CN102980896B - High ferro overhead contact line device auricle fracture detection method - Google Patents

Abstract

The invention discloses one utilizes the detection method of SIFT (Scale-Invariant Feature Transform, Scale invariant features transform) to detect rotation ears auricle fracture defect.Comprise the following steps: first utilize SIFT algorithm that contact net is connect to support and suspender image and rotates ears template image and carry out local feature Point matching; RANSAC (Random Sample Consensus, random sampling is consistent) algorithm after recycling improvement screens the feature point pairs that the match is successful and realizes rotating the location of ears; Then utilize the method for boundary tracking to extract the coboundary curve rotating ears, draw out the flexibility curve of each point on curve; Finally the flexibility curve drawn out and flexibility curve are under normal circumstances done difference at corresponding point position, judged whether that auricle fracture defect occurs.The inventive method accurately can identify the auricle that fracture defect occurs in the overhead contact line installation drawing picture of complexity, greatly can improve the efficiency of detection compared with the method for artificial examination.

Description

High ferro overhead contact line device auricle fracture detection method

Technical field

The present invention relates to high-speed railway field of image detection, in particular to SIFT (Scale-Invariant Feature Transform, Scale invariant features transform) image invariant feature extraction, Feature Points Matching and screening, the methods such as auricle fracture defect identification.

Background technology

Rotate the junction that ears are positioned at high ferro contact net supporting and location device, be load parts important in net-fault supporting construction, vital effect is played to safe train operation.In the actual operation of railway, often cause auricle fracture defect because of train vibrations, cause contact net support device structure intensity to reduce, time serious, even have the danger that steady arm comes off.Therefore be necessary to detect rotation ears parts, Timeliness coverage also changes trouble unit.

At present, China railways department mainly takes manual type to check elements of contacting net.This method efficiency is lower, can not Timeliness coverage fault, is therefore necessary to study automatic testing method.At present, the Automatic Measurement Technique based on image procossing is applied to some extent on railway, is mainly used in the detection realizing contact net geometric parameter and pantograph wearing and tearing.But not yet occur at present for the fault detection method rotating ears.

Summary of the invention

A kind of high ferro contact net is the object of the present invention is to provide to rotate the detection method of ears and sheet fracture defect.The method utilizes the convergent-divergent of SIFT, rotation, brightness and affine-invariant features, can support with suspender image from the contact net of different shooting environmental, different shooting angles, accurate identification also extracts rotation ears, carries out automatic decision to whether there is auricle fracture defect simultaneously.Its concrete means are:

One utilizes the high ferro contact net bracing or strutting arrangement auricle fracture detection method of Scale invariant features transform SIFT (Scale-Invariant Feature Transform), carry out local feature Point matching, cluster, location and fracture defect to gathered image after utilizing the CCD industrial camera being arranged on inspection vehicle roof to gather contact net support and suspender image to detect, comprise the following steps:

A, CCD industrial camera and shot-light are arranged on inspection vehicle roof, inspection vehicle downline runs, and when running into catenary mast, automatic shooting contact net supports and suspender image, adopts the style of shooting adding shot-light night to disturb to reduce daylight.

B, by A obtain contact net that imagery exploitation SIFT collects at the scene and support and suspender image and rotate between ears standard form image and realize local feature Point matching;

C, by B obtain support with all unique points that the match is successful in suspender image spatially distance carry out cluster, the image-region of the corresponding doubtful rotation ears of each cluster;

Random sampling unification algorism RANSAC (Random Sample Consensus) after D, utilization improve screens the feature point pairs in described cluster, judge whether comprise rotation ears in the image-region corresponding to this cluster, calculation template image and contact net support and rotate the affine transformation matrix between ears region in suspender image, realize rotating ears to support and the location in suspender image at contact net, its algorithm flow is as follows:

A1: randomly draw 3 unique points from described cluster, utilizes formula (1) to calculate contact net support and the affine transformation matrix H between suspender image and template image ₁.

H ₁＝BA ^T(AA ^T) ^-1(1)

In formula, $A=\left(\begin{array}{cccc}{x}_1& x_2& K& x_N\\ y_1& y_2& K& y_N\\ 1& 1& K& 1\end{array}\right),$ $B=\left(\begin{array}{cccc}{x_1}^{\′}& {x_2}^{\′}& K& {x_N}^{\′}\\ {y_1}^{\′}& {y_2}^{\′}& K& {y_N}^{\′}\\ 1& 1& K& 1\end{array}\right),$ N is for calculating H ₁time the number of unique point used, now have N=3, (x _i, y _i), i ∈ [1, N] is contact net support and the coordinate of unique point in suspender image, (x _i', y _i') in template image with (x _i, y _i) coordinate of point that matches, by the S that counts in optimum estimate _ivalue set to 0;

A2: (2) formula of utilization to calculate in this cluster the match point of each point in template image (x ', y ') through H ₁the result that obtains after determined affined transformation (x ", y ").

$\left(\begin{array}{c}{x}^{\′\′}\\ y^{\′\′}\\ 1\end{array}\right)=H_1\left(\begin{array}{c}{x}^{\′}\\ y^{\′}\\ 1\end{array}\right)---\left(2\right)$

If (x ", y ") is less than T with the distance of the point (x, y) in cluster, then (x, y) is judged to be an interior point;

A3: judge H ₁whether corresponding all interior point supports identical with the relative position of unique point that matches with it in template image with the relative position in suspender image at contact net.If not identical, then think and comprise erroneous matching in interior point, if identical, then judge H ₁count whether be greater than in optimum estimate the S that counts in corresponding _iif be greater than S _i, then by H ₁as the current optimum estimate of affine transformation matrix, and upgrade S _i;

A4: return A1) continue to run, after moving to stipulated number, circulation stops, if S _ivalue be not 0, then comprise rotation ears in the image-region thinking representated by this cluster.The current optimum estimate H of affine transformation matrix ₁namely finally H is estimated as it.Bring the coordinate on template image four summits into formula (1), and make H ₁=H, can calculate the region that contact net supports and rotates ears in suspender image;

E, extract and rotate ears subimage from support and suspender image, utilize the method based on coboundary embroidery to detect auricle fracture defect, detection method comprises following steps:

E1: the rotation ears subimage supported from contact net and extract in suspender image is carried out binaryzation, rotation ears are separated with image background.From subimage edge, utilize the method for boundary tracking to obtain the coordinate of each point on the curve of coboundary successively, form coboundary curvilinear coordinates sequence;

E2: on the curve of coboundary, the flexibility of kth point is defined as the inclination angle of it and kth+50 some line, and its computing formula is as follows:

$c_k=\left\{\begin{array}{cc}\arctan \left[\frac{y_{k+50}-y_k}{x_{k+50}-x_k}\right]& \left(\mathrm{atrcan}\right[\frac{y_{k+50}-y_k}{x_{k+50}-x_k}]>0)\\ \arctan \left[\frac{y_{k+50}-y_k}{x_{k+50}-x_k}\right]+\mathrm{\&pi;}& \left(\mathrm{atrcan}\right[\frac{y_{k+50}-y_k}{x_{k+50}-x_k}]<0)\end{array}\right.---\left(3\right)$

In formula, c _krepresent the flexibility of a kth point on the curve of coboundary, x _k, y _k, x _k+50, y _k+50, represent that kth on coboundary curve is put and the transverse and longitudinal coordinate of kth+50 point respectively, c _kunit be radian, span is 0 ~ π.

Calculate the flexibility of each point on the curve of coboundary successively, draw out the flexibility curve of each point on the curve of coboundary;

E3: above-mentioned flexibility curve and flexibility curve are under normal circumstances made difference at corresponding point position, if comprise the point being greater than 0.5 in difference curves, then thinks and there occurs auricle fracture defect, send fracture defect alarm signal.

Beneficial effect of the present invention is:

1, the present invention directly rotates ears parts by image processing method to high ferro contact net and detects, and can obtain objective, testing result accurately, decrease workload and the error of artificial cognition.

2, the present invention makes full use of the convergent-divergent of SIFT, rotation, brightness and affine-invariant features, and testing result is by the impact of shooting environmental and shooting angle, and the accuracy rate of Fault Identification is high.

Accompanying drawing explanation

Fig. 1 is algorithm flow sketch of the present invention.

Fig. 2 is that the contact net comprising auricle fracture defect that ccd video camera photographs supports and suspender image.

Fig. 3 is the rotation ears template image for local feature coupling.

Fig. 4 is the feature points clustering figure (Fig. 4 a, Fig. 4 b, Fig. 4 c represent a cluster respectively) that the match is successful.

Fig. 5 is the rotation ears subimage supporting from contact net and extract suspender image.

Fig. 6 is the rotation ears coboundary curve utilizing boundary tracking process to obtain.

Fig. 7 is the flexibility curve of the coboundary curve shown in Fig. 6.

The difference curves that Fig. 8 obtains when corresponding point position does difference with flexibility curve under normal circumstances for the flexibility curve shown in Fig. 7.

Fig. 9 is the rotation ears that there is not auricle fracture defect, the flexibility difference curves utilizing same method to obtain.

Embodiment:

Below in conjunction with accompanying drawing, embodiments of the present invention are described in further detail.

SIFT algorithm is utilized to carry out local feature Point matching to the rotation ears template image shown in the contact net support shown in Fig. 2 and suspender image and Fig. 3.Matching result as shown in Figure 3.The unique point that the match is successful mainly concentrates on 3 regions, forms 3 clusters altogether, as shown in Fig. 4 a, Fig. 4 b and Fig. 4 c.In figure, the large figure of left-half is contact net support to be analyzed and suspender image, and the little figure in the upper right corner rotates ears template images, and dark green line represents the coupling of local feature region in two width figure.

RANSAC algorithm after improving the cluster in Fig. 4 a, Fig. 4 b and Fig. 4 c successively carries out error hiding elimination, and count in the optimum estimate that the cluster in Fig. 4 b and Fig. 4 c finally obtains S _ibe 0, therefore think and do not comprise rotation ears in the image-region representated by them.Count in the optimum estimate that Fig. 4 a obtains after carrying out error hiding elimination S _ibe not 0, therefore think and comprise rotation ears in the image-region representated by it.Bringing the final estimated result H of affine transformation matrix and Prototype drawing formula (1) into as four apex coordinates and can realize rotating ears and support and the location in suspender image at contact net.Contact net supports and rotates the subimage in binaural localization region as shown in Figure 5 in suspender image.

$\left(\begin{array}{c}{x}^{\&prime;\&prime;}\\ y^{\&prime;\&prime;}\\ 1\end{array}\right)=H\left(\begin{array}{c}{x}^{\&prime;}\\ y^{\&prime;}\\ 1\end{array}\right)---\left(1\right)$

By support from contact net with the rotation ears subimage binaryzation that extracts in suspender image after, the rotation ears coboundary curve utilizing boundary tracking process to obtain.Result as shown in Figure 6.The flexibility of each point on the curve of coboundary is calculated according to formula (2).Draw flexibility curve as shown in Figure 7.

$c_k=\left\{\begin{array}{cc}\arctan \left[\frac{y_{k+50}-y_k}{x_{k+50}-x_k}\right]& \left(\mathrm{atrcan}\right[\frac{y_{k+50}-y_k}{x_{k+50}-x_k}]>0)\\ \arctan \left[\frac{y_{k+50}-y_k}{x_{k+50}-x_k}\right]+\mathrm{\&pi;}& \left(\mathrm{atrcan}\right[\frac{y_{k+50}-y_k}{x_{k+50}-x_k}]<0)\end{array}\right.---\left(2\right)$

Flexibility curve in Fig. 7 and flexibility curve are under normal circumstances done difference in corresponding point, and result as shown in Figure 8.As can be seen from Figure 8, difference curves comprise the peak value that two are greater than threshold value 0.5, therefore judge that these rotation ears exist auricle fracture defect.

For the rotation ears that there is not auricle fracture defect, the flexibility difference curves utilizing same method to obtain usually as shown in Figure 9, on curve value a little all in threshold value less than 0.5.

Claims (1)

1. a high ferro overhead contact line device auricle fracture detection method, Scale invariant features transform SIFT (Scale-Invariant Feature Transform) is utilized to detect the fracture of high ferro contact net bracing or strutting arrangement auricle, carry out local feature Point matching, cluster, location and fracture defect to gathered image after utilizing the CCD industrial camera being arranged on inspection vehicle roof to gather contact net support and suspender image to detect, comprise the following steps:

A, CCD industrial camera and shot-light are arranged on inspection vehicle roof, inspection vehicle downline runs, and when running into catenary mast, automatic shooting contact net supports and suspender image, adopts the style of shooting adding shot-light night to disturb to reduce daylight;

B, by A obtain contact net that imagery exploitation SIFT collects at the scene and support and suspender image and rotate between ears standard form image and realize local feature Point matching;

C, by B obtain support with all unique points that the match is successful in suspender image spatially distance carry out cluster, the image-region of the corresponding doubtful rotation ears of each cluster;

Random sampling unification algorism RANSAC (Random Sample Consensus) after D, utilization improve screens the feature point pairs in described cluster, judge whether comprise rotation ears in the image-region corresponding to this cluster, simultaneously calculation template image and contact net supports and the affine transformation matrix that rotates in suspender image between ears region, realize rotating ears to support and the location in suspender image at contact net, its algorithm flow is as follows:

D1: by the S that counts in optimum estimate _ivalue set to 0;

D2: randomly draw 3 unique points from described cluster, utilizes formula (1) to calculate contact net support and the affine transformation matrix H between suspender image and template image ₁;

H ₁＝BA ^T(AA ^T) ^-1(1)

In formula, $A=\left(\begin{array}{cccc}{x}_1& x_2& ...& x_N\\ y_1& y_2& ...& y_N\\ 1& 1& ...& 1\end{array}\right),$ $B=\left(\begin{array}{cccc}{x_1}^{\&prime;}& {x_2}^{\&prime;}& ...& {x_N}^{\&prime;}\\ {y_1}^{\&prime;}& {y_2}^{\&prime;}& ...& {y_N}^{\&prime;}\\ 1& 1& ...& 1\end{array}\right),$ N is for calculating H ₁time the number of unique point used, now have N=3, (x _i, y _i), i ∈ [1, N] is contact net support and the coordinate of unique point in suspender image, (x _i', y _i') in template image with (x _i, y _i) coordinate of point that matches;

D3: (2) formula of utilization to calculate in this cluster the match point (x', y') of each point in template image through H ₁the result that obtains after determined affined transformation (x ", y ");

$\left(\begin{array}{c}{x}^{\&prime;\&prime;}\\ y^{\&prime;\&prime;}\\ 1\end{array}\right)=H_1\left(\begin{array}{c}{x}^{\&prime;}\\ y^{\&prime;}\\ 1\end{array}\right)---\left(2\right)$

If (x ", y ") be less than threshold value T with the distance of the point (x, y) in cluster, then (x, y) is judged to be an interior point;

D4: judge H ₁whether corresponding all interior point supports identical with the relative position of unique point that matches with it in template image with the relative position in suspender image at contact net; If not identical, then think and comprise erroneous matching in interior point, if identical, then judge H ₁count whether be greater than in optimum estimate the S that counts in corresponding _iif be greater than S _i, then by H ₁as the current optimum estimate of affine transformation matrix, and upgrade S _i;

D5: return D2 and continue to run, after moving to stipulated number, circulation stops, if S _ivalue be not 0, then comprise rotation ears in the image-region thinking representated by this cluster; The current optimum estimate H of affine transformation matrix ₁namely finally H is estimated as it; Bring the coordinate on template image four summits into formula (1), and make H ₁=H, can calculate the region that contact net supports and rotates ears in suspender image;

E, extract and rotate ears subimage from support and suspender image, utilize the method based on coboundary embroidery to detect auricle fracture defect, detection method comprises following steps:

E1: the rotation ears subimage supported from contact net and extract in suspender image is carried out binaryzation, rotation ears are separated with image background; From subimage edge, utilize the method for boundary tracking to obtain the coordinate of each point on the curve of coboundary successively, form coboundary curvilinear coordinates sequence;

E2: on the curve of coboundary, the flexibility of kth point is defined as the inclination angle of it and kth+50 some line, and its computing formula is as follows:

$c_k=\left\{\begin{array}{cc}\arctan \left[\frac{y_{k+50}-y_k}{x_{k+50}-x_k}\right]& \left(\mathrm{atrcan}\right[\frac{y_{k+50}-y_k}{x_{k+50}-x_k}]>0)\\ \arctan \left[\frac{y_{k+50}-y_k}{x_{k+50}-x_k}\right]+\mathrm{\&pi;}& \left(\mathrm{atrcan}\right[\frac{y_{k+50}-y_k}{x_{k+50}-x_k}]<0)\end{array}\right.---\left(3\right)$

In formula, c _krepresent the flexibility of a kth point on the curve of coboundary, x _k, y _k, x _k+50, y _k+50, represent that kth on coboundary curve is put and the transverse and longitudinal coordinate of kth+50 point respectively, c _kunit be radian, span is 0 ~ π;

Calculate the flexibility of each point on the curve of coboundary successively, draw out the flexibility curve of each point on the curve of coboundary;

E3: above-mentioned flexibility curve is poor at corresponding point position with flexibility curve under normal circumstances, if comprise the point being greater than 0.5 in difference curves, then thinks and there occurs auricle fracture defect, send fracture defect alarm signal.