CN110378879B

CN110378879B - Bridge crack detection method

Info

Publication number: CN110378879B
Application number: CN201910560732.9A
Authority: CN
Inventors: 张巨勇; 王云; 周洪强; 何凯; 陈志平; 李蓉
Original assignee: Hangzhou Dianzi University
Current assignee: Hangzhou Dianzi University
Priority date: 2019-06-26
Filing date: 2019-06-26
Publication date: 2021-03-02
Anticipated expiration: 2039-06-26
Also published as: CN110378879A

Abstract

The invention discloses a bridge crack detection method. Because the manual marking has certain subjectivity, the pavement crack detection precision depends on the experience knowledge of experts, and the experience lacks objectivity in quantitative analysis. The invention is as follows: firstly, acquiring images of the detected bridge deck one by one to obtain a bridge deck image set. And secondly, image splicing. And thirdly, detecting cracks. And fourthly, extracting fracture parameters. The method disclosed by the invention replaces human eyes to finish the automatic nondestructive detection of the bridge cracks by an image acquisition and processing technology, and has very important practical significance for the research of the bridge crack detection technology in a complex terrain environment. On one hand, the construction safety is enhanced, and on the other hand, the operation maneuverability and flexibility are improved. The invention realizes the fidelity splicing of a plurality of groups of bridge deck images, improves the image splicing precision and efficiency, lays a work foundation for the subsequent bridge crack image detection, and provides a technical reference for the image splicing detection in other fields.

Description

Bridge crack detection method

Technical Field

The invention belongs to the technical field of image detection, and particularly relates to a bridge crack detection method.

Background

The bridge is built, the traffic is improved, and the development of national economy and society of China is powerfully promoted. The bridge is used as an important carrier for connecting two position points with larger span, and the generated internal stress can be transmitted to weak parts along the bridge structure in long-term sun-rain and load operation, so that cracks are easy to generate and develop on the surface of the position structure. The damage degree of the surface cracks with different trends to the bridge structure is also different, and the damage influence is the largest if the extension trend of the surface cracks is perpendicular to the bearing surface of the structure.

Engineering practice and theoretical analysis show that most of bridges in service work with cracks, and potential hazards caused by the cracks of the bridges are not inconstant. In case comparatively serious crack appears in the concrete bridge, outside air and harmful medium can permeate the inside production carbonate through chemical reaction of concrete very easily, cause the basicity environment of reinforcing bar wherein to reduce, and the purification membrane on surface is suffered to destroy the back and is changeed and produce the corrosion, and in addition, concrete carbonization also can aggravate the shrinkage crack, produces serious harm to the safe handling of concrete bridge. As the most common disease characteristic in bridge construction, extremely fine cracks (smaller than 0.05mm) generally have little influence on the structural performance and can be allowed to exist; larger cracks can be continuously generated and expanded under the action of load or external physical and chemical factors to form through seams and deep seams, and indirectly or even directly influence the service life and the safety performance of the beam structure; if the width of the crack reaches more than 0.3mm, the integrity of the structure can be directly damaged, concrete carbonization, protective layer peeling and steel bar corrosion are caused, and mechanical discontinuities are formed in the bridge, so that the bearing capacity of the bridge is greatly reduced, and even collapse accidents happen in severe cases, and the normal use of the structure is damaged.

Therefore, bridge construction damage is greatly related to the generation and development of surface cracks. In order to find the cracks in time and take remedial measures to eliminate potential safety hazards, manual inspection and manual marking are generally adopted, and surface cracks are manually measured and recorded by experienced inspectors through visual observation. However, the detection mode has poor mobility, high risk and low efficiency, and because the manual marking has certain subjectivity, the detection precision depends on the experience knowledge of experts, and the experience lacks objectivity in quantitative analysis.

Disclosure of Invention

The invention aims to provide a bridge crack detection method.

The method comprises the following specific steps:

step1, collecting the images of the detected bridge deck one by one to obtain a bridge deck image set.

Step2, image splicing

2-1. mosaic image preprocessing

Extracting and counting the brightness component information of each bridge deck image in the bridge deck image set, and respectively equalizing the brightness component information of each bridge deck image; and then transforming the bridge deck images into a frequency domain through Fourier transform, and obtaining translation parameters among the images by adopting phase information of a normalized cross-power spectrum in a phase correlation algorithm to complete pre-estimation of an overlapping area between adjacent images.

2-2 image registration

Firstly, extracting SIFT feature points in an overlapping area between every two adjacent images; then sifting SIFT feature points by a self-adaptive contrast threshold method to obtain a feature descriptor consisting of matching point pairs; and calculating a projective transformation matrix between every two adjacent images by using a RANSAC algorithm.

2-3. image fusion

Firstly, performing projection transformation on corresponding bridge deck images according to projection transformation matrixes between adjacent images; and then, respectively carrying out weighted smooth transition on the RGB three-color channels of each adjacent bridge deck image by adopting a gradual-in and gradual-out fusion algorithm to obtain a bridge deck spliced image.

Step3, crack detection

And 3-1, carrying out initial positioning and graying of a crack region through crack detection pretreatment.

And carrying out gray processing on the bridge deck splicing image. And dividing the bridge deck splicing image into a plurality of local areas through uniform grid division. Counting the gray accumulated value in each local area to obtain the gray distribution of all grid areas and extracting the average gray value of the bridge deck image as a gray threshold; and screening out a grid area with the gray accumulated value lower than the gray threshold value as an interested area.

And 3-2, performing median filtering denoising and image enhancement processing based on a fuzzy set on each region of interest to obtain a plurality of crack gray level images to be segmented.

And 3-3, performing binary segmentation on each crack image to be segmented through a PCNN simplified model to obtain a crack segmentation image.

And 3-4, removing residual isolated noise of the crack segmentation image to obtain a crack binary image.

Step4, extracting crack parameters

And 4-1, performing fracture characteristic connection based on seed point growth on the fracture binary image obtained in the step 3.

4-2, classifying and identifying the target cracks based on a projection method, which comprises the following specific steps:

firstly, carrying out inversion operation on the fracture binary image processed in the step 4-1. And respectively projecting the crack binary image in the horizontal direction and the vertical direction by adopting a projection method, and respectively accumulating and summing pixel values of each row and column to obtain a row projection array and a column projection array.

According to the actual size characteristics of the cracks in the crack binary image, calculating the length-width ratio of the crack part in the crack binary image

And delta x is the horizontal projection length of the crack part in the crack binary image, and delta y is the vertical projection length of the crack part in the crack binary image.

If it is

Judging the cracks in the crack binary image to be transverse cracks; if it is

Judging the cracks in the crack binary image to be longitudinal cracks; if it is

And all elements in the row projection array and the column projection array are more than 10, judging that the cracks in the crack binary image are net-shaped cracks; otherwise, judging that the cracks in the crack binary image are oblique cracks;

and 4-3, extracting bridge deck crack characteristic data to obtain the length, width and area information of the actual bridge cracks.

(1) And extracting a fracture skeleton line on the fracture in the fracture binary image.

And counting the number of pixel points with the numerical value of 1 in the crack binary image as the crack pixel area. And extracting a fracture skeleton line according to the fracture image on the fracture binary image. And taking the number of pixel points on the crack skeleton line as the length of the crack pixel.

The method for calculating the width of the bridge crack comprises the following steps:

(1) setting a 5 multiplied by 5 retrieval template, sweeping all crack pixel points on a crack skeleton line, taking two crack pixel points with the farthest distance in the retrieval template, and taking a connecting line of the two crack pixel points as a local trend line of cracks in the template. And then the normal line of the local trend line of each crack obtained after sweeping is carried out, and the normal angle is calculated.

(2) According to the normal angle of each crack pixel point on the crack skeleton line, the corresponding two-side edge points (x) of each crack pixel point on the crack are obtained_ij,y_ij) And (x'_ij,y′_ij) (ii) a The Euclidean distance of the corresponding edge points at two sides of the crack pixel point on the crack is taken as the crack pixel width d on the position point_ijI.e. by

The obtained width d of each slit pixel_ijThe maximum value of (3) is taken as the maximum pixel width of the crack. Width d of each slit_ijThe average value in (a) is taken as the average pixel width of the crack.

(3) Respectively substituting each pixel parameter A' of the crack into an actual parameter conversion formula

And obtaining each actual parameter A of the crack. Wherein u is the object distance, f is the focal length, k is the conversion coefficient, L is the long-side physical dimension of the CCD photosensitive chip, and L is the number of pixels on the long side of the shot image. The pixel parameters of the crack include crack pixel length, crack maximum pixel width, crack average pixel width, and crack pixel area.

Preferably, the method for acquiring the image in step1 specifically comprises the following steps:

1-1, calculating an internal reference matrix of the CCD camera by adopting a Zhangyingyou plane calibration method, and then obtaining a radial distortion coefficient by a least square method.

And 1-2, arranging the CCD camera calibrated in the step 1-1 on a bridge detection platform, and acquiring an image of the complete bridge floor according to a preset shooting track. The preset shooting track is S-shaped.

And 1-3, respectively carrying out image calibration on each bridge deck image acquired in the step 1-2 according to the internal reference matrix and the distortion coefficient of the CCD camera obtained in the step 1-1.

Preferably, in step 2-2, the adaptive contrast threshold method is specifically as follows:

(1) setting a lower limit N of the number of characteristic points _min200, upper limit N _max300, contrast threshold T_c＝T₀；T₀The initial threshold value is 0.02-0.04.

(2) Detecting the characteristic points and counting that the contrast is higher than T_cN.

(3) If N is present_min≤N≤N_maxThen contrast will be higher than T_cThe characteristic points are brought into the initial matching point set, and the rejection contrast is lower than a threshold value T_cAnd directly entering the step (5); otherwise, executing step (4).

(4) If N is less than N_minThen the contrast threshold T is set_cReduced to the original value

And executing the step (3); if N > N_maxThen the contrast threshold is increased to 2 times the original value and step (3) is performed.

(5) And eliminating error feature points in the initial matching point set by a nearest neighbor to second nearest neighbor method, and generating a feature descriptor. The feature descriptor includes a plurality of matching point pairs each composed of a pair of feature points, and distance and direction information between the matching point pairs.

Preferably, in step 2-2, the process of solving the projective transformation matrix by the RANSAC algorithm is as follows:

(1) an initial sample set S is constructed with each matching point pair within the feature descriptor. Counting Euclidean distances between each matching point pair in the initial sample set S, and sequencing from small to large;

(2) taking the first 85% of matching point pairs of the sequence obtained in the step (1) to construct a new sample set S';

(3) randomly extracting 4 groups of matching point pairs from the new sample set S' to form an inner point set S_iAnd calculating the set S of the points in the matrix model_iH of (A) to (B)_iEntering the step (4);

(4) the rest matching point pairs in the new sample set S' are corresponding to the matrix model H_iCarrying out adaptability test; if the matched points with the detection errors smaller than the error threshold exist, adding the matched point pairs with the detection errors smaller than the threshold into the inner point set S_iAnd executing the step (5); otherwise, the matrix model H is discarded_iAnd (3) is re-executed.

(5) If the inner point set S_iIf the number of the middle elements is larger than the specified threshold value, a reasonable parameter model is considered to be obtained, and the updated interior point set S is subjected to_iRecalculating matrix model H_iAnd minimizing the cost function by using an LM algorithm; otherwise, the matrix model H is discarded_iAnd re-executing the step (3).

(6) Repeating steps (3) to (5) for l times, wherein l is the maximum iteration number. Then, comparing the inner point set S obtained in the iteration of the time I_iSet S of interior points with the largest number of elements_iTaking the matrix model H as the final internal point set and calculating_iAs a projective transformation matrix between adjacent bridge deck images.

Preferably, in step 2-3, the detailed steps of projective transformation are as follows:

(1) and according to the transmissibility of the projective transformation matrix between the adjacent images, respectively taking the first bridge deck image of each row as the reference image of the corresponding row for splicing. For the transformation matrix H between the adjacent bridge floor images_ii-1Carrying out transmission transformation to obtain a transmission transformation matrix H between each bridge deck image and the reference image_i1(ii) a Then through each transformation matrix H_i1Mapping the corresponding bridge deck images into a reference plane coordinate system respectively to finish Image splicing and fusion between every two adjacent images in the horizontal direction to form a plurality of transverse panoramic Image images with wide visual angles_i。

(2) The first horizontal panoramic Image obtained in the step (1) is processed₁And splicing the images as reference panoramic images. For transformation matrix T between horizontal panoramic images_jj-1Performing transfer transformation to obtain horizontal panoramic images_iTransfer transformation matrix T with reference panoramic image_j1(ii) a Transforming the matrix T by each transfer_j1And respectively mapping the corresponding transverse panoramic images into a reference plane coordinate system to complete image splicing and fusion between every two adjacent transverse panoramic images in the vertical direction to form a final bridge deck panoramic image.

In step 2-3, in the fade-in and fade-out fusion algorithm, a fade-in and fade-out weighting formula of each fusion point pixel value I (x, y) in the overlapping region of adjacent images is as follows:

wherein, I₁(x,y)、I₂And (x, y) are pixel values of corresponding fusion points of two adjacent bridge deck images in the overlapping area respectively. d₁、d₂Respectively is a gradual change weight factor of two adjacent bridge deck images at the corresponding fusion point;

and

x₁、x₂respectively are the horizontal coordinates of the boundaries at the two sides of the overlapping area; x is the abscissa of the corresponding fusion point; and t is the gray level difference threshold value of the overlapped area of the two adjacent images on the corresponding fusion point.

Preferably, the median filtering and denoising step in step 3-2 is as follows:

(1) traversing each pixel point in the interested region to be respectively used as a target pixel point f_ijAnd performing steps (2) to (4) respectively.

(2) Respectively aiming at the target pixel point f_ijAll pixel points and target pixel points f in eight neighborhoods_ijComparing the gray values, and taking 2 neighborhood pixel points with the minimum gray difference absolute value

And

(3) will be provided with

And

as the adjacent gray scale extension direction of the target pixel point, respectively extending the target pixel point to the eight neighborhoods by one level in the two directions to obtain f_pAnd f_qTwo pixel points;

(4) get target pixel point f_ijAll pixel points and pixel points f in eight neighborhoods_pAnd pixel point f_qThe median value of (a) is taken as a target pixel point f_ijThe replacement value of (a).

Preferably, in step 3-3, a specific method for performing binary segmentation on each fracture image to be segmented by using the PCNN simplified model is as follows:

step 1: initializing PCNN model parameters

(1) Inputting grey values of all normalized pixel points (I, j) of a crack image to be segmented as an external stimulation signal I_ij。

(2) Setting decay time constant delta t as 0.2 and maximum information entropy value H _max0, 30 maximum iteration number n and link strength coefficient matrix

(3) An optimal threshold value obtained by a two-dimensional Otsu method is used as an initial threshold value theta of the PCNN model_ij[0](ii) a Determining a link strength factor beta by a local gray variance method_ij。

Step 2: value k is assigned to 1.

Step 3: carrying out segmentation iteration on a crack image to be segmented by adopting a PCNN simplified model to obtain a binary segmentation image Y (k); computing each neuron in a PCNN simplified modelInternal activation U_ijAnd pulse output Y_ijAnd by comparing U_ijAnd Y_ijThe size of the neuron determines the activation state of each neuron.

Step 4: calculating information entropy value H (k) corresponding to the binary segmentation image Y (k), if H (k) is more than H_maxThen, assign H (k) to H_maxAnd taking the binary segmentation image Y (k) as a new preferred segmentation image Y(s). Otherwise, step5 is entered directly.

Step 5: if k is less than n, increasing k by 1, and then repeatedly executing the steps of Step3 and Step 4; otherwise, outputting the current preferred segmentation image Y(s) as a fracture segmentation image.

Preferably, in step 3-4, the residual isolated noise includes discrete noise points and concentrated noise spots.

The method for removing the discrete noise point specifically comprises the following steps:

(1) and marking the connected regions divided by each edge in the crack segmentation image obtained in the step 3-3.

(2) Respectively calculating the pixel area of each connected region in the optimal crack segmentation image; rejecting connected regions with pixel areas smaller than an area threshold; the area threshold is 20.

(3) And (3) searching by adopting linear segment scanning: setting the length of the line segment template to be 5 pixels, scanning and detecting the boundary line of the communication area reserved in the step (2) from four directions of 0 degrees, 45 degrees, 90 degrees and 135 degrees respectively, and carrying out 'AND' operation. And if the straight line segment scanning retrieval return values of the boundary line of one connected region in the directions of 0 degree, 45 degrees, 90 degrees and 135 degrees are all 0, rejecting the connected region.

The method for removing the aggregated noise spots comprises the following specific steps:

(1) the circularity e (x, y) of each connected region is calculated as follows:

in the formula, C (x, y) and a (x, y) are the perimeter and area of the corresponding connected region, respectively.

If a communicating zoneThe circularity e (x, y) of the domain is above a circularity threshold T_e(x, y), then the connected region is eliminated, T_e(x,y)＝0.3。

(2) The aspect ratio r (x, y) of each connected region is calculated as follows:

in the formula, L (x, y) and W (x, y) are the length and width of the projection matrix of the corresponding connected region in the horizontal direction and the vertical direction respectively.

Setting T_rl(x, y) is the lower limit of the length-width ratio, and the value is 0.1; t is_rh(x, y) is the upper limit of the aspect ratio, and is 10. If the aspect ratio r (x, y) of a connected region satisfies the following inequality: t is_rl(x,y)＜r(x,y)＜T_rh(x, y), then the connected region is culled.

Preferably, the fracture joining method described in step 4-1 is as follows:

(1) and mapping the position information of the fracture break in the fracture binary image to the fracture gray level image obtained in the step 3-2.

(2) And respectively selecting an initial seed point and a guide seed point at the opposite ends of the two sections of cracks corresponding to the fracture.

(3) The initial seed point is a first growing point for growing; and the next growth point is a pixel point with the minimum gray value in the eight neighborhoods of the previous growth point until the grown growth point is coincided with the guide seed point.

(4) And (4) mapping each growth point obtained in the step (3) to a crack binary image from the crack gray image.

Preferably, after the step 4-3 is executed, the sum of the crack pixel areas in all the crack binary images is divided by the number of pixels of the bridge deck panoramic image to obtain the face crack rate.

The invention has the beneficial effects that:

1. the method disclosed by the invention replaces human eyes to finish the automatic nondestructive detection of the bridge cracks by an image acquisition and processing technology, and has very important practical significance for the research of the bridge crack detection technology in a complex terrain environment. On one hand, the construction safety is enhanced, and on the other hand, the operation maneuverability and flexibility are improved.

2. Aiming at the problems of large operation amount and insufficient precision of the traditional SIFT feature image splicing algorithm, the invention provides an improved large-area image splicing algorithm for more completely and accurately extracting bridge deck image crack feature data, realizes fidelity splicing of multiple groups of bridge deck images, improves the image splicing precision and efficiency, lays a work foundation for subsequent bridge crack image detection, and provides a technical reference for image splicing detection in other fields.

3. The invention aims at the unique space ductility and the gray level distinguishability of the bridge crack, and the improved median filtering algorithm is provided to introduce the idea of similar gray level expansion direction on the basis of the original median filtering according to the characteristic of the linear distribution of the similar gray level of the crack edge, thereby not only effectively inhibiting the interference of bridge deck image composite noise and improving the image signal-to-noise ratio, but also keeping the continuity of crack edge detail information, increasing the reliability of target crack fidelity filtering and improving the anti-noise performance of the crack edge feature extraction algorithm.

4. The optimal crack segmentation image obtained by the improved PCNN crack detection algorithm not only ensures complete and clear edge contour, but also retains more accurate detail texture characteristics, and effectively eliminates background noise interference accompanying most crack image segmentation. Compared with the traditional image segmentation algorithm, the improved pulse coupling neural network algorithm reduces the PCNN parameters needing to be set manually, greatly improves the anti-interference performance and the segmentation performance stability in the bridge crack image segmentation under the complex background, and has better robustness.

5. The invention provides a filtering criterion aiming at discrete noise points and aggregated noise spots in the bridge deck image, can further eliminate isolated noise areas in the binary image after the gray threshold segmentation processing, thoroughly eliminate irrelevant noise information in the bridge deck binary image, completely reserve the edge information of the target crack, and ensure the accuracy and precision of the subsequent crack characteristic data extraction.

6. According to the invention, algorithm designs of splicing a plurality of groups of bridge deck images, processing large-area crack images and extracting crack characteristic data are respectively carried out according to the acquired bridge crack image morphology and the pixel distribution characteristics, and reliability analysis research is carried out on key calculation in the algorithm designs, so that the method has higher algorithm innovation and bridge detection engineering reference value.

Drawings

FIG. 1 is a schematic diagram of a bridge crack detection method according to the present invention;

FIG. 2 is a schematic diagram of a bridge deck image acquisition trajectory in the present invention;

FIG. 3 is a flow chart of a process for stitching a plurality of bridge images according to the present invention;

FIG. 4 is a flow chart of adaptive contrast threshold calculation in the present invention;

FIG. 5 is a schematic diagram of a line-by-line image stitching strategy according to the present invention;

FIG. 6 is a schematic diagram of a column-by-column image stitching strategy according to the present invention;

FIGS. 7a and 7b are schematic diagrams of weighted fusion between adjacent images according to the present invention;

FIG. 8 is a schematic view of the initial positioning of the fracture zone in the present invention;

FIGS. 9a and 9b are schematic diagrams of improved median filtering templates in the present invention;

FIG. 10 is a schematic diagram of a simplified PCNN model according to the present invention;

FIG. 11 is a flow chart of an improved PCNN fracture image segmentation algorithm of the present invention;

FIGS. 12a and 12b are schematic views of crack connection based on a seed point growth method in the present invention;

FIG. 13 is a schematic illustration of the calculation of local normal direction within 5 by 5 templates on a fracture skeleton according to the present invention;

FIG. 14 is a schematic view illustrating the calculation of the width of a bridge crack in the present invention.

Detailed Description

The invention is further described below with reference to the accompanying drawings.

As shown in fig. 1, a bridge crack detection method specifically includes the following steps:

step1, image acquisition

1-1, adopting a Zhangyingyou plane calibration method, using a CCD camera to sample images of a calibration plate from different angles, calculating an internal reference matrix of the CCD camera through the detected coordinates of the checkerboard corner points of the calibration plate, and then obtaining a radial distortion coefficient through a least square method.

And 1-2, arranging the CCD camera calibrated in the step 1-1 on a bridge detection platform, and acquiring a plurality of groups of images of the complete bridge floor according to a preset shooting track. The preset shooting trajectory is S-shaped as shown in fig. 2.

And 1-3, respectively carrying out image calibration on each bridge deck image acquired in the step 1-2 according to the internal reference matrix and the distortion coefficient of the CCD camera obtained in the step 1-1 so as to eliminate distortion influence caused by lens distortion.

Step2, image splicing

2-1. mosaic image preprocessing

The size of the panoramic image is first pre-estimated. The size is taken according to the resolution and the number of the images to be spliced, and the invalid area is removed after splicing is completed; then, brightness component information of all bridge deck images to be spliced is extracted and counted, and the brightness component information of each bridge deck image is equalized respectively, so that the brightness difference influence caused by uneven illumination is eliminated; and finally, transforming the bridge deck images into a frequency domain through Fourier transform, and obtaining translation parameters among the images by adopting the phase information of the normalized cross-power spectrum in a phase correlation algorithm so as to complete pre-estimation of an overlapping area between adjacent images.

The phase correlation algorithm is that Fourier transform is adopted to firstly transform the images to be spliced into a frequency domain, and then translation parameters of the two images are calculated through a normalized cross-power spectrum, so that a two-dimensional impulse function is obtained: the peak value of the two-dimensional impulse function reflects the content correlation between the adjacent bridge deck images, the value of the peak value is 1, which indicates that the two images are completely the same, and the value of the peak value is 0, which indicates that the two images are completely different. Although the energy of the impulse function is dispersed from a single peak to a plurality of small peaks due to changes caused by perspective transformation and position movement between adjacent images acquired by the bridge image detection device, the translation parameter corresponding to the maximum peak position is still relatively stable. Therefore, the translation amount obtained by the phase correlation algorithm can roughly acquire the overlapping area between the images to be spliced, the algorithm is insensitive to the change of the light illumination degree, and the detected maximum peak point of the correlation has better robustness and stability.

2-2 image registration

Firstly, SIFT feature points are extracted from the overlapping area between every two adjacent images so as to reduce a large amount of unnecessary feature point detection calculation amount and improve the detection efficiency of the SIFT feature points; then, controlling the number of detected SIFT feature points in a reasonable range by a self-adaptive contrast threshold method so as to screen out stable feature descriptors consisting of matching point pairs; and a modified RANSAC algorithm (random sample consensus algorithm) is adopted to calculate a projective transformation matrix H between every two adjacent images.

The adaptive contrast threshold method is specifically as follows:

after determining the overlapping area (delta x, delta y) between every two adjacent bridge deck images, SIFT feature point detection is carried out only on the overlapping area. Because the characteristic point with low contrast is sensitive to the bridge floor background noise, a contrast threshold value is set, and a stable characteristic point set is screened out and marked as C.

In the prior art, the contrast of each SIFT feature point is calculated by a gaussian difference taylor expansion, and a fixed contrast threshold is set to retain feature points higher than the contrast threshold as stable feature points. However, the contrast threshold T is_cThe value is a fixed value, and is generally between 0.02 and 0.04. However, in crack image detection of different concrete bridges, candidate feature point sets detected by SIFT have great difference, the surfaces of partial bridges are smooth and clean, acquired image digital signals are smooth, and a scale space factor sigma is small, so that the detected feature points are few, and the number requirement of feature point matching can not be met, and the final splicing precision is influenced. (contrast thresholding step used in conventional stitching algorithms). The invention sets a variable contrast threshold to ensure that the detected SIFT feature point number is controlled within a reasonable range. Multiple groups of experimental verification show that the characteristic points of the bridge crack image detection are kept between 200 and 300And better splicing precision can be met.

As shown in FIG. 4, the contrast threshold T is determined in the present invention_cThe method specifically comprises the following steps:

And after matching the characteristic points between the overlapping areas of the adjacent images, screening enough matching point pairs, and solving a transformation matrix between the bridge crack sequence images through the matching point pairs so as to complete the splicing of the bridge deck images in a large range. In order to further improve the image registration efficiency and precision, the RANSAC algorithm is improved.

The improved RANSAC algorithm solves the projective transformation matrix H by the following process:

(6) Repeating steps (3) to (5) for l times, wherein l is the maximum iteration number. Then, comparing the inner point set S obtained in the iteration of the time I_iSet S of interior points with the largest number of elements_iTaking the matrix model H as the final internal point set and calculating_iAs a projective transformation matrix H between adjacent bridge floor images.

The improved RANSAC algorithm not only reduces the sample set data of the point pairs to be matched and improves the proportion of local points in the sample set, but also reduces the iterative refining times of a projection transformation matrix by calculating the Euclidean distances among all the matched point pairs and performing sequencing and screening so as to improve the matching precision of the bridge deck image. According to the characteristic that the matching similarity is higher when the distance between the image feature point pairs is smaller, the Euclidean distance between all the feature point pairs is calculated and is arranged in the order from small to large for screening. Statistics of splicing test results of multiple groups of bridge deck images shows that after initial matching of the overlapping region feature point pairs, the successful matching rate of the initial sample set S can reach more than 85%, and the feature point pairs of the first 85% of the sequence are taken to construct a new sample set S'. Through sample data screening, the sample set S' contains enough matching point pairs, so that the proportion of local points in the sample set is increased, and the iteration times of the transformation matrix parameter model H are greatly reduced.

2-3. image fusion

The projection transformation comprises the following specific steps:

(1) as shown in fig. 5, based on the transmissibility of the projective transformation matrix between adjacent images, the first bridge deck image of each row is used as the reference image of the corresponding row, and the images are spliced according to the image row splicing strategy. For the transformation matrix H between the adjacent bridge floor images_ii-1Carrying out transmission transformation to obtain a transmission transformation matrix H between each bridge deck image and the reference image_i1(ii) a Then through each transformation matrix H_i1Mapping the corresponding bridge deck images into a reference plane coordinate system respectively to finish Image splicing and fusion between every two adjacent images in the horizontal direction to form a plurality of transverse panoramic Image images with wide visual angles_i。

Each transfer matrix transformation formula is as follows:

H₂₁＝H₂₁

H₃₁＝H₃₂×H₂₁

H_n1＝H_nn-1×H_n-1n-2×…×H₂₁

wherein H_ii-1The transformation matrix between the ith-1 bridge deck image and the ith bridge deck image in the same row is obtained by calculation in the step 2-2; h_i1A transformation matrix between the 1 st bridge deck image and the ith bridge deck image in the same row; n is the number of images on the same line.

(2) As shown in fig. 6Showing that the first horizontal panoramic Image obtained in the step (1)₁And as a reference panoramic image, splicing according to an image column splicing strategy. For transformation matrix T between horizontal panoramic images_jj-1Performing transfer transformation to obtain horizontal panoramic images_iTransfer transformation matrix T with reference panoramic image_j1(ii) a Transforming the matrix T by each transfer_j1And (3) respectively mapping the corresponding transverse panoramic images into a reference plane coordinate system to finish image splicing and fusion between every two adjacent transverse panoramic images in the vertical direction, and referring to the description in the step (1) by a transfer matrix transformation formula in the image splicing and fusion to form a final bridge deck panoramic image.

The fade-in fade-out fusion algorithm in this embodiment is improved, as described in detail below.

After image registration, in order to further eliminate the interference of the bridge deck image splicing seam on the crack detection processing, the pixel values of the adjacent bridge deck images are generally weighted and averaged, as shown in fig. 7b, and the distance from the pixel point in the overlapping area to the suture lines on the two sides is used as the fusion weight judgment basis.

However, due to the change of the image acquisition position, the reflection light on the surface of the bridge may cause the phenomenon of jump of the gray value of the individual pixel point in the overlapping area, and in order to eliminate the influence of the jump on the fused image, a threshold value t is introduced into the traditional gradual-in and gradual-out weighted fusion calculation. Calculating the gray difference value of the target pixel point of the overlapped part in the two original images, if the difference value is smaller than a threshold value, the pixel point does not show obvious difference in the original bridge image, and directly taking the weighted average value as the pixel value of the point; and otherwise, the image to be spliced has a light and shade mutation at the pixel position, and the pixel value with larger weight before smoothing is taken as the fused pixel value of the point.

In the fade-in and fade-out fusion algorithm, the fade-in and fade-out weighting formula of each fused pixel value I (x, y) in the overlapping region of adjacent images is as follows:

wherein, I₁(x,y)、I₂(x, y) are the pixel values of the corresponding fusion points of the two adjacent bridge deck images in the overlapping region, as shown in FIG. 7 a. d₁、d₂Respectively is a gradual change weight factor of two adjacent bridge deck images at the corresponding fusion point; as shown in figure 7b of the drawings,

and

Step3, crack detection

And carrying out gray processing on the bridge deck splicing image. And dividing the bridge deck splicing image into a plurality of local areas through uniform grid division. As the bridge target crack gray value is lower than the local background gray average value, counting the gray integrated value in each local area to obtain the gray distribution of all grid areas and extracting the bridge deck image average gray value as the gray threshold value; the grid area with the accumulated gray value lower than the threshold gray value is screened out as the region of interest, and the initial positioning marking of the crack target area is performed, as shown in fig. 8.

And 3-2, performing improved median filtering denoising and image enhancement processing based on a fuzzy set on each region of interest to obtain a plurality of crack gray level images to be segmented.

Although the crack gray scale image lacks rich color contrast, it has unique spatial ductility and gray scale distinctiveness. By improving the median filtering, according to the linear characteristics in the crack local space and the uniformity of the gray level of the edge pixel, a similar gray level expansion direction is introduced on the basis of the original median filtering, as shown in fig. 9a, adjacent pixel points in the direction are brought into the filtering template, and the filtering interference of adjacent noise to the central pixel under the original filtering template is reduced to a certain extent.

The median filtering and denoising steps are as follows:

And

(3) will be provided with

And

as the adjacent gray scale extension direction of the target pixel point, as shown in fig. 9b, the target pixel point is respectively extended to the eight neighborhoods by one level in the two directions to obtain f_pAnd f_qTwo pixel points;

(4) for target pixel point f_ijAll pixel points and pixel points f in eight neighborhoods_pAnd pixel point f_qThe gray value of the formed array is sorted, and the median of the array is taken as a target pixel point f_ijThe replacement value of (a).

3-3, performing binary segmentation on each crack image to be segmented through a PCNN (pulse coupled neural network) simplified model;

in order to improve the fracture image segmentation efficiency, the PCNN simplified model is used on the premise of ensuring the detection precision, and as shown in FIG. 10, partial parameter influence is removed. According to the spatial position information of target cracks and neighborhood pixels under the complex bridge floor background, the optimal initial threshold value in the gray level image is automatically solved through a two-dimensional Otsu algorithm, so that the correct nerve pulse is generated when the model is iterated for the first time, and the optimization efficiency of the optimal iterative segmentation effect is improved; combining the neuron coupling characteristic and the image space gray distribution characteristic, obtaining a link intensity coefficient between neurons by solving the mean square error of local gray of the bridge fracture image, and taking the link intensity coefficient as more accurate coupling intensity between the neurons in a local area so as to capture neuron synchronous pulses in a local gray range.

A specific method for performing binary segmentation on each to-be-segmented fracture image through the PCNN simplified model is shown in fig. 11, and is specifically described as follows:

step 1: initializing PCNN model parameters

(1) Inputting grey values of all normalized pixel points (I, j) of a crack image to be segmented as an external stimulation signal I_ijI.e. F_ij(n)＝I_ij。

Step 2: value k is assigned to 1.

Step 3: performing segmentation iteration on a crack image to be segmented by adopting a PCNN simplified model to obtain a binary segmentation image Y (k), wherein the binary segmentation images obtained by each segmentation are different; computing internal activation U of each neuron in PCNN simplified model_ijAnd pulse output Y_ijAnd by comparing U_ijAnd Y_ijThe size of the neuron determines the activation state of each neuron.

Step 5: if k is less than n, increasing k by 1Repeatedly executing the steps Step3 and Step 4; otherwise, taking the maximum information entropy H_maxThe corresponding iteration times are used as the optimal iteration times k_HAnd outputting the current preferred segmentation image Y(s) as an optimal fracture segmentation image.

And 3-4, removing residual isolated noise from the optimal crack segmentation image to obtain a crack binary image.

Because the bridge surface image is accompanied by interference information of different degrees, such as self surface stain, interference points caused by uneven illumination during shooting and the like, a small amount of isolated noise still exists in the bridge surface binary image after the crack edge segmentation processing, and the bridge surface binary image is divided into two types of discrete noise points and concentrated noise points according to the noise distribution characteristic.

Discrete noise points in the bridge deck image are sparsely distributed, and the number of pixels is small, so that the bridge deck image is often characterized by small noise spots or short lines. Therefore, according to the morphological characteristics of the concrete bridge cracks, the binary edges are subjected to AND operation by using a long line segment template to judge the length characteristics of the edge lines of the connected regions, the areas of the connected regions surrounded by the binary edges are respectively calculated by using an area threshold value, and the area characteristics of the isolated regions are judged. The specific algorithm steps are as follows:

(1) and marking the connected regions divided by each edge in the optimal crack segmentation image obtained in the step 3-3.

(2) Respectively calculating the pixel area of each connected region in the optimal crack segmentation image; taking the connected region with the pixel area smaller than the area threshold as an isolated noise region, and removing; the area threshold is 20.

(3) And (3) searching by adopting linear segment scanning: setting the length of the line segment template to be 5 pixels, scanning and detecting the boundary line of the communication area reserved in the step (2) from four directions of 0 degrees, 45 degrees, 90 degrees and 135 degrees respectively, and carrying out 'AND' operation. If the straight line segment scanning retrieval return values of the boundary line of one connected region in the directions of 0 degrees, 45 degrees, 90 degrees and 135 degrees are all 0 (namely, no straight line with 5 continuous pixels exists in four directions), the connected region is regarded as an isolated noise region and removed.

The aggregated noise spots in the bridge deck image are usually formed by surface dirt, pits, shadow shielding and the like, are distributed in a compact and concentrated manner, have large areas and are blocky, and are divided into regular noise spots and irregular noise spots according to the shape characteristics of the noise spots. The method for removing the aggregated noise spots comprises the following specific steps:

(1) according to the characteristic that the circularity of the crack is small, the circularity is used as a filtering standard of regular noise spots.

The circularity e (x, y) of each connected region is calculated as follows:

If the circularity e (x, y) of a connected component is higher than the circularity threshold T_e(x, y), then the connected region is regarded as a regular noise spot to be removed, T_e(x,y)＝0.3。

(2) According to the shape characteristics of the slender and bent cracks, a certain length-width ratio is adopted as the filtering standard of the noise spot area.

The aspect ratio r (x, y) of each connected region is calculated as follows:

T is set in consideration of extremely high ratio of length to width of transverse cracks and extremely low ratio of longitudinal cracks_rl(x, y) is the lower limit of the length-width ratio, and the value is 0.1; t is_rh(x, y) is the upper limit of the aspect ratio, and is 10. If the aspect ratio r (x, y) of a connected region satisfies the following inequality: t is_rl(x,y)＜r(x,y)＜T_rh(x, y), the connected region is regarded as a small-area irregular noise spot region, and then the connected region is removed.

Step4, data extraction module

And 4-1, performing crack characteristic connection based on seed point growth on the crack binary image obtained in the step3, performing morphological refinement and deburring treatment on the complete crack image, and reserving the most critical crack main body characteristics.

According to the consistency of the edge tangential directions of the front and rear fracture ports, the invention provides a crack connection algorithm based on seed point growth, and the starting seed points and the guide seed points at the adjacent crack ports are arranged to perform connection fitting on the starting seed points and the guide seed points. The initial seed point is the first seed point at which the growth starts and represents the initial position of the region growth; the guide seed point is a seed point guiding the initial seed point to grow contiguously, and represents a termination region of the fitted extension. However, since the fitted growth from the initial seed point to the guide seed point depends on the gray distribution in the extension direction, a growth deviation tends to occur. Therefore, a threshold needs to be set, if growth deviation occurs, only the fitting point of the seed point is within the threshold distance range taking the guide seed point as the center of a circle, and the growth connection is also considered to be successful, otherwise, the seed point needs to be selected again for fitting.

The crack connection method is as follows:

(1) the binary image after edge segmentation does not contain crack neighborhood gray scale information any more, so that the requirement of region growth cannot be met. And (3) acquiring the fracture break position in the fracture binary image obtained in the step (3) by manually selecting or utilizing a branchpoints function in MATBAL software. And mapping the obtained position information of the fracture part to the crack gray level image obtained in the step 3-2, and growing the seed points.

(2) And respectively selecting an initial seed point and a guide seed point at the opposite ends of two sections of cracks corresponding to the fracture breakage, so as to determine the main joining growth direction of the crack, and ensure that the extension direction is more true.

(3) In order to more accurately fit connecting lines that are miscut due to low contrast, seed point growth rules need to be formulated. The pixel values are low due to cracks in the grayscale image. Taking the initial seed point as a first growth point for growth; and the next growth point is a pixel point with the minimum gray value in the eight neighborhoods of the previous growth point until the grown growth point is coincided with the guide seed point.

(4) And (4) mapping each growth point obtained in the step (3) to a crack binary image from the crack gray image to complete the complete connection of the crack fracture.

In this embodiment, a crack seed point growth legend is used as an example for demonstration, as shown in the left side of fig. 12a, a grid matrix represents image pixel points, the numerical values of the image pixel points represent gray values, gray squares represent pixel points at two disconnected crack ports, and p and q points with black circles represent selected starting seed points and guiding seed points, respectively. The eight neighboring directions of the former growing point are shown in fig. 12b, and if the crack connecting line direction is determined to be the second extending direction, the second extending direction is defined as the main direction, and according to the counterclockwise rule, the first and third extending directions are respectively set as the secondary first and secondary second directions, and the priority of the neighboring growing directions is sequentially reduced. Thereby continuously determining each growth point p in the growth direction_iThe coordinate points are represented by black thick line frames and mapped into the crack binary image, as shown in the right side of fig. 12, that is, the coordinate points can be used as the crack connecting lines.

4-2, classifying and identifying the target cracks based on a projection method, and automatically judging the types of the target cracks according to the morphological characteristics of the bridge deck cracks;

firstly, carrying out inversion operation on the fracture binary image processed in the step 4-1. And then according to the characteristics that the pixel of the crack edge line in the crack binary image is 1 and the pixel of the bridge floor background is 0, respectively projecting the crack binary image in the horizontal direction and the vertical direction by adopting a projection method, respectively accumulating and summing the pixel values of each row and column, and storing the sum value into an array to obtain a row projection array and a column projection array. And judging the crack types according to the characteristics of the two groups of obtained projection arrays, and dividing the concrete bridge cracks into transverse cracks, longitudinal cracks, oblique cracks and reticular cracks. In the common bridge deck crack category, the apparent characteristics of different types of cracks are obvious, transverse cracks, longitudinal cracks and oblique cracks are represented as good linear characteristics, the extending and developing directions are clear, and reticular cracks are represented as poor directionality. And taking different morphological characteristics of various cracks in the binary image as the basis for classifying and identifying the target cracks.

Due to the continuity of the edge lines of the cracks, the arrays projected in the horizontal direction and the vertical direction also have continuity, so that x (i) and y (j) are the ith numerical value and the jth numerical value of the row projection array and the column projection array of the crack binary image respectively. The expression is as follows:

wherein f (i, j) is a crack binary value diagram and is a pixel value of the ith row and the jth column; m, N are the length and width dimensions of the crack binary image, respectively.

The method for specifically judging the crack type of the crack binary image comprises the following steps:

If it is

Judging the cracks in the crack binary image to be transverse cracks; if it is

And all elements in the row projection array and the column projection array are larger than 10, which shows that the projections of the crack edge in two directions are overlapped at multiple positions, and the crack in the crack binary image is judged to be a net-shaped crack; otherwise, judging a crack binary imageThe middle cracks are oblique cracks;

The bridge crack detection is finally used for obtaining the disease characteristic information of the target crack and is used as an important basis for observation, control, overhaul and treatment. Therefore, after the target fracture region is positioned and the morphological feature is identified and classified, quantitative data calculation is also needed to be carried out on fracture morphological feature parameters. The morphological characteristic parameters of the bridge crack mainly comprise an area, a length, a width and a maximum width, wherein crack width information is a key index for evaluating the damage degree of the bridge crack.

And a fracture skeleton line of the fracture image on the fracture binary image after fracture edge segmentation and burr branch removal is a fracture center line consisting of most representative single pixels. Because the measurement precision requirements of bridge detection on the length and the area of the crack are not high, in order to improve the calculation efficiency, the number of pixel points on the skeleton line of the crack is directly counted to be used as the length of the crack pixel; and counting the number of pixel points with the numerical value of 1 in the crack binary image as the crack pixel area.

Due to the crack framework after edge thinning and burr removal, the real trend and development condition of the bridge deck crack can be reflected. The vertical direction of each position point is obtained by retrieving the crack skeleton pixel points, and then the vertical direction is mapped into the crack edge binary image, and the bridge crack width information is calculated, as shown in fig. 13, the specific solving process is as follows:

(1) as shown in fig. 14, a 5 × 5 search template is set, all fracture pixel points on the fracture skeleton line are swept, two fracture pixel points (i.e., pixel points with a value of 1) with the farthest distance in the search template are taken, and a connecting line of the two fracture pixel points is used as a local fracture strike line in the template. And then the normal line of the local trend line of each crack obtained after sweeping is carried out, and the normal angle is calculated.

(2) Solving each crack image according to the normal angle of each crack pixel point on the crack skeleton lineTwo side edge points (x) corresponding to the plain points on the crack_ij,y_ij) And (x'_ij,y′_ij) (ii) a And the edge points at the two sides of the crack pixel point corresponding to the crack are the intersection points of the normal line of the crack pixel point and the edge lines at the two sides of the crack. The Euclidean distance of the corresponding edge points at two sides of the crack pixel point on the crack is taken as the crack pixel width d on the position point_ijI.e. by

And obtaining each actual parameter A of the crack. Wherein u is the object distance, f is the focal length, k is the conversion coefficient, L is the long-side physical dimension of the CCD photosensitive chip, and L is the number of pixels on the long side of the shot image. The pixel parameters of the crack include crack pixel length, crack maximum pixel width, crack average pixel width, and crack pixel area. And dividing the sum of the crack pixel areas in all the crack binary images by the number of pixels of the bridge deck panoramic image to obtain the plane crack rate.

The derivation process of the actual parameter conversion formula is as follows:

the image distance can be obtained by a lens imaging formula:

the optical magnification is then:

the imaged dimensions of the target fracture were: a' ═ MA

U is the object distance, v is the image distance, f is the focal length, A is the actual size of the crack target, A 'is the target pixel size, however, since A is in mm unit, A' is in pixel unit, unit conversion is needed, k value is the conversion coefficient, L is the long edge physical size of the CCD photosensitive chip, L is the number of long edge pixel points of the shot image.

Therefore, the actual physical information can be obtained by extracting the data information of the image crack:

therefore, the detection and calculation of the position information and the size information of the bridge cracks on the real detection surface are realized.

The conversion of detection data can be carried out through the size of a photosensitive chip of a shooting camera, the pixel resolution and the bridge deck shooting distance, and the equipment can be adjusted in advance according to the precision requirement of a bridge detection project.

Claims

1. A bridge crack detection method is characterized by comprising the following steps: step1, acquiring images of a detected bridge deck one by one to obtain a bridge deck image set;

step2, image splicing

2-1. mosaic image preprocessing

Extracting and counting the brightness component information of each bridge deck image in the bridge deck image set, and respectively equalizing the brightness component information of each bridge deck image; then transforming the bridge deck images into a frequency domain through Fourier transform, obtaining translation parameters among the images by adopting phase information of a normalized cross-power spectrum in a phase correlation algorithm, and completing pre-estimation of an overlapping area between adjacent images;

2-2 image registration

Firstly, extracting SIFT feature points in an overlapping area between every two adjacent images; then sifting SIFT feature points by a self-adaptive contrast threshold method to obtain a feature descriptor consisting of matching point pairs; calculating a projection transformation matrix between every two adjacent images by adopting a random sampling consistency algorithm;

2-3. image fusion

Firstly, performing projection transformation on corresponding bridge deck images according to projection transformation matrixes between adjacent images; then, respectively carrying out weighted smooth transition on RGB three-color channels of each adjacent bridge deck image by adopting a gradual-in and gradual-out fusion algorithm to obtain a bridge deck splicing image;

step3, crack detection

3-1, carrying out initial positioning and graying of a crack area through crack detection pretreatment;

carrying out gray processing on the bridge deck splicing image; dividing the bridge deck splicing image into a plurality of local areas through uniform grid division; counting the gray accumulated value in each local area to obtain the gray distribution of all grid areas and extracting the average gray value of the bridge deck image as a gray threshold; screening out a grid area with the gray accumulated value lower than a gray threshold value as an interested area;

3-2, performing median filtering denoising and image enhancement processing based on a fuzzy set on each region of interest to obtain a plurality of crack gray level images to be segmented;

3-3, performing binary segmentation on each crack image to be segmented through a pulse coupling neural network simplified model to obtain a crack segmentation image;

3-4, removing residual isolated noise from the crack segmentation image to obtain a crack binary image;

step4, extracting crack parameters

4-1, performing crack characteristic connection based on seed point growth on the crack binary image obtained in the step 3;

firstly, carrying out negation operation on the fracture binary image processed in the step 4-1; respectively projecting the crack binary image in the horizontal direction and the vertical direction by adopting a projection method, and respectively accumulating and summing pixel values of each row and column to obtain a row projection array and a column projection array;

Δ x is the horizontal projection length of the crack part in the crack binary image, and Δ y is the vertical projection length of the crack part in the crack binary image;

if it is

Judging the cracks in the crack binary image to be transverse cracks; if it is

4-3, extracting bridge deck crack characteristic data to obtain the length, width and area information of the actual bridge crack;

(1) extracting a fracture skeleton line on the fracture in the fracture binary image;

counting the number of pixel points with the numerical value of 1 in the crack binary image as the crack pixel area; extracting a crack skeleton line according to a crack image on the crack binary image; taking the number of pixel points on the crack skeleton line as the length of the crack pixel;

(1) setting a 5 multiplied by 5 retrieval template, sweeping all crack pixel points on a crack skeleton line, taking two crack pixel points with the farthest distance in the retrieval template, and taking a connecting line of the two crack pixel points as a local trend line of a crack in the template; then, the normal line of the local trend line of each crack obtained after sweeping is carried out, and the normal angle is calculated;

(2) according to the normal angle of each crack pixel point on the crack skeleton line, the corresponding two-side edge points (x) of each crack pixel point on the crack are obtained_ij,y_ij) And (x'_ij,y′_ij) (ii) a The Euclidean distance of the corresponding edge points at two sides of the crack pixel point on the crack is used as the crack pixel width d on the crack pixel point_ijI.e. by

The obtained width d of each slit pixel_ijThe maximum value of (3) is taken as the maximum pixel width of the crack; width d of each slit_ijThe average value in (a) is taken as the average pixel width of the crack;

Obtaining each actual parameter A of the crack; wherein u is the object distance, f is the focal length, k is the conversion coefficient, L is the physical size of the long side of the CCD photosensitive chip, and L is the number of pixels of the long side of the shot image; the pixel parameters of the crack include crack pixel length, crack maximum pixel width, crack average pixel width, and crack pixel area.

2. The bridge crack detection method according to claim 1, characterized in that: the method for acquiring the image in the step1 specifically comprises the following steps:

1-1, calculating an internal reference matrix of the CCD camera by adopting a Zhangyingyou plane calibration method, and then obtaining a radial distortion coefficient by adopting a least square method;

1-2, arranging the CCD camera calibrated in the step 1-1 on a bridge detection platform, and acquiring an image of the complete bridge floor according to a preset shooting track; the preset shooting track is S-shaped;

3. The bridge crack detection method according to claim 1, characterized in that: in step 2-2, the adaptive contrast threshold method is specifically as follows:

(1) setting a lower limit N of the number of characteristic points_min200, upper limit N_max300, contrast threshold T_c＝T₀；T₀The initial threshold value is 0.02-0.04;

(2) detecting characteristic points and counting pairsRatio higher than T_cThe number of feature points N;

(3) if N is present_min≤N≤N_maxThen contrast will be higher than T_cThe characteristic points are brought into the initial matching point set, and the rejection contrast is lower than a threshold value T_cAnd directly entering the step (5); otherwise, executing the step (4);

And executing the step (3); if N > N_maxIncreasing the contrast threshold to 2 times of the original value and executing the step (3);

(5) rejecting error feature points in the initial matching point set by a nearest neighbor comparison neighbor method, and generating a feature descriptor; the feature descriptor includes a plurality of matching point pairs each composed of a pair of feature points, and distance and direction information between the matching point pairs.

4. The bridge crack detection method according to claim 1, characterized in that: in step 2-2, the process of solving the projective transformation matrix by the random sampling consistency algorithm is as follows:

(1) constructing an initial sample set S by using each matching point pair in the feature descriptor; counting Euclidean distances between each matching point pair in the initial sample set S, and sequencing from small to large;

(3) randomly extracting 4 groups of matching point pairs from the new sample set S' to form an inner point set S_iAnd calculating an interior point set S_iMatrix model H of_iEntering the step (4);

(4) the rest matching point pairs in the new sample set S' are corresponding to the matrix model H_iCarrying out adaptability test; if the matched points with the detection errors smaller than the error threshold exist, adding the matched point pairs with the detection errors smaller than the threshold into the inner point set S_iAnd executing the step (5); otherwise, the matrix model H is discarded_iRe-executing (3);

(5) if the inner point set S_iIf the number of the middle elements is larger than the specified threshold value, a reasonable parameter model is considered to be obtained, and the updated interior point set S is subjected to_iRecalculating matrix model H_iMinimizing a cost function by using a steepest descent algorithm; otherwise, the matrix model H is discarded_iAnd re-executing the step (3);

(6) repeating steps (3) to (5) for l times, wherein l is the maximum iteration number; then, comparing the inner point set S obtained in the iteration of the time I_iSet S of interior points with the largest number of elements_iTaking the matrix model H as the final internal point set and calculating_iAs a projective transformation matrix between adjacent bridge deck images.

5. The bridge crack detection method according to claim 1, characterized in that: in step 2-3, the projection transformation comprises the following specific steps:

(1) according to the transmissibility of the projective transformation matrix between the adjacent images, the first bridge deck image of each row is respectively used as a reference image of the corresponding row for splicing; for the transformation matrix H between the adjacent bridge floor images_ii-1Carrying out transmission transformation to obtain a transmission transformation matrix H between each bridge deck image and the reference image_i1(ii) a Then through each transformation matrix H_i1Mapping the corresponding bridge deck images into a reference plane coordinate system respectively to finish Image splicing and fusion between every two adjacent images in the horizontal direction to form a plurality of transverse panoramic Image images with wide visual angles_i；

(2) The first horizontal panoramic Image obtained in the step (1) is processed₁Splicing as a reference panoramic image; for transformation matrix T between horizontal panoramic images_jj-1Performing transfer transformation to obtain horizontal panoramic images_iTransfer transformation matrix T with reference panoramic image_j1(ii) a Transforming the matrix T by each transfer_j1Respectively mapping the corresponding transverse panoramic images into a reference plane coordinate system to complete image splicing and fusion between every two adjacent transverse panoramic images in the vertical direction to form a final bridge deck panoramic image;

wherein, I₁(x,y)、I₂(x, y) are pixel values of corresponding fusion points of two adjacent bridge deck images in the overlapping area respectively; d₁、d₂Respectively is a gradual change weight factor of two adjacent bridge deck images at the corresponding fusion point;

and

6. The bridge crack detection method according to claim 1, characterized in that: the median filtering and denoising steps in the step 3-2 are as follows:

(1) traversing each pixel point in the interested region to be respectively used as a target pixel point f_ijAnd performing steps (2) to (4) respectively;

And

(3) will be provided with

And

7. The bridge crack detection method according to claim 1, characterized in that: in the step 3-3, a specific method for performing binary segmentation on each crack image to be segmented through the pulse coupling neural network simplified model is as follows:

step 1: initializing pulse coupled neural network model parameters

(1) Inputting grey values of all normalized pixel points (I, j) of a crack image to be segmented as an external stimulation signal I_ij；

(2) Setting decay time constant delta t as 0.2 and maximum information entropy value H_max0, 30 maximum iteration number n and link strength coefficient matrix

(3) The optimal threshold value obtained by the two-dimensional Otsu method is used as the initial threshold value theta of the pulse coupling neural network model_ij[0](ii) a Determining a link strength factor beta by a local gray variance method_ij；

Step 2: assigning 1 to k;

step 3: segmenting and iterating a crack image to be segmented by adopting a pulse coupled neural network simplified model to obtain a binary segmentation image Y (k); computing internal activation U of each neuron in pulse coupled neural network simplified model_ijAnd pulse output Y_ijAnd by comparing U_ijAnd Y_ijJudging the activation state of each neuron according to the size of the neuron;

step 4: calculating information entropy value H (k) corresponding to the binary segmentation image Y (k), if H (k) is more than H_maxThen, assign H (k) to H_maxAnd taking the binary segmentation image Y (k) as a new preferred segmentation image Y(s); otherwise, directly enter step 5;

8. The bridge crack detection method according to claim 1, characterized in that: in step 3-4, the residual isolated noise comprises discrete noise points and concentrated noise spots;

(1) marking the connected regions divided by each edge in the crack segmentation image obtained in the step 3-3;

(2) respectively calculating the pixel area of each connected region in the optimal crack segmentation image; rejecting connected regions with pixel areas smaller than an area threshold; the area threshold is 20;

(3) and (3) searching by adopting linear segment scanning: setting the length of a line segment template to be 5 pixels, respectively scanning and detecting the boundary line of the communication area reserved in the step (2) from four directions of 0 degree, 45 degrees, 90 degrees and 135 degrees, and carrying out 'AND' operation; if the straight line segment scanning retrieval return values of the boundary line of one connected region in the directions of 0 degree, 45 degrees, 90 degrees and 135 degrees are all 0, rejecting the connected region;

(1) the circularity e (x, y) of each connected region is calculated as follows:

in the formula, C (x, y) and A (x, y) are the perimeter and the area of the corresponding communication area respectively;

if the circularity e (x, y) of a connected component is higher than the circularity threshold T_e(x, y), then the connected region is eliminated, T_e(x,y)＝0.3；

in the formula, L (x, y) and W (x, y) are the length and width of a projection matrix of the corresponding communication area in the horizontal direction and the vertical direction respectively;

setting T_rl(x, y) is the lower limit of the length-width ratio, and the value is 0.1; t is_rh(x, y) is the upper limit of the length-width ratio, and the value is 10; if the aspect ratio r (x, y) of a connected region satisfies the following inequality: t is_rl(x,y)＜r(x,y)＜T_rh(x, y), then the connected region is culled.

9. The bridge crack detection method according to claim 1, characterized in that: the fracture joining method described in step 4-1 is as follows:

(1) mapping the position information of the fracture break in the fracture binary image to the fracture gray level image obtained in the step 3-2;

(2) respectively selecting an initial seed point and a guide seed point at the opposite ends of two sections of cracks corresponding to the fracture;

(3) the initial seed point is a first growing point for growing; the next growth point is a pixel point with the minimum gray value in the eight neighborhoods of the previous growth point until the grown growth point is superposed with the guide seed point;

10. The bridge crack detection method according to claim 1, characterized in that: and 4-3, after the step is executed, dividing the sum of the crack pixel areas in all the crack binary images by the number of pixels of the bridge deck panoramic image to obtain the surface crack rate.