CN104298993A - Pole number positioning and identification method suitable for railway line in complex scene - Google Patents
Pole number positioning and identification method suitable for railway line in complex scene Download PDFInfo
- Publication number
- CN104298993A CN104298993A CN201410010810.5A CN201410010810A CN104298993A CN 104298993 A CN104298993 A CN 104298993A CN 201410010810 A CN201410010810 A CN 201410010810A CN 104298993 A CN104298993 A CN 104298993A
- Authority
- CN
- China
- Prior art keywords
- image
- bar
- electric pole
- line
- carried out
- 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 57
- 238000012549 training Methods 0.000 claims abstract description 24
- 239000011159 matrix material Substances 0.000 claims abstract description 19
- 238000001914 filtration Methods 0.000 claims abstract description 16
- 230000011218 segmentation Effects 0.000 claims abstract description 11
- 230000009466 transformation Effects 0.000 claims abstract description 10
- 230000004807 localization Effects 0.000 claims description 9
- 239000000284 extract Substances 0.000 claims description 7
- 238000006243 chemical reaction Methods 0.000 claims description 4
- 238000012706 support-vector machine Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 abstract description 3
- 238000012958 reprocessing Methods 0.000 abstract 1
- 238000001514 detection method Methods 0.000 description 9
- 238000010586 diagram Methods 0.000 description 7
- 238000004458 analytical method Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000012544 monitoring process Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 2
- 239000012535 impurity Substances 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 206010016256 fatigue Diseases 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011897 real-time detection Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 230000036548 skin texture Effects 0.000 description 1
- 239000000725 suspension Substances 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/09—Recognition of logos
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a pole number positioning and identification method suitable for railway line in a complex scene. The method is characterized by to begin with, carrying out reduction and median filter preprocessing on scene images collected along a railway line; then, finding edges by utilizing a large-dimension sobel filter, and detecting the strongest edge through Hough transformation, that is, determining the position of a telegraph pole; next, specifically positioning number regions by utilizing number plate co-occurrence matrix textural features; carrying out digit identification on pole number: carrying out filtering, binaryzation and segmentation on the images to obtain 64-dimension feature of each digital image, carrying out identification through a training model, and after the number plate identification is finished, reprocessing all identification results; and carrying out number plate self-correction according to the unique characters of the number plates, and through the self-correction, the number plate identification rate reaches 90% and the above.
Description
Technical field
The present invention relates to a kind of bar number location and recognition methods be applicable under track complex scene along the line.
Background technology
Along with improving constantly of railway total size, speed is constantly accelerated, and run interval and constantly shorten, the security that result in railway is more and more important, and Detection task is more and more heavier.Traditional mode by artificial check along the line can not adapt to the fast development of railway far away, and uses machine vision to be trend of the times.
For guaranteeing high-speed railway operation order, improve Supply Security, reliability, the Ministry of Railways constructs high-speed railway power supply safety and detects monitoring system (6C system), is below referred to as 6C system.Its objective is and carry out comprehensive to the tractive power supply system of high-speed railway, the comprehensive detection monitoring of all standing, major function comprises the detection to high speed contact net suspension parameter and bow net operational factor, to overhead contact line, cantilever structure, the detection of attached clue and parts, to the real-time detection of catenary ' s parameters, to the Real-Time Monitoring in motor train unit pantograph pan state and contact net special cross section and place, to the real-time online detection etc. of contact net operational factor and power-supply unit parameter, for the safe operation of high ferro power-supply unit, the comprehensive analysis of running status and parameter, the maintenance of equipment provides technical basis.The scope of application of bar recognition system refers to whether system normally can be run under different image-forming conditions.Whether more common dislikes slightly because have weather condition, and whether available light is sufficient, and whether bar pattern is consistent etc.In general Vision Builder for Automated Inspection is all carry out in indoor, rarely has outdoor.Outdoor difficulty is, weather condition is uncontrollable, and ambient light is uncontrollable.In addition, bar number and camera distance are not fixed, and bar pattern is fixing etc. both increases the difficulty of machine recognition.Not yet find that there is the associated software product that bar number identifies in the market, the identification of Railway Bureau to bar number can only by manually identifying thousands of picture.Manual detection efficiency is low, dull in content, easily tired out, and testing result is larger by artificial affecting.Common target localization mainly contains License Plate, but during License Plate, background is generally ground, and car plate self color is relatively more special, and in bar recognition technology, bar location is a vital step.This is because in natural scene, background complexity is various, just becomes abnormal difficult to the location of bar number.
Summary of the invention
The object of this invention is to provide a kind of bar number location and recognition methods be applicable under track complex scene along the line, to solve the low problem in current bar number location and the location difficulty existed respectively in identifying and recognition efficiency.
The present invention is for solving the problems of the technologies described above and providing a kind of bar localization method be applicable under track complex scene along the line, and this localization method comprises the following steps:
1) scene image that acquisition trajectory is along the line, and pre-service is carried out to the scene image gathered, make image be convenient to segmentation and identify;
2) edge extracting is carried out to pretreated image, determine electric pole region;
3) utilize number plate co-occurrence matrix textural characteristics, the numeric area of bar number is located out.
Pre-service in described step 1) comprises image scaling and filtering two steps, and wherein image scaling carries out convergent-divergent according to the width of electric pole in image.
Described step 2) comprise the following steps:
A. edge extracting is carried out to pretreated image;
B. Hough transformation is adopted to ask for the maximum linear at extracted edge, and calculated line angle;
C. shift initial acquisition picture position onto according to the position of required maximum linear is counter, centered by required straight line, one times of electric pole width is respectively expanded in left and right, obtains preliminary electric pole region;
D. rotating obtaining the angle that image calculates according to step in step C, making electric pole be in vertical direction in the picture;
E. according to the numerical projection left hand edge of local maximum determination electric pole and right hand edge, the region between left and right edges is referred to as electric pole region.
The pixel joint probability of what the co-occurrence matrix in described step 3) represented is two positions is close, for reacting the integrated information of gradation of image about direction, adjacent spaces and amplitude of variation, f (x, y) be a width two-dimensional digital image, its size is M × N, grey level is Ng, then the gray level co-occurrence matrixes meeting certain space relation is:
P(i,j)=#｛(x1,y1),(x2,y2)∈M×N｜f(x1,y1)=i,f(x2,y2)=j｝
Wherein # (x) represents the element number in set x, and P is the matrix of Ng × Ng, if (x1, y1) and (x2, y2) spacing is d, and the angle of both and abscissa line is θ, the gray level co-occurrence matrixes P (i of various spacing and angle, j, d, θ).
The present invention solves the problems of the technologies described above to additionally provide a kind of bar recognition methods be applicable under track complex scene along the line, and this recognition methods comprises the following steps:
1) scene image that acquisition trajectory is along the line, and pre-service is carried out to the scene image gathered, make image be convenient to segmentation and identify;
2) edge extracting is carried out to pretreated image, determine electric pole region;
3) utilize number plate co-occurrence matrix textural characteristics, the numeric area of bar number is located out;
4) location digital picture is out carried out 64 dimensional features that filtering, binaryzation and segmentation obtain each digital picture, identify by training pattern, determine bar number.
Described recognition methods also comprises carries out self-tuning step to identified bar number, this step generates correct bar sequence according to the initial bar manually determined number, then mate with the actual bar identified number, interpolation is carried out for the bar number missed in reality, and extrapolates undetected bar number corresponding frame number.
Pre-service in described step 1) comprises image scaling and filtering two steps, and wherein image scaling carries out convergent-divergent according to the width of electric pole in image, makes the electric pole width after convergent-divergent be.
Described step 2) comprise the following steps:
A. edge extracting is carried out to pretreated image;
B. Hough transformation is adopted to ask for the maximum linear at extracted edge, and calculated line angle;
C. shift initial acquisition picture position onto according to the position of required maximum linear is counter, centered by required straight line, one times of electric pole width is respectively expanded in left and right, obtains preliminary electric pole region;
D. rotating obtaining the angle that image calculates according to step in step C, making electric pole be in vertical direction in the picture;
E. according to the numerical projection left hand edge of local maximum determination electric pole and right hand edge, the region between left and right edges is referred to as electric pole region;
F. co-occurrence matrix calculating is carried out to the electric pole region determined, find the region that textural characteristics is the most close.
Bar identification in described step 3) comprises the following steps:
A. the bar region gathering image arranges, as training sample;
B. binary conversion treatment is carried out to image;
C. the image after binaryzation is carried out horizontal projection, cut out each digital picture;
D. each digital image scaling cut out is become 8*8 size, extract 64 dimensional features, identify by training pattern.
Training pattern in described steps d is obtained by off-line training, and described off-line training comprises the following steps:
A) the bar area image gathering image carries out manual sorting, as training sample;
B) adopt maximum variance between clusters that figure is carried out binaryzation;
C) image of binaryzation is carried out horizontal projection, cut out each digital picture, collect 0-9 totally 10 class digital pictures;
D) the 10 class digital image scalings cut out are become 8*8 size, extract 64 dimensional features, by support vector machine, feature is trained, obtain the training pattern of each numeral.
The invention has the beneficial effects as follows: the scene image that acquisition trajectory of the present invention is along the line, first whole image is reduced and the pre-service of medium filtering, then edge is found with the sobel wave filter that dimension is larger, the strongest edge is detected again by Hough transformation, namely the position of electric pole is determined, then utilize number plate co-occurrence matrix textural characteristics, numeric area is specifically located out; Then the numeral identification of bar number is carried out, image is carried out 64 dimensional features that filtering, binaryzation and segmentation obtain each digital picture, identify by training pattern, and after number plate identification completes, again can process all recognition results, character specific to number plate, carries out number plate self-correcting, by self-correcting, the discrimination of number plate is made to obtain more than 90%.
Accompanying drawing explanation
Fig. 1 is the process flow diagram being applicable to the recognition methods of track complex scene lower beam along the line of the present invention;
Fig. 2 is the schematic diagram in electric pole region and number plate region in the embodiment of the present invention;
Fig. 3 is the process flow diagram of embodiment of the present invention king-rod number location;
Fig. 4 is the embodiment of the present invention king-rod number process flow diagram identified;
Fig. 5 is the embodiment of the present invention king-rod number off-line training process flow diagram identified;
Fig. 6 is the embodiment of the present invention king-rod number ONLINE RECOGNITION process flow diagram identified.
Embodiment
Below in conjunction with accompanying drawing, the specific embodiment of the present invention is further described.
A kind of embodiment being applicable to track complex scene lower beam along the line localization method of the present invention
First the present invention adopts the means such as image filtering, morphology to carry out pre-service to picture, then according to edge feature and textural characteristics, bar number is located out, carrying out the numeral identification of bar number, after whole segment data identification completes, improve discrimination by aftertreatment.Whole process as shown in Figure 3, mainly comprises the following steps:
1. the scene image that acquisition trajectory is along the line, and pre-service is carried out to the scene image gathered, make image be convenient to segmentation and identify.
First this preprocessing process carries out convergent-divergent according to the width of electric pole in image, after making convergent-divergent, electric pole width is 10 ~ 20, general minification is the integral multiple of 2, because background in image is complicated, disturb more, but the general width of electric pole is larger, by reducing (being sample drawn point in essence), the effect that little impurity disappears can be reached, therefore very crucial to raising alignment quality; Medium filtering is carried out to the image after reducing.
2. bar number location
This step comprise to process after image carry out edge extracting, edge extracting adopt sobel realize, the size of convolution kernel suitably strengthens; To the edge that sobel extracts, ask for maximum linear with Hough transformation, and calculated line angle.Initial acquisition picture position is shifted onto according to required position is counter, and centered by required straight line, the bar width of each expansion in left and right one times of electric wire, obtain preliminary electric pole region, namely Image semantic classification is to some process of image, is only used to the position of positioning wires bar, due in the process of Image semantic classification, image self-information is lost huge, more pinpoint time, adopt subregion in original image.Angle according to calculating rotates, electric pole is corrected as vertical direction, in actual picture gatherer process, to pitch due to train and actual electric pole also has inclination conditions, therefore in picture, the angle of standing of electric pole is inconsistent, needs it samely to rotate to vertical direction.After rotating to vertical direction, edge is extracted with sobel, left hand edge and the right hand edge of electric pole is obtained with vertical projection local maximum, region between left and right edges is referred to as electric pole region, as shown in Figure 2, electric pole region is more accurate, and scope is less, fewer to the interference of Subsequent number board regional search.
Co-occurrence matrix calculating is carried out to the image obtained by said process, find the most proximate region of textural characteristics, the co-occurrence matrix joint probability density of the pixel of two positions defines, the i.e. distribution character of reaction brightness, also reflection have same brightness or close to brightness pixel between position distribution characteristic, it is the second-order statistics feature of relevant brightness of image change, it is the basis of definition one group of textural characteristics, the gray level co-occurrence matrixes of piece image can reflect that gradation of image is about direction, the integrated information of adjacent spaces and amplitude of variation, it is the local mode of analysis chart picture and the basis of their queueing disciplines.
One of the present invention is applicable to the recognition methods of track complex scene lower beam along the line
First recognition methods in the present embodiment adopts the means such as image filtering, morphology to carry out pre-service to picture, then according to edge feature and textural characteristics, bar number is located out, carrying out the numeral identification of bar number, after whole segment data identification completes, improve discrimination by aftertreatment.Whole process as shown in Figure 1, mainly comprises the following steps:
1. the scene image that acquisition trajectory is along the line, and pre-service is carried out to the scene image gathered, make image be convenient to segmentation and identify.
First this preprocessing process carries out convergent-divergent according to the width of electric pole in image, after making convergent-divergent, electric pole width is 10 ~ 20, general minification is the integral multiple of 2, because background in image is complicated, disturb more, but the general width of electric pole is larger, by reducing (being sample drawn point in essence), the effect that little impurity disappears can be reached, therefore very crucial to raising alignment quality; Medium filtering is carried out to the image after reducing.
2. bar number location
This step as shown in Figure 3, comprise to process after image carry out edge extracting, edge extracting adopt sobel realize, the size of convolution kernel suitably strengthens; To the edge that sobel extracts, ask for maximum linear with Hough transformation, and calculated line angle.Initial acquisition picture position is shifted onto according to required position is counter, and centered by required straight line, the bar width of each expansion in left and right one times of electric wire, obtain preliminary electric pole region, namely Image semantic classification is to some process of image, is only used to the position of positioning wires bar, due in the process of Image semantic classification, image self-information is lost huge, more pinpoint time, adopt subregion in original image.Angle according to calculating rotates, electric pole is corrected as vertical direction, in actual picture gatherer process, to pitch due to train and actual electric pole also has inclination conditions, therefore in picture, the angle of standing of electric pole is inconsistent, needs it samely to rotate to vertical direction.After rotating to vertical direction, edge is extracted with sobel, left hand edge and the right hand edge of electric pole is obtained with vertical projection local maximum, region between left and right edges is referred to as electric pole region, as shown in Figure 2, electric pole region is more accurate, and scope is less, fewer to the interference of Subsequent number board regional search.
Co-occurrence matrix calculating is carried out to the image obtained by said process, find the most proximate region of textural characteristics, the co-occurrence matrix joint probability density of the pixel of two positions defines, the i.e. distribution character of reaction brightness, also reflection have same brightness or close to brightness pixel between position distribution characteristic, it is the second-order statistics feature of relevant brightness of image change, it is the basis of definition one group of textural characteristics, the gray level co-occurrence matrixes of piece image can reflect that gradation of image is about direction, the integrated information of adjacent spaces and amplitude of variation, it is the local mode of analysis chart picture and the basis of their queueing disciplines.
The gray level co-occurrence matrixes of piece image can reflect the integrated information of gradation of image about direction, adjacent spaces, amplitude of variation, and it is the local mode of analysis chart picture and the basis of their queueing disciplines.
If f (x, y) is a width two-dimensional digital image, its size is M × N, and grey level is Ng, then the gray level co-occurrence matrixes meeting certain space relation is
P(i,j)=#｛(x1,y1),(x2,y2)∈M×N｜f(x1,y1)=i,f(x2,y2)=j｝
Wherein # (x) represents the element number in set x, obvious P is the matrix of Ng × Ng, if (x1, y1) be d with (x2, y2) spacing, the angle of both and abscissa line is θ, then can obtain the gray level co-occurrence matrixes P (i of various spacing and angle, j, d, θ).
The textural characteristics of number field, needs to intercept a large amount of number plate regions in advance, carries out calculating and on average obtain.Main variance and these two features of correlativity using co-occurrence matrix in this example.
When carrying out images match search, first determine the width of electric pole, then estimate the width in number plate region, according to fixing depth-width ratio, thus number plate region height determined, determined by d further, θ is 90 °.From left to right, from top to bottom, calculate co-occurrence matrix successively, find the place meeting threshold value, bar region is determined.
3. bar number identifies
This process comprises off-line training and ONLINE RECOGNITION two parts, and process flow diagram is see Fig. 4, Fig. 5, Fig. 6.Off-line training step mainly comprises following steps: first carry out manual sorting, as training sample to the bar area image gathering image; Then with maximum variance between clusters by image binaryzation; Then Iamge Segmentation, carries out horizontal projection to image, cuts out each digital picture, needs to collect 0-9 totally 10 class digital pictures; Finally to the numeral cut out, digital image scaling is become 8*8 size, extract 64 dimensional features, by support vector machine, feature is trained, obtain the training pattern of each numeral.
The ONLINE RECOGNITION stage mainly comprises following steps: during ONLINE RECOGNITION, first carries out image denoising.By medium filtering, picture noise is removed.Then the image binaryzation operation described in carry out step 31, the image segmentation operations described in then carry out step 31.Finally to 64 dimensional features described in the digital extraction step 31 cut out, identify by training pattern.In this example, for the numeral of identification error as sample, again train, Renewal model is to improve discrimination.
4. aftertreatment
Under normal circumstances, differ from the integral multiple of 2, be namely all odd number or even number between bar number, the monotonicity of bar number is consistent, or increases progressively, or successively decreases, and adds up large gauge rod number, obtains parity and the monotonicity of leading portion bar number, is abandoned by wrong bar number.In this example, select 200 recognition results.If the result after screening is compared with result before screening, quantity reduces huge, illustrates that this section of result identification may have problems, do not process, to prevent from makeing mistakes, threshold value is set to 70% in this example, and namely the result of at least 70% is retained and just can processes.According to the initial bar number manually determined, automatically generate correct bar sequence.Then mate with the actual bar detected number, for the bar number missed in reality, carry out interpolation, and extrapolate undetected bar number corresponding frame number.When coupling, situation process be divided.Such as should 0126, but recognition result is 0166.Because result difference is huge, so directly judge that recognition result is as error result.Both requiring in this example, difference is no more than 20.Such as should 0126, but recognition result is 0128.When there is this situation, matching process to consider two kinds may, a) 0128 is correct, inserts 0126b) 0128 be wrong.When resolution these two kinds is possible, generally needs special processing, delay and provide result of determination.Namely also need to judge result several after this result, adopt 15 recognition results in this example.
The proposed by the invention bar recognition methods be applicable under track complex scene along the line, by scale transformation cleverly, medium filtering, large filtering core rim detection, Hough transformation straight-line detection all serial of methods, fast and accurately by electric pole zone location.Afterwards in electric pole region, find concrete number plate region by skin texture detection, these are that the identification of number plate lays the foundation.After number plate identification completes, again can process all recognition results.Character specific to number plate, carries out number plate self-correcting, by self-correcting, makes the discrimination of number plate obtain more than 90%.
The foregoing is only preferred embodiment of the present invention, do not limit to scope of patent protection of the present invention, therefore the equivalence change of such as using instructions of the present invention and graphic content to do, be all in like manner all contained in the scope of the present invention.
Claims (10)
1. be applicable to the bar localization method under track complex scene along the line, it is characterized in that, this localization method comprises the following steps:
1) scene image that acquisition trajectory is along the line, and pre-service is carried out to the scene image gathered, make image be convenient to segmentation and identify;
2) edge extracting is carried out to pretreated image, determine electric pole region;
3) utilize number plate co-occurrence matrix textural characteristics, the numeric area of bar number is located out.
2. the bar localization method be applicable under track complex scene along the line according to claim 1, it is characterized in that, pre-service in described step 1) comprises image scaling and filtering two steps, and wherein image scaling carries out convergent-divergent according to the width of electric pole in image.
3. the bar localization method be applicable under track complex scene along the line according to claim 1, is characterized in that, described step 2) comprise the following steps:
A. edge extracting is carried out to pretreated image;
B. Hough transformation is adopted to ask for the maximum linear at extracted edge, and calculated line angle;
C. shift initial acquisition picture position onto according to the position of required maximum linear is counter, centered by required straight line, one times of electric pole width is respectively expanded in left and right, obtains preliminary electric pole region;
D. rotating obtaining the angle that image calculates according to step in step C, making electric pole be in vertical direction in the picture;
E. according to the numerical projection left hand edge of local maximum determination electric pole and right hand edge, the region between left and right edges is referred to as electric pole region.
4. the bar localization method be applicable under track complex scene along the line according to claim 1, it is characterized in that, the pixel joint probability of what the co-occurrence matrix in described step 3) represented is two positions is close, for reacting the integrated information of gradation of image about direction, adjacent spaces and amplitude of variation, f (x, y) is a width two-dimensional digital image, and its size is M × N, grey level is Ng, then the gray level co-occurrence matrixes meeting certain space relation is:
P(i,j)=#｛(x1,y1),(x2,y2)∈M×N｜f(x1,y1)=i,f(x2,y2)=j｝
Wherein # (x) represents the element number in set x, and P is the matrix of Ng × Ng, if (x1, y1) and (x2, y2) spacing is d, and the angle of both and abscissa line is θ, the gray level co-occurrence matrixes P (i of various spacing and angle, j, d, θ).
5. be applicable to the bar recognition methods under track complex scene along the line, it is characterized in that, this recognition methods comprises the following steps:
1) scene image that acquisition trajectory is along the line, and pre-service is carried out to the scene image gathered, make image be convenient to segmentation and identify;
2) edge extracting is carried out to pretreated image, determine electric pole region;
3) utilize number plate co-occurrence matrix textural characteristics, the numeric area of bar number is located out;
4) location digital picture is out carried out 64 dimensional features that filtering, binaryzation and segmentation obtain each digital picture, identify by training pattern, determine bar number.
6. the bar recognition methods be applicable under track complex scene along the line according to claim 5, it is characterized in that, described recognition methods also comprises carries out self-tuning step to identified bar number, this step generates correct bar sequence according to the initial bar manually determined number, then mate with the actual bar identified number, interpolation is carried out for the bar number missed in reality, and extrapolates undetected bar number corresponding frame number.
7. the bar recognition methods be applicable under track complex scene along the line according to claim 5 or 6, it is characterized in that, pre-service in described step 1) comprises image scaling and filtering two steps, and wherein image scaling carries out convergent-divergent according to the width of electric pole in image.
8. the bar recognition methods be applicable under track complex scene along the line according to claim 5 or 6, is characterized in that, described step 2) comprise the following steps:
A. edge extracting is carried out to pretreated image;
B. Hough transformation is adopted to ask for the maximum linear at extracted edge, and calculated line angle;
C. shift initial acquisition picture position onto according to the position of required maximum linear is counter, centered by required straight line, one times of electric pole width is respectively expanded in left and right, obtains preliminary electric pole region;
D. rotating obtaining the angle that image calculates according to step in step C, making electric pole be in vertical direction in the picture;
E. according to the numerical projection left hand edge of local maximum determination electric pole and right hand edge, the region between left and right edges is referred to as electric pole region;
F. co-occurrence matrix calculating is carried out to the electric pole region determined, find the region that textural characteristics is the most close.
9. the bar recognition methods be applicable under track complex scene along the line according to claim 5 or 6, it is characterized in that, the bar identification in described step 3) comprises the following steps:
A. the bar region gathering image arranges, as training sample;
B. binary conversion treatment is carried out to image;
C. the image after binaryzation is carried out horizontal projection, cut out each digital picture;
D. each digital image scaling cut out is become 8*8 size, extract 64 dimensional features, identify by training pattern.
10. the bar recognition methods be applicable under track complex scene along the line according to claim 9, it is characterized in that, the training pattern in described steps d is obtained by off-line training, and described off-line training comprises the following steps:
A) the bar area image gathering image carries out manual sorting, as training sample;
B) adopt maximum variance between clusters that figure is carried out binaryzation;
C) image of binaryzation is carried out horizontal projection, cut out each digital picture, collect 0-9 totally 10 class digital pictures;
D) the 10 class digital image scalings cut out are become 8*8 size, extract 64 dimensional features, by support vector machine, feature is trained, obtain the training pattern of each numeral.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410010810.5A CN104298993B (en) | 2014-01-09 | 2014-01-09 | A kind of bar number positioning and recognition methods suitable under complex scene along track |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410010810.5A CN104298993B (en) | 2014-01-09 | 2014-01-09 | A kind of bar number positioning and recognition methods suitable under complex scene along track |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104298993A true CN104298993A (en) | 2015-01-21 |
| CN104298993B CN104298993B (en) | 2018-11-09 |
Family
ID=52318714
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410010810.5A Expired - Fee Related CN104298993B (en) | 2014-01-09 | 2014-01-09 | A kind of bar number positioning and recognition methods suitable under complex scene along track |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104298993B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105719302A (en) * | 2016-01-25 | 2016-06-29 | 成都国铁电气设备有限公司 | Contact net post device recognition and location method and system |
| CN107084666A (en) * | 2017-05-10 | 2017-08-22 | 中国计量大学 | Brake block dimension synthesis detection method based on machine vision |
| CN108416354A (en) * | 2018-03-09 | 2018-08-17 | 国电南瑞科技股份有限公司 | Shaft tower serial number extracting method based on helicopter routing inspection transmission line of electricity |
| WO2018195741A1 (en) * | 2017-04-24 | 2018-11-01 | 李卓希 | Driving environment identification method and system |
| CN109584220A (en) * | 2018-11-15 | 2019-04-05 | 锐捷网络股份有限公司 | A kind of silk-screen localization method, silk-screen defect inspection method and relevant apparatus |
| CN109816253A (en) * | 2019-01-29 | 2019-05-28 | 成都国铁电气设备有限公司 | A kind of defect analysis method based on bar number identification |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1845605A (en) * | 2006-05-15 | 2006-10-11 | 华北电力大学（北京） | Image processing and recognizing method for real-time monitoring power high voltage circuit breaker switch state |
| CN101383003A (en) * | 2008-10-31 | 2009-03-11 | 江西赣粤高速公路股份有限公司 | Real-time precise recognition method for vehicle number board |
| CN101739556A (en) * | 2010-01-21 | 2010-06-16 | 上海交通大学 | Method for automatically identifying number of steel billet |
| CN101753846A (en) * | 2008-12-05 | 2010-06-23 | 三星电子株式会社 | Use camera to adjust the apparatus and method of character boundary automatically |
| CN102419820A (en) * | 2011-08-18 | 2012-04-18 | 电子科技大学 | Method for rapidly detecting car logo in videos and images |
| CN102782707A (en) * | 2010-03-11 | 2012-11-14 | 微软公司 | Resolution adjustment of an image that includes text undergoing an OCR process |
| CN102855499A (en) * | 2012-08-23 | 2013-01-02 | 西南交通大学 | Method for determining railway contact net main column identification information based on combination of multiple clue data pipelines |
| CN102902974A (en) * | 2012-08-23 | 2013-01-30 | 西南交通大学 | Image based method for identifying railway overhead-contact system bolt support identifying information |
| CN103106395A (en) * | 2012-12-15 | 2013-05-15 | 长春理工大学 | Geometry normalization kernel device of license plate character real-time recognition system |
-
2014
- 2014-01-09 CN CN201410010810.5A patent/CN104298993B/en not_active Expired - Fee Related
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1845605A (en) * | 2006-05-15 | 2006-10-11 | 华北电力大学（北京） | Image processing and recognizing method for real-time monitoring power high voltage circuit breaker switch state |
| CN101383003A (en) * | 2008-10-31 | 2009-03-11 | 江西赣粤高速公路股份有限公司 | Real-time precise recognition method for vehicle number board |
| CN101753846A (en) * | 2008-12-05 | 2010-06-23 | 三星电子株式会社 | Use camera to adjust the apparatus and method of character boundary automatically |
| CN101739556A (en) * | 2010-01-21 | 2010-06-16 | 上海交通大学 | Method for automatically identifying number of steel billet |
| CN102782707A (en) * | 2010-03-11 | 2012-11-14 | 微软公司 | Resolution adjustment of an image that includes text undergoing an OCR process |
| CN102419820A (en) * | 2011-08-18 | 2012-04-18 | 电子科技大学 | Method for rapidly detecting car logo in videos and images |
| CN102855499A (en) * | 2012-08-23 | 2013-01-02 | 西南交通大学 | Method for determining railway contact net main column identification information based on combination of multiple clue data pipelines |
| CN102902974A (en) * | 2012-08-23 | 2013-01-30 | 西南交通大学 | Image based method for identifying railway overhead-contact system bolt support identifying information |
| CN103106395A (en) * | 2012-12-15 | 2013-05-15 | 长春理工大学 | Geometry normalization kernel device of license plate character real-time recognition system |
Non-Patent Citations (3)
| Title |
|---|
| 张泊平等: "基于运动车辆的摄像机自适应控制系统研究", 《计算机应用与软件》 * |
| 编码代号47: "共生矩阵", 《百度百科》 * |
| 陈烁华等: "Sobel-Hough检测方法在机器视觉倒车辅助系统的应用研究", 《机械设计与制造》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105719302A (en) * | 2016-01-25 | 2016-06-29 | 成都国铁电气设备有限公司 | Contact net post device recognition and location method and system |
| CN105719302B (en) * | 2016-01-25 | 2018-08-10 | 成都国铁电气设备有限公司 | The recognition positioning method and system of catenary mast device |
| WO2018195741A1 (en) * | 2017-04-24 | 2018-11-01 | 李卓希 | Driving environment identification method and system |
| CN107084666A (en) * | 2017-05-10 | 2017-08-22 | 中国计量大学 | Brake block dimension synthesis detection method based on machine vision |
| CN108416354A (en) * | 2018-03-09 | 2018-08-17 | 国电南瑞科技股份有限公司 | Shaft tower serial number extracting method based on helicopter routing inspection transmission line of electricity |
| CN109584220A (en) * | 2018-11-15 | 2019-04-05 | 锐捷网络股份有限公司 | A kind of silk-screen localization method, silk-screen defect inspection method and relevant apparatus |
| CN109816253A (en) * | 2019-01-29 | 2019-05-28 | 成都国铁电气设备有限公司 | A kind of defect analysis method based on bar number identification |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104298993B (en) | 2018-11-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109191459B (en) | Automatic identification and rating method for continuous casting billet macrostructure center segregation defect | |
| CN104298993A (en) | Pole number positioning and identification method suitable for railway line in complex scene | |
| CN102289676B (en) | Method for identifying mode of switch of substation based on infrared detection | |
| CN112434695B (en) | Upper pull rod fault detection method based on deep learning | |
| CN110310255B (en) | Point switch notch detection method based on target detection and image processing | |
| CN109118479A (en) | Defects of insulator identification positioning device and method based on capsule network | |
| CN112101182B (en) | Railway wagon floor damage fault identification method based on improved SLIC method | |
| CN104268505A (en) | Automatic cloth defect point detection and recognition device and method based on machine vision | |
| CN109785301B (en) | Rail corrugation period evaluation method based on image processing | |
| CN103234976A (en) | Warp knitting machine cloth flaw on-line visual inspection method based on Gabor transformation | |
| CN102542289A (en) | Pedestrian volume statistical method based on plurality of Gaussian counting models | |
| CN109946304A (en) | Surface defects of parts on-line detecting system and detection method based on characteristic matching | |
| CN111079734B (en) | Method for detecting foreign matters in triangular holes of railway wagon | |
| CN102073846A (en) | Method for acquiring traffic information based on aerial images | |
| Gui et al. | Automated defect detection and visualization for the robotic airport runway inspection | |
| CN107967681A (en) | Defect inspection method is hindered in a kind of elevator compensation chain punching based on machine vision | |
| CN109190624A (en) | Kitchen fume concentration detection method based on image procossing | |
| CN109447063A (en) | A kind of kitchen fume concentration detection method based on image procossing | |
| CN104537651A (en) | Proportion detecting algorithm and system for cracks in road surface image | |
| CN109540925A (en) | Complicated ceramic tile surface defect inspection method based on difference shadow method and local variance measurement operator | |
| CN109740609A (en) | A kind of gauge detection method and device | |
| CN109781737A (en) | A kind of detection method and its detection system of hose surface defect | |
| CN114926410A (en) | Method for detecting appearance defects of brake disc | |
| CN101013126A (en) | Method for predicting intensity of wooden artificial panel by computer | |
| CN112198170A (en) | Detection method for identifying water drops in three-dimensional detection of outer surface of seamless steel pipe |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20181109 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |