CN110363706A - A kind of large area bridge floor image split-joint method - Google Patents
A kind of large area bridge floor image split-joint method Download PDFInfo
- Publication number
- CN110363706A CN110363706A CN201910561334.9A CN201910561334A CN110363706A CN 110363706 A CN110363706 A CN 110363706A CN 201910561334 A CN201910561334 A CN 201910561334A CN 110363706 A CN110363706 A CN 110363706A
- Authority
- CN
- China
- Prior art keywords
- image
- bridge floor
- adjacent
- point
- matrix
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 32
- 238000001514 detection method Methods 0.000 claims abstract description 26
- 239000011159 matrix material Substances 0.000 claims description 63
- 230000009466 transformation Effects 0.000 claims description 55
- 230000004927 fusion Effects 0.000 claims description 13
- 230000003044 adaptive effect Effects 0.000 claims description 10
- 238000002156 mixing Methods 0.000 claims description 8
- 241000023320 Luma <angiosperm> Species 0.000 claims description 6
- 230000008859 change Effects 0.000 claims description 6
- OSWPMRLSEDHDFF-UHFFFAOYSA-N methyl salicylate Chemical compound COC(=O)C1=CC=CC=C1O OSWPMRLSEDHDFF-UHFFFAOYSA-N 0.000 claims description 6
- 238000012549 training Methods 0.000 claims description 6
- 238000012360 testing method Methods 0.000 claims description 5
- 238000013519 translation Methods 0.000 claims description 5
- 238000010606 normalization Methods 0.000 claims description 4
- 230000008569 process Effects 0.000 claims description 4
- 238000001228 spectrum Methods 0.000 claims description 4
- HUTDUHSNJYTCAR-UHFFFAOYSA-N ancymidol Chemical compound C1=CC(OC)=CC=C1C(O)(C=1C=NC=NC=1)C1CC1 HUTDUHSNJYTCAR-UHFFFAOYSA-N 0.000 claims description 3
- 238000007430 reference method Methods 0.000 claims description 3
- 230000007704 transition Effects 0.000 claims description 3
- 230000000007 visual effect Effects 0.000 claims 1
- 230000006378 damage Effects 0.000 abstract description 4
- 238000005070 sampling Methods 0.000 abstract description 4
- 230000007547 defect Effects 0.000 abstract description 3
- 238000012545 processing Methods 0.000 abstract description 3
- 238000011160 research Methods 0.000 abstract description 3
- 238000011161 development Methods 0.000 abstract description 2
- 238000005516 engineering process Methods 0.000 abstract description 2
- 238000007781 pre-processing Methods 0.000 abstract description 2
- 238000009659 non-destructive testing Methods 0.000 abstract 1
- 201000010099 disease Diseases 0.000 description 8
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 8
- 238000010586 diagram Methods 0.000 description 5
- 235000013350 formula milk Nutrition 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 238000012216 screening Methods 0.000 description 3
- 238000010276 construction Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005286 illumination Methods 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 238000012546 transfer Methods 0.000 description 2
- BVKZGUZCCUSVTD-UHFFFAOYSA-L Carbonate Chemical compound [O-]C([O-])=O BVKZGUZCCUSVTD-UHFFFAOYSA-L 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000003763 carbonization Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000007797 corrosion Effects 0.000 description 1
- 238000005260 corrosion Methods 0.000 description 1
- 238000005336 cracking Methods 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000007598 dipping method Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 231100001261 hazardous Toxicity 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 239000012528 membrane Substances 0.000 description 1
- 230000035772 mutation Effects 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000000746 purification Methods 0.000 description 1
- 238000004445 quantitative analysis Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000003014 reinforcing effect Effects 0.000 description 1
- 230000000246 remedial effect Effects 0.000 description 1
- 230000006641 stabilisation Effects 0.000 description 1
- 238000011105 stabilization Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4038—Image mosaicing, e.g. composing plane images from plane sub-images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
- G06T7/33—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/80—Analysis of captured images to determine intrinsic or extrinsic camera parameters, i.e. camera calibration
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20212—Image combination
- G06T2207/20221—Image fusion; Image merging
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of large area bridge floor image split-joint methods.With the development of Computer Vision Detection Technique, gradually appears and image detection is applied in bridge machinery engineering practice.But since pontic space is larger, according to long-range shooting sampling, it will receive the limitation of resolution of video camera, cause to be unable to get satisfied detection accuracy.The present invention is as follows: 1, acquiring the image for being detected bridge floor one by one, obtain bridge floor image collection.Later, image preprocessing is carried out.2, image registration.3, image co-registration.The present invention completes the automation non-destructive testing of bridge defect feature instead of human eye by Image Acquisition and processing technique, has very important realistic meaning to the research of the bridge floor damage detection technology under complicated landform environment.On the one hand working security is enhanced, on the other hand improves operation mobility and flexibility.Invention realizes the fidelity splicing of large area bridge floor image, improves image mosaic precision and efficiency.
Description
Technical field
The invention belongs to technical field of image detection, and in particular to a kind of large area bridge floor image split-joint method.
Background technique
Bridge exposes to the sun and rain by long-term and loads operation, internal stress can be along bridge as important transport hub
Girder construction is transmitted to some weak parts, and the positional structure surface is caused the Disease Characters such as crack easily occur.Due to bridge surface
Disease Characters cause outside air and hazardous medium to be easy to penetrate into inside concrete after chemical reaction to generate carbonate,
Causing the basicity environment of wherein reinforcing bar reduces, and the purification membrane on surface is more also easy to produce corrosion after wrecking, in addition, concrete carbonization
Also it can aggravate shrinkage cracking, serious harm is generated to the safe handling of concrete-bridge.Therefore, in order to ensure that the use of girder construction
Service life and security performance need that bridge surface Disease Characters are detected and administered in time.
It in order to detect deck disease feature in time, and adopts remedial measures to eliminate safe hidden trouble, generallys use artificial
The mode of inspection and manual markings.However bridge bottom surface detection working environment is often more dangerous, this detection mode mobility
Difference, risk is big, low efficiency.And bridge floor damage by veteran reviewer's hand dipping and detects by an unaided eye to do and remembers
Record, have certain subjectivity, detection accuracy compared with depend on expert Heuristics, and experience lack in quantitative analysis it is objective
Property.
With the development of Computer Vision Detection Technique, gradually appears and image detection is applied to bridge machinery engineering practice
In.But since pontic space is larger, according to long-range shooting sampling, it will receive the limitation of resolution of video camera, cause not obtaining
To satisfied detection accuracy.Therefore need to carry out bridge surface the continuous multiple groups sampling of short distance, and to collected sample
Image carries out the image mosaic of large area, and splicing effect will have a direct impact on the detection accuracy of deck disease feature.In order to
On the basis of meeting wide viewing angle, high-resolution, reduction cumulative errors, raising bridge floor image mosaic precision are realized true as far as possible
The overall situation of bridge machinery face information shows.
Summary of the invention
The purpose of the present invention is to provide a kind of large area bridge floor image split-joint methods.
The specific steps of the present invention are as follows:
Step 1 acquires the image for being detected bridge floor one by one, obtains bridge floor image collection.Later, it extracts and counts bridge floor figure
The luma component information of each bridge floor image in image set conjunction, and the luma component information of each bridge floor image is equalized respectively.
Then each bridge floor image is transformed in frequency domain by Fourier transformation, and using the normalization cross-power in phase related algorithm
The phase information of spectrum obtains the translation parameters between image, completes the pre- estimation between overlapping region adjacent image.
Step 2, image registration
Firstly, extracting SIFT feature in overlapping region between each adjacent image.Then pass through adaptive contrast threshold
Method screens SIFT feature, obtains the feature descriptor being made of matching double points.And each neighbor map is calculated using RANSAC algorithm
Projective transformation matrix as between.
Adaptive contrast threshold method is specific as follows:
(1) characteristic point numerical lower limits N is setmin=200, upper limit Nmax=300, contrast threshold Tc=T0。T0For initial threshold
Value, value are 0.02~0.04.
(2) characteristic point detection is carried out, and Statistical Comparison degree is higher than TcCharacteristic point quantity N.
(3) if Nmin≤N≤Nmax, then contrast is higher than TcCharacteristic point be included in initial matching point set, reject contrast
Lower than threshold value TcCharacteristic point, and be directly entered step (5).Otherwise, step (4) are executed.
(4) if N < Nmin, then by contrast threshold TcIt is reduced to former numerical valueAnd execute step (3).If N > Nmax, then
Contrast threshold is increased to 2 times of former numerical value, and executes step (3).
(5) the mistake characteristic point that initial matching point is concentrated is rejected than secondary near neighbor method by arest neighbors, and generates feature description
Symbol.In feature descriptor comprising by pairs of feature point group at multiple matching double points and the distance between each matching double points
And directional information.
Step 3, image co-registration
First according to the projective transformation matrix between adjacent image, projective transformation is carried out to corresponding bridge floor image.Then it uses
Be fade-in gradually go out blending algorithm smooth transition is weighted respectively to tri- Color Channel of RGB of each adjacent bridge floor image, obtain bridge floor
Stitching image.
It is fade-in and gradually goes out in blending algorithm, being fade-in for each merging point pixel value I (x, y) gradually goes out to add in adjacent image overlapping region
It is as follows to weigh formula:
Wherein, I1(x,y)、I2(x, y) is respectively correspondence merging point of the two adjacent bridge floor images in overlapping region
Pixel value.d1、d2Gradual change weight factor of the two respectively adjacent bridge floor images in corresponding merging point.Withx1、x2The respectively abscissa on overlapping region two sides boundary.X is the abscissa of corresponding merging point.T is two adjacent
Gray difference threshold of the image overlapping region on corresponding merging point.
Preferably, the method for acquiring image in step 1 is specific as follows:
(1) after the internal reference matrix for calculating CCD camera using Zhang Zhengyou plane reference method, radial direction is obtained by least square method
Distortion factor.
(2) on bridge machinery platform placement by step 1-1 calibration CCD camera, according to preset shooting track into
The Image Acquisition of the complete bridge floor of row.Preset shooting track is S-shaped.
(3) according to the obtained CCD camera internal reference matrix of step 1-1 and distortion factor each bridge floor collected to step 1-2
Image carries out image calibration respectively.
Preferably, the process that RANSAC algorithm solves projective transformation matrix is as follows in step 2:
(1) original training set S is constructed with each matching double points in feature descriptor.It is respectively matched in statistics original training set S
Euclidean distance between point pair, and by sorting from small to large.
(2) the preceding 85% matching double points building new samples collection S ' of sequence obtained by step (1) is taken.
(3) 4 groups of matching double points are randomly selected from new samples collection S ' forms an interior point set Si, and calculating matrix model
Interior point set SiHi, enter step (4).
(4) remaining interior each matching double points of new samples collection S ' are directed to matrix model HiCarry out adaptive test.It examines if it exists
The match point that error is less than error threshold is tested, then interior point set S is added in the matching double points for examining error to be less than threshold valuei, and hold
Row step (5).Otherwise, give up matrix model Hi, re-execute (3).
(5) if interior point set SiMiddle element number is greater than defined threshold, then it is assumed that reasonable parameter model is obtained, to update
Interior point set S afterwardsiRecalculate matrix model Hi, and use LM algorithmic minimizing cost function.Otherwise, give up the matrix norm
Type Hi, and it re-execute the steps (3).
(6) l step (3) is repeated to (5), and l is maximum number of iterations.Later, interior point set obtained in l iteration is compared
Close Si, with the maximum interior point set S of element numberiAs final interior point set, and the matrix model H for taking it to calculateiAs adjacent
Projective transformation matrix between bridge floor image.
Preferably, specific step is as follows for projective transformation in step 3:
(1) according to the transitivity of the projective transformation matrix between adjacent image, made respectively with first bridge floor image of every row
Benchmark image for corresponding row is spliced.Between the transformation matrix H each adjacent bridge floor imageii-1Transmitting transformation is carried out, is obtained
Transmitting transformation matrix H between each bridge floor image and benchmark imagei1.Pass through each transformation matrix H againi1By corresponding bridge floor image
It is respectively mapped in datum plane coordinate system, to complete the image mosaic fusion in horizontal direction between each adjacent image, is formed more
Open the lateral panoramic picture Image of wide viewing anglei。
(2) by first obtained in step (1) lateral panoramic picture Image1Spliced as reference panorama image.
Between the transformation matrix T each lateral panoramic picturejj-1Transmitting transformation is carried out, each lateral panoramic picture Image is obtainediIt is complete with benchmark
Transmitting transformation matrix T between scape imagej1.Pass through each transmitting transformation matrix T againj1Respectively by corresponding lateral panoramic picture point
It is not mapped in datum plane coordinate system, to complete the image mosaic fusion on vertical direction between each adjacent transverse panoramic picture,
Form final bridge floor panoramic picture.
The invention has the advantages that:
1, the present invention completes the lossless inspection of automation of bridge defect feature instead of human eye by Image Acquisition and processing technique
It surveys, there is very important realistic meaning to the research of the bridge floor damage detection technology under complicated landform environment.On the one hand enhance
Working security, on the other hand improves operation mobility and flexibility.
2, the present invention is directed to traditional images registration Algorithm in the problem that operand is larger and precision is insufficient, in order to more complete
And bridge floor image Disease Characters data are accurately extracted, and a kind of adjacent bridge floor image registration algorithm of improved multiple groups is proposed, it is real
The fidelity splicing of existing large area bridge floor image, improves image registration accuracy and efficiency, examines for subsequent bridge defect characteristic image
Working foundation has been established in survey, also provides a Technical Reference for the image mosaic detection of other field.
3, the present invention detection environment special for bridge surface proposes that improved be fade-in gradually goes out Image Fusion, draws
The gray difference threshold for entering adjacent bridge floor image, can be effectively suppressed the influence of bridge floor image uncorrelated noise, farthest retains bridge
The minutia information of face disease improves the signal-to-noise ratio of stitching image on the basis of realizing the fusion of multiple groups bridge floor image fidelity.
4, the present invention improves the anti-interference and stability of bridge floor merging algorithm for images under complex background, has preferable
Robustness.Guarantee accuracy and precision that subsequent Disease Characters data are extracted.
5, the present invention carries out the design of multiple groups bridge floor merging algorithm for images according to bridge surface Image Acquisition working environment, and
Reliability Analysis Research is carried out to wherein crucial calculate, algorithm novelty with higher and bridge machinery engineering reference price
Value.
Detailed description of the invention
Fig. 1 is large area bridge floor image mosaic flow chart in the present invention;
Fig. 2 is bridge floor Image Acquisition track schematic diagram in the present invention;
Fig. 3 is several bridge image mosaic process flow diagrams in the present invention;
Fig. 4 is adaptive contrast threshold calculation flow chart in the present invention;
Fig. 5 is that progressive image splices tactful schematic diagram in the present invention;
Fig. 6 is image mosaic strategy schematic diagram by column in the present invention;
Fig. 7 a, 7b the Weighted Fusion schematic diagram between adjacent image in the present invention.
Specific embodiment
Below in conjunction with attached drawing, the invention will be further described.
As shown in Figure 1, a kind of large area bridge floor image split-joint method, the specific steps are as follows:
Step 1, splicing pretreatment
1-1. uses Zhang Zhengyou plane reference method, carries out image sampling to scaling board from different perspectives using CCD camera, leads to
After crossing the internal reference matrix of scaling board X-comers coordinate calculating CCD camera of detection, obtained by least square method radial abnormal
Variable coefficient.
1-2. disposed on bridge machinery platform by step 1-1 calibration CCD camera, according to preset shooting track into
The multiple series of images acquisition of the complete bridge floor of row.Preset shooting track is as shown in Fig. 2, S-shaped.
1-3. is according to the obtained CCD camera internal reference matrix of step 1-1 and distortion factor each bridge floor collected to step 1-2
Image carries out image calibration respectively, to eliminate lens distortion bring distortion effect.
1-4. image preprocessing
Estimate the size of panoramic picture in advance first.Resolution ratio sum number measurement value of the size according to image to be spliced, splicing
Inactive area is removed after the completion;Then by extracting and counting the luma component information of all bridge floor images to be spliced, and divide
The other luma component information to each bridge floor image equalizes, to eliminate the influence of uneven illumination bring difference in brightness;Finally
Each bridge floor image is transformed in frequency domain by Fourier transformation, and using the normalization crosspower spectrum in phase related algorithm
Phase information obtains the translation parameters between image, completes the pre- estimation between overlapping region adjacent image with this.
Phase related algorithm is first to be transformed to image to be spliced in frequency domain using Fourier transformation, then pass through normalization mutually
The translation parameters of two image of spectra calculation obtains two-dimensional impulse function: the peak value size reflection of the two-dimensional impulse function
Content relevance between adjacent bridge floor image, value are that 1 two images of expression are identical, indicate entirely different for 0.Bridge figure
Change as brought by being moved between the collected adjacent image of detection device institute there are perspective transform and position, although impulse can be made
The energy of function is dispersed to numerous small leaks from single peak value, but its corresponding translation parameters in peak-peak position can still keep phase
To stabilization.Therefore, the translational movement obtained by phase related algorithm can obtain roughly the overlapping region between image to be spliced, and
The algorithm is insensitive to illumination brightness change, and correlation maximum peak point detected has preferable robustness and stability.
Step 2, image registration
Firstly, SIFT feature is extracted in overlapping region between each adjacent image, to reduce a large amount of unnecessary characteristic points
Calculation amount is detected, the detection efficiency of SIFT feature is improved;Then it by adaptive contrast threshold method, will test
SIFT feature quantity controls in a reasonable range, to filter out stable feature point set;And using improved
RANSAC algorithm (RANSAC algorithm) calculates the projective transformation matrix H between each adjacent image.
Adaptive contrast threshold method is specific as follows:
Overlapping region between determination two-by-two adjacent bridge floor image (after Δ x, Δ y), is carried out only for the overlapping region
SIFT feature detection.Since wherein the lower characteristic point of contrast is more sensitive to bridge floor ambient noise, therefore set contrast
Threshold value filters out stable feature point set, is denoted as C.
In the prior art, the contrast of each SIFT feature is calculated by difference of Gaussian Taylor expansion, and fixation is set
Contrast threshold retain the characteristic point higher than the contrast threshold as invariant feature point.However above-mentioned contrast threshold Tc
For fixed value, general value is between 0.02 to 0.04.But in the crack image detection of different concrete-bridges, SIFT detection
To candidate feature point set have very big difference, Partial Bridges surface is more smooth bright and clean, and acquired image digital signal is more
Smoothly, scale space factor sigma is smaller, causes the characteristic point detected less, may be unable to satisfy the number of Feature Points Matching instead
Amount demand influences final splicing precision.(the contrast threshold step used in traditional stitching algorithm).The present invention is arranged one
The contrast threshold of variation, to guarantee the SIFT feature number detected control in a reasonable range.It is tested through multiple groups
Verifying, which shows that the characteristic point of Bridge Crack image detection is maintained between 200 to 300, can meet preferable splicing precision.
As shown in figure 4, determining contrast threshold T in the present inventioncMethod, it is specific as follows:
(1) characteristic point numerical lower limits N is setmin=200, upper limit Nmax=300, contrast threshold Tc=T0;T0For initial threshold
Value, value are 0.02~0.04.
(2) characteristic point detection is carried out, and Statistical Comparison degree is higher than TcCharacteristic point quantity N.
(3) if Nmin≤N≤Nmax, then contrast is higher than TcCharacteristic point be included in initial matching point set, reject contrast
Lower than threshold value TcCharacteristic point, and be directly entered step (5);Otherwise, step (4) are executed.
(4) if N < Nmin, then by contrast threshold TcIt is reduced to former numerical valueAnd execute step (3);If N > Nmax, then
Contrast threshold is increased to 2 times of former numerical value, and executes step (3).
(5) the mistake characteristic point that initial matching point is concentrated is rejected than secondary near neighbor method by arest neighbors, and generates feature description
Symbol.In feature descriptor comprising by pairs of feature point group at multiple matching double points and the distance between each matching double points and
Directional information.
After the Feature Points Matching between adjacent image overlapping region, enough matching double points are filtered out, matching double points are passed through
The transformation matrix between Bridge Crack sequence image is solved, large-scale bridge floor image mosaic is completed with this.To further increase figure
As registration efficiency and precision, RANSAC algorithm is improved.
The process that improved RANSAC algorithm solves projective transformation matrix H is as follows:
(1) original training set S is constructed with each matching double points in feature descriptor.It is respectively matched in statistics original training set S
Euclidean distance between point pair, and by sorting from small to large;
(2) the preceding 85% matching double points building new samples collection S ' of sequence obtained by step (1) is taken;
(3) 4 groups of matching double points are randomly selected from new samples collection S ' forms an interior point set Si, and calculating matrix model
Interior point set SiHi, enter step (4);
(4) remaining interior each matching double points of new samples collection S ' are directed to matrix model HiCarry out adaptive test;It examines if it exists
The match point that error is less than error threshold is tested, then interior point set S is added in the matching double points for examining error to be less than threshold valuei, and hold
Row step (5);Otherwise, give up matrix model Hi, re-execute (3).
(5) if interior point set SiMiddle element number is greater than defined threshold, then it is assumed that reasonable parameter model is obtained, to update
Interior point set S afterwardsiRecalculate matrix model Hi, and use LM algorithmic minimizing cost function;Otherwise, give up the matrix norm
Type Hi, and it re-execute the steps (3).
(6) l step (3) is repeated to (5), and l is maximum number of iterations.Later, interior point set obtained in l iteration is compared
Close Si, with the maximum interior point set S of element numberiAs final interior point set, and the matrix model H for taking it to calculateiAs adjacent
Projective transformation matrix H between bridge floor image.
Improved RANSAC algorithm is by the Euclidean distance between calculating all matching double points and is ranked up screening, not only subtracts
The sample set data for having lacked point pair to be matched improve intra-office point proportion in sample set, and reduce projective transformation square
The iteration of battle array refines number, to improve the matching precision of bridge floor image.According to image characteristic point to the distance between it is smaller,
With the higher characteristic of similarity, calculate herein the Euclidean distance between all characteristic points pair and according to sequence from small to large arrange into
Row screening.After showing overlapped provincial characteristics point to first matching by multiple groups bridge floor image mosaic test result statistics, initial sample
The successful match rate of this collection S then takes preceding 85% characteristic point of its sequence to building new samples collection S ' up to 85% or more.Through sample
Notebook data screening, sample set S ' they include enough matching double points, not only increase intra-office point proportion in sample set, and
Greatly reduce the number of iterations of transformation matrix parameter model H.
Step 3, image co-registration
First according to the projective transformation matrix between adjacent image, projective transformation is carried out to corresponding bridge floor image;Then it uses
Be fade-in gradually go out blending algorithm smooth transition is weighted respectively to tri- Color Channel of RGB of each adjacent bridge floor image, obtain bridge floor
Stitching image.
Specific step is as follows for projective transformation:
(1) as shown in figure 5, according to the transitivity of the projective transformation matrix between adjacent image, with first bridge floor of every row
Image is spliced respectively as the benchmark image of corresponding row according to image line splicing strategy.Between each adjacent bridge floor image
Transformation matrix Hii-1Transmitting transformation is carried out, the transmitting transformation matrix H between each bridge floor image and benchmark image is obtainedi1;Pass through again
Each transformation matrix Hi1Corresponding bridge floor image is respectively mapped in datum plane coordinate system, it is each adjacent in horizontal direction to complete
Image mosaic fusion between image, forms the lateral panoramic picture Image of multiple wide viewing anglesi。
Each transfer matrix transformation for mula is as follows:
H21=H21
H31=H32×H21
Hn1=Hnn-1×Hn-1n-2×…×H21
Wherein, Hii-1For the transformation matrix between (i-1)-th bridge floor image and i-th bridge floor image of same a line, value exists
It is calculated in step 2-2;Hi1For the transformation matrix between the 1st bridge floor image of same a line and i-th bridge floor image;N is same
Amount of images in a line.
(2) as shown in fig. 6, first lateral panoramic picture Image obtained in step (1)1As reference panorama image,
According to image column splicing strategy, spliced.Between the transformation matrix T each lateral panoramic picturejj-1Transmitting transformation is carried out, it obtains
To each lateral panoramic picture ImageiTransmitting transformation matrix T between reference panorama imagej1;Pass through each transmitting transformation matrix again
Tj1Corresponding lateral panoramic picture is respectively mapped in datum plane coordinate system respectively, to complete each adjacent cross on vertical direction
It is merged to the image mosaic between panoramic picture, the description in image mosaic fusion in transfer matrix transformation for mula reference step (1),
Form final bridge floor panoramic picture.
It is fade-in in the present embodiment and gradually goes out blending algorithm by improving, referring specifically to following.
After image registration, further to eliminate the interference that bridge floor image mosaic stitches fracture detection processing, usually to phase
Adjacent bridge floor image pixel value is weighted and averaged, as shown in Figure 7b, distance of the pixel to both sides suture in overlapping region
As fusion weight distinguishing rule.
But since image capture position changes, bridge surface reflection is likely to result in respective pixel in overlapping region
There are hopping phenomenons for point gray value, to eliminate the influence that it generates blending image, are fade-in in tradition and gradually go out Weighted Fusion calculating
One threshold value t of middle introducing.Lap target pixel points are calculated in the corresponding gray scale difference value of two width original images, if the difference is small
In threshold value, illustrate the pixel in former bridge floor image and not shown notable difference, can directly take its weighted average as should
Point pixel value;Conversely, illustrating that image to be spliced, there are light and shade mutation, should take its smooth preceding weight larger under the pixel position
Pixel value as the fusion pixel values.
It is fade-in and gradually goes out in blending algorithm, being fade-in for each fusion pixel values I (x, y) gradually goes out to weight in adjacent image overlapping region
Formula is as follows:
Wherein, I1(x,y)、I2(x, y) is respectively correspondence merging point of the two adjacent bridge floor images in overlapping region
Pixel value, as shown in Figure 7a.d1、d2Gradual change weight factor of the two respectively adjacent bridge floor images in corresponding merging point;Such as figure
Shown in 7b,Withx1、x2The respectively abscissa on overlapping region two sides boundary;X is corresponding merging point
Abscissa;T is gray difference threshold of the two adjacent image overlapping regions on corresponding merging point.
Claims (4)
1. a kind of large area bridge floor image split-joint method, it is characterised in that: step 1 acquires the image for being detected bridge floor one by one, obtains
To bridge floor image collection;Later, the luma component information of each bridge floor image in bridge floor image collection is extracted and counts, and right respectively
The luma component information of each bridge floor image is equalized;Then each bridge floor image is transformed to by frequency domain by Fourier transformation
It is interior, and the translation parameters between image, completion pair are obtained using the phase information of the normalization crosspower spectrum in phase related algorithm
The pre- estimation of overlapping region between adjacent image;
Step 2, image registration
Firstly, extracting SIFT feature in overlapping region between each adjacent image;Then it is sieved by adaptive contrast threshold method
SIFT feature is selected, the feature descriptor being made of matching double points is obtained;And using between each adjacent image of RANSAC algorithm calculating
Projective transformation matrix;
Adaptive contrast threshold method is specific as follows:
(1) characteristic point numerical lower limits N is setmin=200, upper limit Nmax=300, contrast threshold Tc=T0;T0For initial threshold,
Value is 0.02~0.04;
(2) characteristic point detection is carried out, and Statistical Comparison degree is higher than TcCharacteristic point quantity N;
(3) if Nmin≤N≤Nmax, then contrast is higher than TcCharacteristic point be included in initial matching point set, reject contrast and be lower than threshold
Value TcCharacteristic point, and be directly entered step (5);Otherwise, step (4) are executed;
(4) if N < Nmin, then by contrast threshold TcIt is reduced to former numerical valueAnd execute step (3);If N > Nmax, then will be right
Than 2 times that degree threshold value is increased to former numerical value, and execute step (3);
(5) the mistake characteristic point that initial matching point is concentrated is rejected than secondary near neighbor method by arest neighbors, and generates feature descriptor;It is special
Levy in descriptor comprising by pairs of feature point group at multiple matching double points and the distance between each matching double points and direction
Information;
Step 3, image co-registration
First according to the projective transformation matrix between adjacent image, projective transformation is carried out to corresponding bridge floor image;Then it uses and is fade-in
Gradually go out blending algorithm and smooth transition is weighted respectively to tri- Color Channel of RGB of each adjacent bridge floor image, obtains bridge floor splicing
Image;
It is fade-in and gradually goes out in blending algorithm, being fade-in for each merging point pixel value I (x, y) gradually goes out to weight public affairs in adjacent image overlapping region
Formula is as follows:
Wherein, I1(x,y)、I2(x, y) is respectively the pixel of correspondence merging point of the two adjacent bridge floor images in overlapping region
Value;d1、d2Gradual change weight factor of the two respectively adjacent bridge floor images in corresponding merging point;Withx1、x2The respectively abscissa on overlapping region two sides boundary;X is the abscissa of corresponding merging point;T is two adjacent
Gray difference threshold of the image overlapping region on corresponding merging point.
2. a kind of large area bridge floor image split-joint method according to claim 1, it is characterised in that: acquire figure in step 1
The method of picture is specific as follows:
(1) after the internal reference matrix for calculating CCD camera using Zhang Zhengyou plane reference method, radial distortion is obtained by least square method
Coefficient;
(2) CCD camera of the placement by step 1-1 calibration on bridge machinery platform, has carried out according to preset shooting track
The Image Acquisition of whole bridge floor;Preset shooting track is S-shaped;
(3) according to the obtained CCD camera internal reference matrix of step 1-1 and distortion factor each bridge floor image collected to step 1-2
Image calibration is carried out respectively.
3. a kind of large area bridge floor image split-joint method according to claim 1, it is characterised in that: in step 2, RANSAC
The process that algorithm solves projective transformation matrix is as follows:
(1) original training set S is constructed with each matching double points in feature descriptor;Count each matching double points in original training set S
Between Euclidean distance, and by sorting from small to large;
(2) the preceding 85% matching double points building new samples collection S ' of sequence obtained by step (1) is taken;
(3) 4 groups of matching double points are randomly selected from new samples collection S ' forms an interior point set Si, and point in calculating matrix model
Set SiHi, enter step (4);
(4) remaining interior each matching double points of new samples collection S ' are directed to matrix model HiCarry out adaptive test;Error is examined if it exists
Less than the match point of error threshold, then interior point set S is added in the matching double points for examining error to be less than threshold valuei, and execute step
(5);Otherwise, give up matrix model Hi, re-execute (3);
(5) if interior point set SiMiddle element number is greater than defined threshold, then it is assumed that reasonable parameter model is obtained, to updated
Interior point set SiRecalculate matrix model Hi, and use LM algorithmic minimizing cost function;Otherwise, give up matrix model Hi,
And it re-execute the steps (3);
(6) l step (3) is repeated to (5), and l is maximum number of iterations;Later, interior point set S obtained in l iteration is comparedi,
With the maximum interior point set S of element numberiAs final interior point set, and the matrix model H for taking it to calculateiAs adjacent bridge floor
Projective transformation matrix between image.
4. a kind of large area bridge floor image split-joint method according to claim 1, it is characterised in that: in step 3, projection becomes
Change that specific step is as follows:
(1) according to the transitivity of the projective transformation matrix between adjacent image, using first bridge floor image of every row as right
The benchmark image that should be gone is spliced;Between the transformation matrix H each adjacent bridge floor imageii-1Transmitting transformation is carried out, each bridge is obtained
Transmitting transformation matrix H between face image and benchmark imagei1;Pass through each transformation matrix H againi1Corresponding bridge floor image is distinguished
It is mapped in datum plane coordinate system, to complete the image mosaic fusion in horizontal direction between each adjacent image, forms multiple width
The lateral panoramic picture Image at visual anglei;
(2) by first obtained in step (1) lateral panoramic picture Image1Spliced as reference panorama image;To each
Transformation matrix T between lateral panoramic picturejj-1Transmitting transformation is carried out, each lateral panoramic picture Image is obtainediWith reference panorama figure
Transmitting transformation matrix T as betweenj1;Pass through each transmitting transformation matrix T againj1Corresponding lateral panoramic picture is reflected respectively respectively
It is mapped in datum plane coordinate system, to complete the image mosaic fusion on vertical direction between each adjacent transverse panoramic picture, is formed
Final bridge floor panoramic picture.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910561334.9A CN110363706B (en) | 2019-06-26 | 2019-06-26 | Large-area bridge deck image splicing method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910561334.9A CN110363706B (en) | 2019-06-26 | 2019-06-26 | Large-area bridge deck image splicing method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110363706A true CN110363706A (en) | 2019-10-22 |
CN110363706B CN110363706B (en) | 2023-03-21 |
Family
ID=68216594
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910561334.9A Active CN110363706B (en) | 2019-06-26 | 2019-06-26 | Large-area bridge deck image splicing method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110363706B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112734720A (en) * | 2021-01-08 | 2021-04-30 | 沈阳工业大学 | Ship hull laser cleaning in-place detection method and system based on visual identification |
CN115266282A (en) * | 2021-04-30 | 2022-11-01 | 武汉兰丁云医学检验实验室有限公司 | Pathological image automatic identification method based on combination of three methods of HE staining, ki67 and P16 |
CN116543267A (en) * | 2023-07-04 | 2023-08-04 | 宁德时代新能源科技股份有限公司 | Image set processing method, image segmentation device, image set processing apparatus, image segmentation device, and storage medium |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101916445A (en) * | 2010-08-25 | 2010-12-15 | 天津大学 | Affine parameter estimation-based image registration method |
US20130216089A1 (en) * | 2010-04-22 | 2013-08-22 | The University Of North Carolina At Charlotte | Method and System for Remotely Inspecting Bridges and Other Structures |
CN104599249A (en) * | 2015-01-13 | 2015-05-06 | 重庆大学 | Cable rope bearing bridge deck vehicle load distribution real-time detection method |
CN106910208A (en) * | 2017-03-07 | 2017-06-30 | 中国海洋大学 | A kind of scene image joining method that there is moving target |
CN107424181A (en) * | 2017-04-12 | 2017-12-01 | 湖南源信光电科技股份有限公司 | A kind of improved image mosaic key frame rapid extracting method |
EP3319038A1 (en) * | 2015-06-30 | 2018-05-09 | Baidu Online Network Technology (Beijing) Co., Ltd. | Panoramic image generation method and apparatus for user terminal |
CN108921848A (en) * | 2018-09-29 | 2018-11-30 | 长安大学 | Bridge Defect Detecting device and detection image joining method based on more mesh cameras |
CN109754368A (en) * | 2019-01-23 | 2019-05-14 | 郑州工程技术学院 | A kind of crack joining method in bridge quality testing |
-
2019
- 2019-06-26 CN CN201910561334.9A patent/CN110363706B/en active Active
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20130216089A1 (en) * | 2010-04-22 | 2013-08-22 | The University Of North Carolina At Charlotte | Method and System for Remotely Inspecting Bridges and Other Structures |
CN101916445A (en) * | 2010-08-25 | 2010-12-15 | 天津大学 | Affine parameter estimation-based image registration method |
CN104599249A (en) * | 2015-01-13 | 2015-05-06 | 重庆大学 | Cable rope bearing bridge deck vehicle load distribution real-time detection method |
EP3319038A1 (en) * | 2015-06-30 | 2018-05-09 | Baidu Online Network Technology (Beijing) Co., Ltd. | Panoramic image generation method and apparatus for user terminal |
US20180365803A1 (en) * | 2015-06-30 | 2018-12-20 | Baidu Online Network Technology (Beijing) Co., Ltd | Panoramic image generation method and apparatus for user terminal |
CN106910208A (en) * | 2017-03-07 | 2017-06-30 | 中国海洋大学 | A kind of scene image joining method that there is moving target |
CN107424181A (en) * | 2017-04-12 | 2017-12-01 | 湖南源信光电科技股份有限公司 | A kind of improved image mosaic key frame rapid extracting method |
CN108921848A (en) * | 2018-09-29 | 2018-11-30 | 长安大学 | Bridge Defect Detecting device and detection image joining method based on more mesh cameras |
CN109754368A (en) * | 2019-01-23 | 2019-05-14 | 郑州工程技术学院 | A kind of crack joining method in bridge quality testing |
Non-Patent Citations (4)
Title |
---|
LARAQUI, A.1;BAATAOUI, A.1;SAAIDI, A.1;JARRAR, A.1;MASRAR, M.1;S: "Image mosaicing using voronoi diagram", 《MULTIMEDIA TOOLS & APPLICATIONS》 * |
宋佳乾等: "基于改进SIFT特征点匹配的图像拼接算法", 《计算机测量与控制》 * |
张静等: "基于SIFT特征和误匹配逐次去除的图像拼接", 《半导体光电》 * |
肖映彩等: "无人机航拍图像拼接的研究与实现", 《通信技术》 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112734720A (en) * | 2021-01-08 | 2021-04-30 | 沈阳工业大学 | Ship hull laser cleaning in-place detection method and system based on visual identification |
CN112734720B (en) * | 2021-01-08 | 2024-03-05 | 沈阳工业大学 | Ship hull laser cleaning in-situ detection method and system based on visual identification |
CN115266282A (en) * | 2021-04-30 | 2022-11-01 | 武汉兰丁云医学检验实验室有限公司 | Pathological image automatic identification method based on combination of three methods of HE staining, ki67 and P16 |
CN116543267A (en) * | 2023-07-04 | 2023-08-04 | 宁德时代新能源科技股份有限公司 | Image set processing method, image segmentation device, image set processing apparatus, image segmentation device, and storage medium |
CN116543267B (en) * | 2023-07-04 | 2023-10-13 | 宁德时代新能源科技股份有限公司 | Image set processing method, image segmentation device, image set processing apparatus, image segmentation device, and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN110363706B (en) | 2023-03-21 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Lei et al. | New crack detection method for bridge inspection using UAV incorporating image processing | |
CN110378879B (en) | Bridge crack detection method | |
CN108764257B (en) | Multi-view pointer instrument identification method | |
CN112818988B (en) | Automatic identification reading method and system for pointer instrument | |
CN109583324A (en) | A kind of pointer meters reading automatic identifying method based on the more box detectors of single-point | |
CN105069743B (en) | Detector splices the method for real time image registration | |
CN104599258B (en) | A kind of image split-joint method based on anisotropic character descriptor | |
CN109145958B (en) | A kind of real scene wisp detection method generating confrontation network based on multitask | |
CN104134200B (en) | Mobile scene image splicing method based on improved weighted fusion | |
CN110363706A (en) | A kind of large area bridge floor image split-joint method | |
Li et al. | Automatic bridge crack identification from concrete surface using ResNeXt with postprocessing | |
CN102297660A (en) | Measuring method of shield tunnel lining segment seam open width and apparatus thereof | |
CN115409069A (en) | Village and town building identification method, classification method, device, electronic equipment and medium | |
CN110232387A (en) | A kind of heterologous image matching method based on KAZE-HOG algorithm | |
CN111598942A (en) | Method and system for automatically positioning electric power facility instrument | |
CN109341668A (en) | Polyphaser measurement method based on refraction projection model and beam ray tracing method | |
CN102494663A (en) | Measuring system of swing angle of swing nozzle and measuring method of swing angle | |
Yang et al. | Deep neural network based visual inspection with 3d metric measurement of concrete defects using wall-climbing robot | |
CN103942786B (en) | The self adaptation block objects detection method of unmanned plane visible ray and infrared image | |
CN114419421A (en) | Subway tunnel crack identification system and method based on images | |
CN116630267A (en) | Roadbed settlement monitoring method based on unmanned aerial vehicle and laser radar data fusion | |
CN117131441B (en) | Night light pollution monitoring method, device, computer equipment and storage medium | |
Li et al. | Rotating box multi-objective visual tracking algorithm for vibration displacement measurement of large-span flexible bridges | |
CN109740598A (en) | Object localization method and device under structuring scene | |
CN117422699A (en) | Highway detection method, highway detection device, computer equipment and storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |