP registration area is guaranteed, finally, the vertical deviation introduced by the rail web local registration is eliminated through the vertical position constraint of rail jaw points on the contour after fine registration, and the accuracy of the CICP registration is effectively improved.

Description

Steel rail contour robustness registration method based on constraint iteration closest point method

Technical Field

The invention relates to the field of rail transit detection, in particular to a steel rail contour robust registration method based on a constrained iteration closest point method.

Background

Accurate measurement of the rail profile plays a crucial role in rail quality inspection. The measuring result can visually reflect the geometrical form of the rail section, and a scientific basis is provided for rail maintenance.

Laser image method (LIM, also called photogrammetry) and Laser Displacement Method (LDM) are two main methods for nondestructive testing of rail cross-section profile. The former involves complicated image processing procedures, and the former has been gradually replaced by the latter in practical applications.

The traditional profile measurement is based on the ideal profile curve form of the measurement profile, namely, the profile comprises a complete railhead and a complete railweb, and strong outlier interference does not exist on the curve. At this time, the outline registration method mainly includes a Double circle center fitting (DCS) method and an Iterative Closest Point (ICP) method, and the principle is as shown in fig. 1.

The double-circle-center fitting method is used for calculating a rotation matrix and a translation vector by fitting the circle center coordinates of the double arcs of the rail web of the measured contour and comparing the circle center coordinates with the circle center coordinates of the corresponding position of the standard contour, so that the accurate registration of the measured contour and the standard contour is realized. The iterative closest point method is used for estimating the optimal rotation matrix and translation vector parameters based on least square by searching a plurality of registration characteristic point pairs on the non-polluted areas of the rail web of the measured contour and the standard contour, thereby realizing contour registration.

In the actual operation process, due to the complexity of field working conditions, the measurement profile curve form may be seriously damaged, and the main interference factors are as follows:

(1) Outlier interference

In order to make the laser plane stably cover the rail head and the rail web area all the time in the dynamic operation process, a certain margin is usually left in the width of the laser plane. In this case, in addition to projecting on the rail, it may also project on other irrelevant areas, such as on both sides of the ballast and the clip, which may introduce a large number of outliers into the measurement profile, an example of which is shown in fig. 2. In addition, the reflection of the bright areas on the surface of the steel rail further increases the interference degree of the outliers. The existence of outliers not only affects the positioning of the rail waist region on the profile, but also further affects the accurate measurement of rail section wear. Therefore, before contour registration, we must detect and remove most of the outlier interference.

(2) Local occlusion or loss of web

Due to the harsh outdoor working environment, the lower half of the web of the rail profile is often obscured by weeds, mud or other contaminants. In addition, when the inspection vehicle passes through a small-radius curve, as the relative distance between the laser displacement sensor fixed to the bottom of the vehicle body and the rail increases, a small-radius arc area of the rail web farthest from the sensor may be locally missing, as shown in fig. 3.

For DCS-based rail profile registration, the integrity of the rail web double arcs is crucial. Considering that local occlusion or deletion of the rail web often occurs in actual working conditions, the method is not suitable for realizing robust registration of the contour.

For the ICP-based rail contour registration, the rail waist area for extracting the registration feature points is not required to cover the complete rail waist, but the non-pollution property of the area is ensured and cannot be interfered by any outside world. Under the actual working condition, a pollution area and a non-pollution area formed by local shielding and interference of the railway ballast and the fastener coexist on the curve of the rail waist. At this time, how to realize accurate division of the two is the key to solve the problem. In addition, how to ensure the convergence of the algorithm in iterative registration is also important.

Disclosure of Invention

Based on the above analysis, the conventional ICP algorithm is still higher in adaptability of contour registration than DCS, but must try to satisfy its two constraints. Therefore, on the basis of the traditional ICP algorithm, a Constrained ICP (Constrained iterative closed point, CICP) algorithm is provided to realize the robust registration of the steel rail outline under the complex working condition.

The technical scheme adopted by the invention is as follows:

a steel rail contour robustness registration method based on a constraint iteration closest point method comprises the following steps:

step S1: after a threshold T1 of the distance between adjacent points is obtained on a curve of a normal contour in a statistical manner, firstly splitting a measured contour to obtain a plurality of curve fragments, and then setting a point number threshold T2 according to the characteristic of dense distribution of data points of the normal contour to test the curve fragments one by one so as to remove sparse outliers;

step S2: setting a segment interval threshold T3, merging curve fragments which belong to the same region and have adjacent intervals not exceeding T3 again, then carrying out concave-convex inspection on each region, and extracting a contour core region based on the principle of continuous concave-convex and point number maximization;

and step S3: roughly registering the measured contour with a standard contour, calculating the slope of a straight line at the side of a rail by positioning the position of the jaw point in the core area of the contour, comparing the slope with corresponding items of the standard contour to obtain a rotation matrix and a translation vector, and transferring the measured contour to an initial position which is close enough to the standard contour;

and step S4: the method comprises the steps of performing contour rail waist weight measurement segmentation and local non-pollution area extraction, setting a distance threshold T4 through a Ramer algorithm, and sequentially finding out the positions where all distance values exceed the threshold T4 from the whole to the local on a contour rail waist measurement curve so as to globally re-segment the contour rail waist weight measurement curve into n fragments, wherein the first fragment obtained by segmentation is a local non-pollution area because the local non-pollution area is usually positioned at the upper part of the rail waist;

step S5: performing fine registration on the original measured contour and the standard contour, and searching a plurality of matching characteristic point pairs on the rail web local pollution-free areas of the measured contour and the standard contour according to an Iterative Closest Point (ICP) algorithm, wherein the matching characteristic point pairs are respectively M = { M = _i I =1,2, \8230;, n } and S = { S = { (S) } _i I =1,2, \8230;, n }, and calculating the optimal rotation matrix R based on the principle of the least sum of squared distances between matched points after registration ₂ And a translation vector T ₂ Which satisfies

Realizing fine contour registration;

step S6: correcting the registration result of the contour fine registration in the step S5 by an iteration closest point method through the constraint of the vertical position of the rail jaw point;

step S7: self-adaptive detection and removal of outliers of the measured profile after fine registration, namely calculating the distance value between the original measured profile after fine registration and the corresponding point of the railhead area of the standard profile, and performing ascending sorting, namely P ₁ 、P ₂ …P _i 、P _i+1 …P _n Then the absolute value of the difference between two adjacent distance values, i.e. | P, is calculated _i -P _i+1 I, will | P _i -P _i+1 Comparing | with a set threshold T5, and if the threshold T5 is exceeded, the distance value is P _i+1 The corresponding point of (1) is the boundary position of the outlier and the normal point on the original contour, and all points after the corresponding point are outliers.

Preferably, in step S1, T1=2 and T2=10.

Preferably, in the above step S2, T3 is set to 30.

Preferably, in step S3, the specific process of calculating the rotation matrix and the translation vector includes:

by locating the position of the jaw point in the core region of the profile, i.e. with the coordinate (x) _m ,y _m ) Calculating the slope k of the rail side straight line _m And coordinates (x) of the jaw point of the standard profile _s ,y _s ) And slope k of the rail-side straight line _s Comparing the measured profile to obtain a rotation angle theta and a rotation matrix R when the measured profile is roughly aligned ₁ And a translation vector T ₁ Are respectively as

Preferably, in the above step S4, T4 is set to 5.

Preferably, in the above step S7, T5 is set to 3.

The invention has the beneficial effects that: according to the steel rail contour robustness registration method based on the constrained iteration closest point method, firstly, a measured contour is moved to an initial position close enough to a standard contour through a contour rough registration method, so that the convergence of a CICP algorithm is guaranteed; then, the rail waist is subjected to self-adaptive and global segmentation through a Ramer algorithm, a local pollution-free area is accurately extracted, the pollution-free performance of a CICP registration area is guaranteed, finally, through the vertical position constraint of rail jaw points, the vertical deviation caused by the local registration of the rail waist is eliminated, and the accuracy of the CICP registration is effectively improved.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.

FIG. 1 is a schematic diagram of conventional contour registration; (a) DCS profile registration principle; (b) ICP profile registration principle;

FIG. 2 is a schematic diagram of outlier interference on a survey profile;

FIG. 3 is a schematic view of a partial occlusion or absence of a rail web; (a) partially shielding the real object image by the rail web; (b) a rail web local occlusion profile graph; (c) the rail web is partially missing the real object map; (d) Profile curve diagram for local deletion of rail web

FIG. 4 is an operation flowchart of the steel rail contour robustness registration method based on the constrained iteration closest point method according to the present invention;

FIG. 5 is a graph of contour coarse registration; (a) a contour segmentation effect; (b) detection removal effect of sparse outliers; (c) a fragmentation merging effect; (d) concave-convex testing and contour core area extraction; (e) Coarse registration effect of contours

FIG. 6 is a graph showing the effect of rail waist weight segmentation and extraction of a local non-polluted area;

FIG. 7 is a contour re-registration map; (a) a contour registration effect without vertical constraint; (b) a contour registration effect under vertical constraint;

FIG. 8 is a graph of adaptive detection removal of outliers on a contour; (a) re-aligning the original profile with the standard profile; (b) statistical sorting of the distance of the corresponding points of the railhead area; (c) detected outliers; (d) removing outliers;

FIG. 9 is a diagram of all sample contours and their corresponding core regions; (a) a rail web local occlusion profile map; (b) locally shielding the contour core area map by the rail web; (c) a local missing profile map of the rail web; (d) the rail web is locally lack of a contour core area graph; (e) a foot-disturbed contour map; (f) a core area map of the foot-disturbed contour;

FIG. 10 is a comparison of the registration effect of contours under different methods; (a) registration effect of local occlusion contours under different methods; (b) registration effect under different methods of local missing contours; (c) And (4) registering effects of different methods of the interfered contour of the rail foot.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. The components of embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations.

Thus, the following detailed description of the embodiments of the present invention, presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

Compared with DCS, the traditional ICP algorithm is still higher in adaptability of contour registration, but two constraint conditions of the traditional ICP algorithm must be met somehow, and for this reason, the invention provides a steel rail contour robust registration method based on a constraint iteration closest point method, as shown in fig. 4. The operation steps of the algorithm will be explained in detail by taking the partially occluded contour of the rail web in fig. 3 as an example.

To ensure CICP convergence, appropriate rotation and translation vectors must be found to shift the measured profile to an initial position sufficiently close to the standard profile. Although the sensors are fixed to the bottom of the car body and the position of the sensors relative to the rails during dynamic operation varies to a substantially limited extent, these parameters can theoretically be obtained from fixed empirical or accurately predicted values, but transient jitter in the car body caused by non-smooth transitions when passing through the rail gap or switch area can still cause the position of the measurement profile to deviate significantly from the previous frame. At this time, the reliability of both methods is difficult to be ensured. Based on the above analysis, the present invention transfers the measured profile to an initial position sufficiently close to the standard profile by coarse registration of the profiles.

The first step is as follows: contour core region extraction

After the threshold T1 of the distance between adjacent points is obtained statistically on the curve of the normal profile (T1 is set to 2), the measured profile is first split to obtain a plurality of curve fragments (as shown in (a) of fig. 5), and then the curve fragments are tested one by one to remove sparse outliers (as shown in (b) of fig. 5) according to the characteristic that the data points of the normal profile are densely distributed and the threshold T2 of the set point number is 10.

Setting the segment pitch threshold T3 to be 30, merging the curve fragments belonging to the same region whose adjacent intervals do not exceed T3 again (see (c) in fig. 5), and then performing a roughness inspection on each region to extract a contour core region based on the principle of continuous roughness and maximum point number (see (d) in fig. 5).

The second step: coarse registration of measured profile with standard profile

Rough registration of the measured profile with the standard profile (e.g. fig. 5), calculation of the slope of the straight line on the rail side by locating the position of the jaw point in the core region of the profile, and comparison with the corresponding entry of the standard profile to obtain the rotation matrix and translation vector, and transfer of the measured profile to an initial position sufficiently close to the standard profile.

The specific calculation process of the rotation matrix and the translation vector is as follows:

by locating the position of the jaw point in the core region of the profile, i.e. with the coordinate (x) _m ,y _m ) Calculating the slope k of the rail side straight line _m And coordinates (x) of the jaw point of the standard profile _s ,y _s ) And slope k of the rail-side straight line _s Comparing the measured profile to obtain a rotation angle theta and a rotation matrix R when the measured profile is roughly aligned ₁ And translation vector T ₁ Are respectively as

The third step: contour rail waist weight segmentation and local pollution-free extraction

As can be seen from fig. 3, the contamination of the web of the gauge profile caused by occlusion or absence is mainly located in the middle-lower part, and the upper part of the web is always free of contamination. In order to realize the adaptive maximum proportion extraction of the region, a distance threshold T4 is set through the Ramer algorithm, and positions where all distance values exceed the threshold T4 are sequentially found from whole to local on the measurement contour waist curve, so that the measurement contour curve is globally re-divided into n pieces, and the result is shown in fig. 6. Here, the distance threshold T4 of the Ramer algorithm is set to 5, which is a larger value, to obtain a better global segmentation effect. Then, the first fragment of the segmented rail web is the local non-pollution area that we want to extract.

The fourth step: fine registration of raw measurement profile with standard profile

According to an Iterative Closest Point (ICP) algorithm, a plurality of registration characteristic point pairs are searched on a rail web local pollution-free area of the measured contour and the standard contour, wherein the registration characteristic point pairs are respectively M = { M = _i I =1,2, \8230;, n } and S = { S = { (S) } _i I =1,2, \8230;, n }, and calculating the optimal rotation matrix R based on the principle of the least sum of squared distances between matched points after registration ₂ And a translation vector T ₂ Which is satisfied with

And realizing contour fine registration. However, because the rail web local pollution-free area is located on a large-radius arc similar to a straight line (taking the 60kg/m steel rail which is most widely applied in China as an example, the radius of the large arc reaches 400 mm), because of the lack of vertical position constraint, the measurement profile registered by the traditional ICP algorithm may be randomly distributed along the rail web longitudinal direction of the standard profile, and the result is shown in fig. 7 (a), thereby generating a vertical registration deviation. Therefore, we propose to use the position of the rail jaw point as a vertical constraint to correct the contour registration result of the traditional ICP algorithm, so as to effectively improve the accuracy of the CICP registration, and the contour registration result after the constraint is added is shown in fig. 7 (b). Obviously, the registration accuracy of the latter is significantly higher than that of the former.

The fifth step: adaptive detection removal of registered contour outliers

After the CICP is used for accurately registering the core area of the measured contour with the standard contour, the distance value of the corresponding point of the railhead area of the original measured contour and the standard contour after fine registration is calculated, andin ascending order, i.e. P ₁ 、P ₂ …P _i 、P _i+1 …P _n Then the absolute value of the difference between two adjacent distance values, i.e. | P, is calculated _i -P _i+1 I, will | P _i -P _i+1 Comparing | with a set threshold T5, if the threshold T5 is exceeded, the distance value is P _i+1 The corresponding point of (a) is the boundary position of the outlier and the normal point on the original contour, and all points behind the corresponding point are the outliers, so that the self-adaptive detection and removal of the outliers on the measured contour are realized, and the results are sequentially shown in fig. 8 (a) - (d).

In order to test the effectiveness of the steel rail contour robust registration method based on the constraint ICP algorithm, except for the local shielding case of FIG. 2, a case that a rail web is locally absent and a case that the tail end of a rail foot is interfered by a fastener are extracted to be used as a test sample together. In consideration of the influence of the interference of outliers on the contour registration, before the test, the same method is used to extract the core region corresponding to each sample contour as shown in fig. 9 (b), 9 (d), and 9 (f). Meanwhile, the performance of the test process is compared with that of the traditional DCS and ICP algorithm. The contour registration results for each sample under different methods are shown in fig. 10.

The results show that: for the DCS, since the integrity of the rail web must be guaranteed, it is only applicable to the 3 rd sample, i.e. the condition that the rail foot is interfered and the rail web is complete, which also makes its adaptability in practical application the worst; for ICP it does not require the integrity of the web, so it is also applicable to the case of local absence of the 2 nd sample web, with enhanced adaptability compared to DCS. However, rail foot interference that may occur destroys the non-pollution of the rail web, reducing the registration performance of ICP; in contrast, the CICP algorithm provided by the invention comprehensively considers the influence of the problems on the contour registration, is applicable to 3 samples, shows remarkable superiority, and can be used for robust registration of the steel rail contour under complex working conditions.

The invention provides a CICP-based steel rail profile robust registration algorithm, aiming at the situation that the traditional steel rail section profile registration algorithm cannot be suitable for rail waist local shielding, local deletion and rail foot interference under complex working conditions. Aiming at various situations, the algorithm can realize accurate registration of the measured profile and the standard profile, and greatly improves the reliability and stability of the measuring system.

Its advantages are as follows:

(1) The measured contour is moved to an initial position close enough to the standard contour through rough contour registration, so that the convergence of the CICP algorithm is ensured;

(2) Self-adaptive and global segmentation is carried out on the rail waist through a Ramer algorithm, a local pollution-free area is accurately extracted, and the pollution-free performance of a CICP registration area is guaranteed;

(3) And the contour after fine registration is restrained by the vertical position of the rail jaw point, so that the vertical deviation caused by local registration of the rail web is eliminated, and the accuracy of CICP registration is effectively improved.

The above description is only for the purpose of illustrating the technical solutions of the present invention and not for the purpose of limiting the same, and other modifications or equivalent substitutions made by those skilled in the art to the technical solutions of the present invention should be covered within the scope of the claims of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A steel rail contour robustness registration method based on a constrained iteration closest point method is characterized by comprising the following steps:

step S1: after a threshold T1 of the distance between adjacent points is obtained through statistics on a normal contour curve, firstly splitting a measured contour to obtain a plurality of curve fragments, and then setting a point number threshold T2 according to the characteristic of dense distribution of normal contour data points and carrying out one-to-one inspection on the curve fragments so as to remove sparse outliers;

step S2: setting a segment interval threshold T3, merging curve fragments which belong to the same region and have adjacent intervals not exceeding T3 again, then carrying out concave-convex inspection on each region, and extracting a contour core region based on continuous concave-convex and point number maximum principle;

and step S4: the method comprises the steps of carrying out contour measurement rail waist weight segmentation and local non-pollution area extraction, setting a distance threshold T4 through a Ramer algorithm, and sequentially finding out the positions of all distance values exceeding the threshold T4 from the whole to the local on a contour measurement rail waist curve so as to globally re-segment the contour measurement rail waist curve into n fragments, wherein the first fragment is a local non-pollution area as the local non-pollution area is usually located at the upper part of the rail waist;

step S5: fine registration is carried out on the original measured contour and the standard contour, a plurality of matching characteristic point pairs are searched on the rail web local pollution-free areas of the measured contour and the standard contour according to an Iterative Closest Point (ICP) algorithm, and the matching characteristic point pairs are respectively M = { M = _i I =1,2,l, n } and S = { S = _i I =1,2,L, n }, and calculating an optimal rotation matrix R based on the principle of least sum of squares of distances between matching points after registration ₂ And translation vector T ₂ Which is satisfied with

Realizing fine contour registration;

step S7: self-adaptive detection and removal of outliers of the measured profile after fine registration, namely calculating the distance value between the original measured profile after fine registration and the corresponding point of the railhead area of the standard profile, and performing ascending sorting, namely P ₁ 、P ₂ …P _i 、P _i+1 …P _n Then the absolute value of the difference between two adjacent distance values, i.e. | P, is calculated _i -P _i+1 I, will | P _i -P _i+1 Comparing | with a set threshold T5, if the threshold T5 is exceeded, the distance value is P _i+1 The corresponding point of (1) is the boundary position of the outlier and the normal point on the original contour, and all points after the corresponding point are outliers.

2. The rail contour robust registration method based on the constrained iterative closest point method according to claim 1, wherein in the step S1, T1=2 and T2=10.

3. The method for rail contour robust registration based on the constrained iterative closest point method according to claim 1, wherein T3 is set to 30 in the step S2.

4. The rail contour robust registration method based on the constrained iterative closest point method according to claim 1, wherein in step S3, the specific calculation process of the rotation matrix and the translation vector is:

5. The rail contour robust registration method based on the constrained iterative closest point method according to claim 1, wherein in the step S4, T4 is set to 5.

6. The rail contour robust registration method based on the constrained iterative closest point method according to claim 1, wherein in step S7, T5 is set to 3.