CN107180416A - Train wheel tread image deformity correction method and system - Google Patents
Train wheel tread image deformity correction method and system Download PDFInfo
- Publication number
- CN107180416A CN107180416A CN201710349621.4A CN201710349621A CN107180416A CN 107180416 A CN107180416 A CN 107180416A CN 201710349621 A CN201710349621 A CN 201710349621A CN 107180416 A CN107180416 A CN 107180416A
- Authority
- CN
- China
- Prior art keywords
- image
- tread
- wheel
- region
- row
- 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 112
- 238000012937 correction Methods 0.000 title claims abstract description 79
- 238000000638 solvent extraction Methods 0.000 claims abstract description 19
- 230000008569 process Effects 0.000 claims abstract description 18
- 230000000007 visual effect Effects 0.000 claims abstract description 17
- 230000000694 effects Effects 0.000 claims abstract description 13
- 238000004422 calculation algorithm Methods 0.000 claims description 24
- 230000011218 segmentation Effects 0.000 claims description 24
- 238000012545 processing Methods 0.000 claims description 20
- 238000012952 Resampling Methods 0.000 claims description 16
- 239000011159 matrix material Substances 0.000 claims description 16
- 238000002372 labelling Methods 0.000 claims description 15
- 239000000284 extract Substances 0.000 claims description 14
- 238000012935 Averaging Methods 0.000 claims description 7
- 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 7
- 238000013507 mapping Methods 0.000 claims description 7
- 238000005530 etching Methods 0.000 claims description 6
- 230000000717 retained effect Effects 0.000 claims description 6
- 239000003550 marker Substances 0.000 claims description 5
- 239000000203 mixture Substances 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 4
- 238000006243 chemical reaction Methods 0.000 claims description 2
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims description 2
- 238000003672 processing method Methods 0.000 abstract description 2
- 239000004575 stone Substances 0.000 abstract description 2
- 238000005259 measurement Methods 0.000 abstract 1
- 239000000306 component Substances 0.000 description 48
- 238000001514 detection method Methods 0.000 description 17
- 230000007547 defect Effects 0.000 description 10
- 238000004458 analytical method Methods 0.000 description 9
- 238000002474 experimental method Methods 0.000 description 6
- 238000000605 extraction Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 238000013461 design Methods 0.000 description 4
- 230000009466 transformation Effects 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 238000003709 image segmentation Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 2
- 238000005286 illumination Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 239000000523 sample Substances 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000009966 trimming Methods 0.000 description 2
- 208000034656 Contusions Diseases 0.000 description 1
- 241001504664 Crossocheilus latius Species 0.000 description 1
- 229910000831 Steel Inorganic materials 0.000 description 1
- 206010052428 Wound Diseases 0.000 description 1
- 208000027418 Wounds and injury Diseases 0.000 description 1
- 238000009825 accumulation Methods 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 208000034526 bruise Diseases 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 238000004040 coloring Methods 0.000 description 1
- 230000001447 compensatory effect Effects 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 239000008358 core component Substances 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 238000003708 edge detection Methods 0.000 description 1
- 238000011010 flushing procedure Methods 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 238000003702 image correction Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000003902 lesion Effects 0.000 description 1
- 238000011068 loading method Methods 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012827 research and development Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000005096 rolling process Methods 0.000 description 1
- 239000010959 steel Substances 0.000 description 1
- 238000000844 transformation Methods 0.000 description 1
- 230000001960 triggered effect Effects 0.000 description 1
Classifications
-
- G06T5/80—
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/20—Image enhancement or restoration by the use of local operators
- G06T5/30—Erosion or dilatation, e.g. thinning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30236—Traffic on road, railway or crossing
Abstract
The invention discloses a kind of train wheel tread image deformity correction method and system, it is related to image processing method technical field.Methods described marks watershed initial partitioning side view visual angle tread surface chart picture using improved coloured image first, then the geometric correction of plane transverse and longitudinal coordinate is sequentially carried out as the different characteristics of horizontal and vertical distortion according to tread surface chart and completes the secondary splitting and visual angle effect in tread region in the process, finally obtain tread front viewing angle correction chart picture.Methods described can accurately be split and correct side view visual angle tread image, to realize that the accurate measurement of tread fault zone lays the first stone.
Description
Technical field
The present invention relates to image processing method technical field, more particularly to a kind of train wheel tread image deformity correction side
Method and system.
Background technology
Develop with the high speed and heavy loading of railway transportation, safety of railway traffic problem becomes increasingly conspicuous.Wheel is to being railway
One of core component of haulage vehicle, wheel tread directly takes on vehicle as the direct contact portion between vehicle and rail
All active forces between rail, working environment is severe, therefore wheel tread defect becomes wheel to one of major failure.Wheel pair
Tread failure can influence Wheel Rail Contact state, and periodic impact, other parts and track road to bogie are caused to car body
Base causes damage, and strong vibration can be to transporting passenger-cargo have adverse effect on.If can not find in time, over time
Accumulation can cause very big potential safety hazard, thus it is significant to guarantee driving safety to strengthen wheel tread state-detection.
Domestic and foreign scholars are conducted in-depth research to wheel tread fault detection technique.The MERMEC companies research and development in the U.S.
W-INSPECT systems obtain the image of the whole tread of wheel and to obtaining using high resolution camera and specific compensatory light
Image analyzed, differentiate scratch and the failure such as stripping of wheel tread, detection speed can reach 30Km/h；Germany
The Argus_II systems of Hegenscheidt-MFD companies to the wheel of contact by two ultrasonic probes by sending Rayleigh face
The ultrasonic pulse of waveshape simultaneously detects that the mode of response of cycle signal obtains wheel information, and utilize certain standard scores
Analysis signal amplitude carries out judging tread failure, and its Detection results is fine, but equipment cost is very high；The tread of Southwest Jiaotong University is damaged
It is to be based on optical bruise detection method to hinder on-line detecting system, and its mentality of designing is necessarily caused when wheel passes through from rail
The deformation of rail, deformation caused by normal wheels is constant, and the wheel that there is thread defect passes through what is necessarily deformed upon
Change, therefore can judge whether wheel tread is defective by the deformation of one section of rail of optical sensor analysis, although its
Threshold decision situation that is larger and being only capable of providing a thread defect is influenceed with degree of precision, but by rail vibration, and can not
Return to definite defect condition；The online tread hurt detecting system of the Tian Lili designs of Northwestern Polytechnical University utilizes 3 groups of faces
Array camera shoots the wheel passed through, realizes the segmentation to each tread image and damage field positioning；Nanjing Aero-Space are big
Sun Ran is detected using three groups of line-scan digital cameras to wheel tread image, and tread image light is avoided compared with area array cameras
According to uneven situation.
Tread fault type is complicated, and simple sound and vibration signal threshold decision can not obtain enough tread states
Information, therefore this full tread infomation detection mode of image detection becomes current main flow.Due to image capture device position
The limitation of setting can not realize the extraction of tread front viewing angle image, often direct root after conventional method side view visual angle collection image
Failure qualitative judgement is carried out according to tread surface chart picture, such method does not consider IMAQ orientation and tread relative position first
Change the influence to imaging, it is often more important that the tread curved surface of extraction carries Severe distortion, it is impossible to which defect part area is carried out
Quantitative calculating, Accurate Analysis.
The content of the invention
The technical problems to be solved by the invention are how to provide one kind can accurately split and correct side view visual angle tread
The image deformity correction method and system of image, are that the accurate calculating of thread defect area lays the first stone.To solve above-mentioned technology
Problem, the technical solution used in the present invention is：A kind of train wheel tread image deformity correction method, it is characterised in that including
Following steps：
Gather the wheel tread image of railway transport vehicle under motion state；
The RGB information that processing is used to recover tread image is carried out to colored tread image using MSRCR algorithms；
Each connected domain crestal line figure of image is obtained using improved coloured image Based On Method of Labeling Watershed Algorithm；
According to collection tread feature of image, find the most connected domain of pixel in image and tread image is divided for the first time
Cut, extract tread curved surface；
According to tread surface chart as the different characteristics of horizontal and vertical distortion sequentially carries out plane horizontal stroke, ordinate geometric correction
And the secondary splitting and visual angle effect in tread region are completed in the process；
Export tread front viewing angle correction chart picture.
Further technical scheme is：The wheel tread image of railway transport vehicle under described collection motion state
Method is as follows：
Several cameras composition camera array is fixed in track external position, and sets identical at equal intervals in the inner side of track
Quantity proximity transducer, every camera one proximity transducer of correspondence, wheel often just triggers correspondence by a proximity transducer
Camera once photo taking is carried out to wheel, the same wheel diverse location image of camera array collection is believed image comprising complete wheel
Breath.
Further technical scheme is：It is each that the improved coloured image Based On Method of Labeling Watershed Algorithm of described use obtains image
The method of connected domain crestal line figure is as follows：
The coloured image gradient map of tread is obtained, morphology opening and closing reconstruct is carried out；
Maximum to image RGB color component is taken or computing, is retained the very big value information of each chrominance component and is obtained
The processing of first opening operation post-etching is carried out after the maximum of region, prospect mark is completed；
Mean operation, OST Threshold segmentations and watershed transform are carried out to image RGB color component, background mark is obtained
Expansion process is carried out using square masterplate complete context marker afterwards；
Amended coloured image gradient map is obtained by minimum scaling method；
Each connected domain crestal line figure of image is obtained after carrying out watershed transform to amended coloured image gradient map；According to adopting
Collect tread feature of image, find the most connected domain of pixel in image and initial partitioning is carried out to tread image, extract tread
Curved surface.
Further technical scheme is that described abscissa geometric correction method is as follows：
First with the gray scale feature of tread curved surface, often the maximum of first half is calculated in each row in row non-zero pixels
Wheel rim side and the point of interface position of tread, wheel rim and tread point of interface position are calculated using the method for averaging；
It is tight that the tread portions that every row is intercepted out from tread area image using two point of interfaces of image R component remove distortion
The wheel rim rim region of weight；
The tread region split by R component carries out dividing processing to G components and B component matrix；
The tread region of each row of each color component is sampled into specified width, which width using arest neighbors method for resampling；
The vector reconstruction of each row obtained using each component resampling goes out tread image.
Further technical scheme is that described ordinate geometric correction method is as follows：
Tread image after being corrected using the mathematical modeling calculated to abscissa carries out Nonlinear Mapping in y direction
Reconstruct；
Bilinear interpolation is carried out to the image ordinate after reconstruct and obtains wide high proportion and actual ratio identical tread figure
Picture.
The invention also discloses a kind of train wheel tread image deformity correction system, it is characterised in that including：
Image capture module, the wheel tread image for gathering railway transport vehicle under motion state；
Image pre-processing module：Colored tread image is handled using MSRCR algorithms, for recovering tread image
RGB information；
Image initial partitioning module：For obtaining each connected domain of image using improved coloured image Based On Method of Labeling Watershed Algorithm
Crestal line figure is simultaneously divided tread image for the first time according to the most connected domain of pixel in tread feature of image, searching image is gathered
Cut, extract tread curved surface；
Image rectification and secondary splitting module：For according to tread surface chart as the different characteristics of horizontal and vertical distortion is suitable
It is secondary to carry out horizontal plane, ordinate geometric correction and the secondary splitting and visual angle effect that complete tread region in the process；
Image output module：For exporting tread front viewing angle correction chart picture.
Further technical scheme is：Described image acquisition module is included in track external position and fixes several cameras
Camera array is constituted, and identical quantity proximity transducer is set in the inner side of track at equal intervals, every camera correspondence one is approached
Sensor, wheel often just triggers corresponding camera by a proximity transducer and carries out once photo taking to wheel, and camera array is adopted
The same wheel diverse location image of collection is included and completely taken turns to image information.
Further technical scheme is：Described image initial partitioning module is used for the coloured image gradient for obtaining tread
Figure, carries out morphology opening and closing reconstruct；
Maximum to image RGB color component is taken or computing, is retained the very big value information of each chrominance component and is obtained
The processing of first opening operation post-etching is carried out after the maximum of region, prospect mark is completed；Average fortune is carried out to image RGB color component
Calculation, OST Threshold segmentations and watershed transform, are obtained after background mark using square masterplate progress expansion process completion background mark
Note；Amended coloured image gradient map is obtained by minimum scaling method；Amended coloured image gradient map is carried out
Each connected domain crestal line figure of image is obtained after watershed transform；According to collection tread feature of image, pixel in image is found most
Connected domain to tread image carry out initial partitioning, extract tread curved surface.
Further technical scheme is：Described image is corrected and secondary splitting module includes abscissa geometric correction module
And ordinate geometric correction module；The abscissa geometric correction module is often gone non-first with the gray scale feature of tread curved surface
The maximum of first half calculates the point of interface position of the side of wheel rim in each row and tread in zero pixel, utilizes the side of averaging
Method calculates wheel rim and tread point of interface position；Every row is intercepted out from tread area image using two point of interfaces of image R component
Tread portions remove the serious wheel rim rim region of distortion；The tread region split by R component is to G components and B component square
Battle array carries out dividing processing；The tread region of each row of each color component is sampled into using arest neighbors method for resampling and specifies width
Degree；The vector reconstruction of each row obtained using each component resampling goes out tread image.
Further technical scheme is：Described image is corrected and secondary splitting module includes abscissa geometric correction module
And ordinate geometric correction module；The ordinate geometric correction module is corrected using the mathematical modeling calculated to abscissa
Tread image afterwards carries out Nonlinear Mapping reconstruct in y direction；Bilinear interpolation is carried out to the image ordinate after reconstruct to obtain
To wide high proportion and actual ratio identical tread image.
It is using the beneficial effect produced by above-mentioned technical proposal：Methods described is using improved coloured image mark point
Water ridge initial partitioning side view visual angle tread surface chart picture, improves the segmentation precision of image, then horizontal according to tread surface chart picture
The geometric correction of plane transverse and longitudinal coordinate is sequentially carried out to the different characteristics with longitudinal distortion and tread region is completed in the process
Secondary splitting and visual angle effect, finally obtain tread front viewing angle correction chart picture, changing traditional detection method can only be according to side
Whether have the shortcomings that to damage depending on multi-view image analysis tread and lesion shape and size can not be determined, relatively directly perceived and accurate analysis
Go out the size of damage field, judge to provide value foundation for station inspector fault level.Experiments verify that, this method have compared with
Strong antijamming capability simultaneously achieves preferable accuracy of detection.
Brief description of the drawings
Fig. 1 is the flow chart of methods described of the embodiment of the present invention；
Fig. 2 be in methods described of the embodiment of the present invention under haze reason to taking turns the image to being acquired；
Fig. 3 is the result figure pre-processed to haze reason lower whorl to image in methods described of the embodiment of the present invention；
Fig. 4 be in methods described of the embodiment of the present invention under half-light reason to taking turns the image to being acquired；
Fig. 5 is the result figure pre-processed to half-light reason lower whorl to image in methods described of the embodiment of the present invention；
Fig. 6 is improved coloured image Based On Method of Labeling Watershed Algorithm flow chart in methods described of the embodiment of the present invention；
Fig. 7 is watershed crestal line figure in methods described of the embodiment of the present invention；
Fig. 8 is prospect mark and background mark stacking chart in methods described of the embodiment of the present invention；
Fig. 9 is colour-coded connected domain design sketch in methods described of the embodiment of the present invention；
Figure 10 is methods described of embodiment of the present invention acceptance of the bid note cut zone superposition artwork；
Figure 11 is tread image rectification flow chart in methods described of the embodiment of the present invention；
Figure 12 is the direct linear transformations figure in tread region in methods described of the embodiment of the present invention；
Figure 13 is wheel rim lateral width change schematic diagram in methods described of the embodiment of the present invention；
Figure 14 is the intensity map of red component in methods described of the embodiment of the present invention；
Figure 15 is the brightness distribution curve figure in red component single file in methods described of the embodiment of the present invention；
Figure 16 is the approximate projection model figure after abscissa correction in methods described of the embodiment of the present invention；
Figure 17 is ordinate correction principle schematic diagram in methods described of the embodiment of the present invention；
Figure 18 is the theory diagram of system described in the embodiment of the present invention；
Figure 19 is tread segmentation result figure in the embodiment of the present invention；
Figure 20 is abscissa correction result figure in the embodiment of the present invention；
Figure 21 is result figure after ordinate correction in the embodiment of the present invention；
Figure 22 is reference pair corner location schematic diagram in the embodiment of the present invention；
Figure 23 is that medium square width of the embodiment of the present invention is compared figure with actual value；
Figure 24 is that medium square of embodiment of the present invention height is compared figure with actual value；
Figure 25 is side view tread image in the embodiment of the present invention；
Figure 26 is front viewing angle tread image in the embodiment of the present invention；
Figure 27 is the tread image after being corrected by methods described in the embodiment of the present invention；
Figure 28 is each connected domain proportion figure in methods described of the embodiment of the present invention；
Figure 29 is template image in methods described of the embodiment of the present invention；
Figure 30 is the result figure for handling artwork in methods described of the embodiment of the present invention using template.
Embodiment
With reference to the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete
Ground is described, it is clear that described embodiment is only a part of embodiment of the present invention, rather than whole embodiments.It is based on
Embodiment in the present invention, it is every other that those of ordinary skill in the art are obtained under the premise of creative work is not made
Embodiment, belongs to the scope of protection of the invention.
Many details are elaborated in the following description to facilitate a thorough understanding of the present invention, still the present invention can be with
It is different from other manner described here using other to implement, those skilled in the art can be without prejudice to intension of the present invention
In the case of do similar popularization, therefore the present invention is not limited by following public specific embodiment.
As shown in figure 1, the embodiment of the invention discloses a kind of train wheel tread image deformity correction method and system, bag
Include following steps：
S101：Gather the wheel tread image of railway transport vehicle under motion state；
S102：The RGB letter that processing is used to recover tread image is carried out to colored tread image using MSRCR algorithms
Breath；
S103：Each connected domain crestal line figure of image is obtained using improved coloured image Based On Method of Labeling Watershed Algorithm.According to collection
Tread feature of image, finds the most connected domain of pixel in image and carries out initial partitioning to tread image, extract tread bent
Face；
S104：According to tread surface chart as the different characteristics of horizontal and vertical distortion sequentially carries out that plane is horizontal, ordinate is several
What corrects and completes the secondary splitting and visual angle effect in tread region in the process；
S105：Export tread front viewing angle correction chart picture.
The characteristics of due to bogies for railway vehicles Machine Design, wheel tread can not be intactly exposed to extraneous visual range
It is interior, while because the requirement of railway security gauge can not complete its front viewing angle IMAQ, thus tread during dynamic detection
Fixing multi-section camera using track side locations more than image detecting method coordinates wheel detector sequentially to gather tread image.
Specifically, in the embodiment of the present invention, the wheel tread image of railway transport vehicle under described collection motion state
Method it is as follows：Several cameras composition camera array is fixed in track external position, and is set at equal intervals in the inner side of track
Identical quantity proximity transducer, every camera one proximity transducer of correspondence, wheel is often just triggered by a proximity transducer
Corresponding camera carries out once photo taking to wheel, and the same wheel diverse location image of camera array collection is taken turns to figure comprising complete
As information.It should be noted that the camera is generally high-speed industrial camera, continuously taking pictures in the short time can be completed.
Often disturbed in image acquisition process by external environment, to follow-up image segmentation and target identification etc.
Reason is impacted.As shown in Fig. 2 the picture contrast and brightness that haze sky is collected are low, color occurs offseting and distortion so that
Image detail thickens.In addition, being in as shown in figure 4, typically having single wheel in bogie at backlight, its color difference
It is smaller, influence picture quality.For both the above common situations, the embodiment of the present invention is using MSRCR algorithms to colored tread figure
Recover the RGB information of image as carry out processing, effect is as shown in Figure 3 and Figure 5.
Examine treatment effect as shown in table 1 using each evaluation index image before and after the processing：
Each picture appraisal standard analysis statistical form of table 1 (edge strength uses Sobel edge detection operators)
Data comparison understands that picture quality is significantly improved after pre-processing from table 1, particularly image detail and color contrast
Degree has very big improvement, is that successive image segmentation is got ready.
Each connected domain crestal line figure of image is obtained and according to collection tread using improved coloured image Based On Method of Labeling Watershed Algorithm
Feature of image, finds the most connected domain of pixel in image and carries out initial partitioning to tread image, extract tread curved surface；
Different from the Based On Method of Labeling Watershed Algorithm in gray scale domain, the method for the invention directly calculates coloured image gradient map, makes
The direct feel of the more accurate and closer human eye of obtained coloured image gradient.The improved Based On Method of Labeling Watershed Algorithm of the present invention is preceding
Scape carries out logic or computing during marking by the maximum figure of 3 chrominance components, and each chrominance component is retained as far as possible
During very big value information obtains region maximum, context marker, three color components of image are subjected to mean value computation and come
Reduce influence of the different colours to mark result between object.Algorithm idiographic flow is as shown in fig. 6, obtain each step design sketch such as figure
Shown in 7-10.
In Fig. 6, the flow is：
S1031：The coloured image gradient map of tread is obtained, morphology opening and closing reconstruct is carried out；
S1032：Maximum to image RGB color component is taken or computing, retains each chrominance component maximum letter
Breath obtains carrying out the processing of first opening operation post-etching after the maximum of region, completes prospect mark；
S1033：Mean operation, OST Threshold segmentations and watershed transform are carried out to image RGB color component, carried on the back
Expansion process is carried out using square masterplate complete context marker after scape mark；
S1034：Amended coloured image gradient map is obtained by minimum scaling method；
S1035：Each connected domain crestal line figure of image is obtained after carrying out watershed transform to amended coloured image gradient map；
According to collection tread feature of image, find the most connected domain of pixel in image and initial partitioning is carried out to tread image, extract
Go out tread curved surface.
The specific method that extraction tread region after crestal line figure is obtained in watershed is as follows：
(1) each connected domain is marked.Each connection produced using labeling function to the watershed transform of previous step
Domain is marked, return label matrix L and corresponding number of labels n.
(2) most connected domains of counting are found.Region interested is found in all connected domains and needs first to analyze and feels emerging
The exclusive feature in interesting region.In the present invention, it is seen that the area in tread region is significantly larger than other from experimental result picture
By asking the most labels of points to find the region corresponding to tread region in region, therefore the present invention.It is first during practical operation
First use the points of the corresponding connected domain of n label of statistics with histogram function pair to be counted, then find out points wherein
Most label Lm.There are 46 connected domains in the crestal line figure obtained in an experiment after mark watershed processing, Figure 28 is each region
The histogram of proportion in the picture, the 29th region is tread region in figure.As can be seen from the figure tread region exists
The ratio accounted in image is significantly larger than other regions, it is possible to found by finding the most connected domain of pixel in image
Tread region.
(3) the corresponding regions of label Lm are found out.The correspondence connection of Lm labels is determined in mark matrix L using search function
Domain coordinate a little.
(4) artwork is handled using connected domain Lm.The coordinate of the point obtained using previous step creates a Lm area
Domain is the template that 1 remaining region is 0, then just can be with using each color component matrix in the pattern matrix dot product artwork created
Finally give single tread area image.Fig. 9 is the tread area image extracted using the above method.
From Figure 29-Figure 30 as can be seen that the method success designed using the present invention from watershed transform to crestal line
Tread region has been isolated in figure, and the cromogram in tread region is successfully extracted from artwork using the tread region of extraction
Picture.
Because the illumination in tread area image and distribution of color complexity are difficult to obtain stable segmentation result, at actual place
In reason, the selection of parameter and template size in the opening and closing operation based on reconstruction in Based On Method of Labeling Watershed Algorithm in algorithm for image enhancement
Segmentation result will be impacted, it is difficult to obtain single tread image.As shown in Fig. 4-14, the tread region being partitioned into
Although relatively more neat, the part of rim section region and wheel rim side is further comprises in addition to comprising tread.Therefore, entering
Also need to simply split the image-region thoroughly removed outside tread to image after line flag watershed processing.
Because the tread image of extraction for curved surface and camera is shot from side, the tread area image obtained is caused to exist
Serious geometric distortion, it is difficult to set up accurate mathematical modeling and quantitatively calculated thread defect area.By analyzing image
The reason for distortion, understands that lateral aberration is mainly linear distortion caused by camera perspective, and longitudinal distortion is mainly wheel shape
Caused nonlinear distortion, it is separate between the two, therefore as illustrated in flow chart figure 11, can be by the geometric correction in tread region
It is divided into abscissa correction and ordinate corrects two steps and carried out.Comprise the following steps that：
S1041：First with the gray scale feature of tread curved surface, often the maximum of first half is calculated in row non-zero pixels
Wheel rim side and the point of interface position of tread, wheel rim and tread point of interface position are calculated using the method for averaging in each row；
S1042：The tread portions that every row is intercepted out from tread area image using two point of interfaces of image R component are removed
The serious wheel rim rim region of distortion；
S1043：The tread region split by R component carries out dividing processing to G components and B component matrix；
S1044：The tread region of each row of each color component is sampled into specified width, which width using arest neighbors method for resampling；
S1045：The vector reconstruction of each row obtained using each component resampling goes out tread image.
S1046：Tread image after being corrected using the mathematical modeling calculated to abscissa carries out non-thread in y direction
Property mapping reconstruction；
S1047：Bilinear interpolation is carried out to the image ordinate after reconstruct and obtains wide high proportion and actual ratio identical
Tread image.
Tread image abscissa is corrected：
According to the calibration principle of linear distortion, it is only necessary to by all lines of tread area image be drawn into it is wide can
To recover the tread image of script.But if directly carrying out stretching obtained result such as Figure 12 institutes to the tread region being partitioned into
Show, it can be seen that image still suffers from serious distortion after stretching, this is due to wheel rim lateral side regions in whole tread peak width
Not consistent, as shown in figure 13, rim width often capable corresponds to the horizontal transversal of wheel rim side annulus in tread area image
Width, and horizontal transversal all in annulus crosses most short at the CD in the center of circle, gradually increases to two ends mobile width at CD.With
This is not on same plane, wheel rim, wheel rim area in the image after conversion on the inside of wheel rim side and wheel rim with tread simultaneously
The ratio in domain and tread region differs greatly with actual ratio.Due to wheel rim side and wheel rim inside region, not tread is wiped
The generation position of the main damages such as wound, stripping, stone roller heap, therefore must go to before Linear Transformation in often being gone except tread area image
Wheel rim side and wheel rim inside region, i.e., to watershed algorithm split after image progress secondary splitting.
If carrying out row linear transformation using 3 color components, the width of transformation results might have nuance, cause
Later stage recovers mistake occur during coloured image.Thus only enter every trade segmentation to red component, then split using red component
The tread portions arrived carry out dividing processing to other Component Matrices, and the coloured image of tread is finally recombinated using all components.Step on
Flushing colouring component Luminance Distribution feature, as shown in figure 14, the ledge that red " ridge " is partly located at wheel is exactly tread
With its brightness ratio peripheral region of the intersection of wheel rim will height, and the female parts that the intersection of wheel rim and tread is located at wheel are yellow
" the lowest point " of color and blueness, its brightness ratio surrounding pixel is low.It can be considered to the gray scale feature using two intersections come
Split tread area image.
As shown in figure 15, the intersection of tread and wheel rim is located at the first half maximum of points in often going, tread and wheel rim
Junction be then located at often near the minimum value of row latter half.Thus can be by seeking first half in every row non-zero pixels
Maximum calculates the point of interface position of the side of wheel rim in each row and tread.Due to wheel rim and tread boundary line substantially and
The width overall variation very little of wheel rim in the picture, the point of interface for directly calculating wheel rim and tread with second half section minimum value can be caused
Very big error, therefore calculate wheel rim and tread point of interface position using averaging method, i.e., every row second half section is calculated first most
Small value distance and the distance at the row end, then calculate being averaged for all row distances and are worth to the approximate width of wheel rim, finally will
Wheel rim point of interface approximate with tread is used as per the point of row distance end of line rim width.Using two point of interfaces obtained from tread
Split often capable tread portions in area image.After wheel rim side and wheel rim inside region is removed, due to varying level line
The width in upper tread region is different, therefore by tread region, resampling is specified width, which width to the present invention line by line, and so neither influence is schemed
As texture, the calculating time is reduced again.Finally, three color components are reassembled as coloured image with regard to that can obtain after abscissa correction
Image.
Tread image ordinate is corrected
The front viewing angle image of camera can approximately be regarded as by completing rear tread pattern image in abscissa correction, as shown in figure 16,
Circle center distance from wheel when shooting is 2.828m, and tread perspective plane is highly 0.727m, it can thus be concluded that tread image midpoint phase
The light path of machine is θ with the maximum angle of camera axis, and shooting image angle, θ is smaller at other positions farther out, therefore is being calculated
Light path can be treated simply as parallel to camera axis during ordinate calibration model.
When setting up the mathematical modeling of ordinate correction as, tread region in image is approximately taken to 120 ° of tread circular arc area
Domain, and picture centre line take as it is concordant with the wheel center of circle, as shown in figure 17.
The nonlinear distortion problem of ordinate can be regarded as point on arc BD according to string BD based on two above condition
What equal interval sampling was caused, thus the Correction Problemss of ordinate can be converted into asking circular arc BD point sampling value is asked at equal intervals
Topic.The point C on point A correspondence strings BD on arc BD as shown in figure 17.If LAOC=θ, radius BO=R, then arc AB length is R θ,
String AB length is 2Rsin (θ/2), and line segment BC length is y, then can obtain
Thus the corresponding relation of point and the point on string BD on arc BD is obtained, specific aligning step is as follows：
(1) matrix that the image after abscissa correction is m*n is assumed, then when calculating beginning according to the ratio of arc length and chord length
It is 120 ° that example, which defines central angle in a correspondingly sized matrix, the present invention, it is therefore desirable to define 1.21m*n matrix.
(2) all coordinates of the matrix created in previous step are mapped to horizontal seat according to formula (1) using arest neighbors interpolation method
In matrix after calibration just.
(3) in order that the ratio of the transverse and longitudinal coordinate of image after correction is identical with actual ratio, by the ordinate of matrix use
The method of bilinear interpolation is by matrix tensile into designated length.
By being corrected to tread image horizontal stroke, ordinate, the correction of tread region color image coordinate is completed.
As shown in figure 18, the embodiment of the invention also discloses a kind of train wheel tread image deformity correction system, including：
Image capture module 101, the wheel tread image for gathering railway transport vehicle under motion state；
Image pre-processing module 102：Colored tread image is handled using MSRCR algorithms, for recovering tread
The RGB information of image；
Image initial partitioning module 103：Each connected domain of image is obtained using improved coloured image Based On Method of Labeling Watershed Algorithm
Crestal line figure.According to collection tread feature of image, find the most connected domain of pixel in image and tread image is divided for the first time
Cut, extract tread curved surface；
Image rectification and secondary splitting module 104：For according to difference spy of the tread surface chart as horizontal and vertical distortion
Point sequentially carries out horizontal plane, ordinate geometric correction and the secondary splitting and visual angle effect that complete tread region in the process；
Image output module 105：For exporting tread front viewing angle correction chart picture.
Further, as shown in figure 18, described image acquisition module 101 is included in several of track external position fixation
Several proximity transducers 1012 set at equal intervals on the inside of camera 1011 and track, every close sensing of camera correspondence one
Device, wheel often just triggers corresponding camera by a proximity transducer and once photo taking is carried out to wheel, camera array collection
Same wheel diverse location image is included and completely taken turns to image information.
Further, described image initial partitioning module 103 is used for the coloured image gradient map for obtaining tread, carries out form
Learn opening and closing reconstruct；Maximum to image RGB color component is taken or computing, is retained the very big value information of each chrominance component and is obtained
The processing of first opening operation post-etching is carried out after to region maximum, prospect mark is completed；Average is carried out to image RGB color component
Computing, OST Threshold segmentations and watershed transform, are obtained after background mark using square masterplate progress expansion process completion background
Mark；Amended coloured image gradient map is obtained by minimum scaling method；Amended coloured image gradient map is entered
Each connected domain crestal line figure of image is obtained after row watershed transform；According to collection tread feature of image, find image in pixel most
Many connected domains carry out initial partitioning to tread image, extract tread curved surface.
Further, described image correction and secondary splitting module 104 include abscissa geometric correction module 1041 and
Ordinate geometric correction module 1042；The abscissa geometric correction module 1041 is every first with the gray scale feature of tread curved surface
The maximum of first half calculates the point of interface position of the side of wheel rim in each row and tread in row non-zero pixels, utilizes averaging
Method calculate wheel rim and tread point of interface position.Intercepted out using two point of interfaces of image R component from tread area image
Often capable tread portions remove the serious wheel rim rim region of distortion；The tread region split by R component is to G components and B points
Moment matrix carries out dividing processing；The tread region of each row of each color component is sampled into using arest neighbors method for resampling specified
Width；The vector reconstruction of each row obtained using each component resampling goes out tread image.
The ordinate geometric correction module 1042 utilizes stepping on after being corrected using the mathematical modeling calculated to abscissa
Face image carries out Nonlinear Mapping reconstruct in y direction；Bilinear interpolation is carried out to the image ordinate after reconstruct and obtains wide height
Ratio and actual ratio identical tread image.
For the accuracy of checking this method segmentation correction tread, 34mm*34mm square nets are pasted at wheel tread
Paper is demarcated as the yardstick evaluation criterion of image detection result.Because demarcation paper color and tread background difference are very big, therefore meeting
Normal segmentation of the Based On Method of Labeling Watershed Algorithm to tread is had influence on, so the surface chart that the initial partitioning used in confirmatory experiment is obtained
Seem with demarcation paper wheel tread region without the wheel tread contours segmentation same position collection obtained in the case of demarcation paper
Image, as shown in figure 19.
What segmentation was extracted is approximate 120 ° of tread region in tread image, and tread is approximately regarded as behind the face of cylinder to count
Calculation extracts tread actually corresponding rectangular region.Wherein 1/3rd of tread rolling circle girth are 880mm, freight car wheel
The width in tread region is 103mm.The width of the tread region extracted in confirmatory experiment in the picture is all on 100 pixel left sides
The right side, therefore in order to ensure image low distortion also for the convenient calculating to damaging real area in tread region, final choice
By the picture that the resampling of tread area image is 880*103 sizes, that is, 1 pixel distance represents 1mm distances in image.
In order to not allow squared paper to have influence on the extraction to two boundary lines, wheel rim and wheel rim in secondary splitting removes image
During region, the computation interval in squared paper region is adjusted, grid spaces are avoided, remaining operation is with extracting normal horizontal seat
Mark trimming process identical, most the image resampling in tread region is 103 row at last.Image after abscissa correction, such as Figure 20 institutes
Show, it is known that the width of the black box of squared paper is roughly the same, is consistent with expected results.Simultaneously it is also seen that in ordinate side
To grid height and the width on abscissa direction and differ, and the height of each grid is also different.
The trimming process of ordinate is pure mathematics computing, therefore is not influenceed by squared paper, calculating process and normal wheels
When it is identical.Figure 20 is subjected to Nonlinear Mapping first with formula (1), Figure 20 pixel is 480* in this experiment
103, it is therefore desirable in the 581*103 matrixes for being first mapped to it.The tread area image pixel specified is 880*103, therefore
Also need to carry out resampling on the vertical scale to image.If Figure 21 is the image after final ordinate is corrected.
The present invention uses the size of black and white grid in demarcation paper to verify the result of geometric correction as object of reference.Specific
It is the method subtracted each other using the diagonal point coordinates of grid when calculating grid size.Point where circle as shown in figure 22 is exactly to use
To calculate the reference angle steel joint of grid size, the grid that the grid of blue circle mark is used when being calculation error.It is artificial to read
Coordinate point coordinates be respectively [61,117], [93,154], [26,149], [60,184], [93,218], [26,218], [59,
253], [94,288], [26,288], [61,323], [95,355].Thus the grid size respectively 35*32,32* calculated
37,34*35,33*30,34*36,33*35,33*35,34*35,33*35,35*35,35*35,33*35,34*32。
The correction result of grid width is mainly corrected by abscissa to be influenceed, the width and reality of grid after such as Figure 23 corrections
Value is more shown, and the result of abscissa correction at most differs the error of 2 pixels, i.e. abscissa 6% or so with actual value.Grid
Height correction result mainly corrected and influenceed by ordinate.As shown in Figure 19, the intersection of the wheel rim on picture top and tread
Do not identify correctly, the demarcation paper of relevant position also has obvious distortion.As Figure 24 grids are high and actual value ratio
Shown in relatively scheming, the more other grids of height error of top-down preceding 4 grids are much bigger, the mistake of up to 4 pixels
Difference, error is 12% or so.
For haze (PM2.5=260-475 μ g/m3), dim light (100-500lx) and normal illumination (500-2000lx) shape
State, each 10 wheels of collection carry out segmentation correction using the above method to tread image respectively to image.Definition detection segmentation figure picture
Each grid element center point abscissa error of middle scaling board is Hx, and ordinate error is Zx, with actual grid center position error
For Fx, thenAccording to grid element center point location AME in single image from accuracy and stably
Property angle is evaluated collecting sample data error under each ambient condition using each index, evaluation result such as table 2 below
The error analysis table of table 2
| Ambient condition | Detect number of times | Central point AME/mm | A classes uncertainty/mm | B classes uncertainty/mm | Composite Seismogram/mm |
| Normally | 10 | 2.098 | 0.812 | 0.1 | 0.818 |
| Haze | 10 | 2.735 | 0.849 | 0.1 | 0.854 |
| Dim light | 10 | 2.472 | 0.904 | 0.1 | 0.910 |
Tread image detection error is maintained between 2.0-2.8mm under the conditions of the set environment as shown in Table 2, by error not
Degree of certainty analysis system worst error uncertainty is 0.910mm, therefore detects that global error can be maintained within 3.71mm,
Therefore this method can meet detection wheel tread length 40mm defects or peel off the requirement of failure, be flat sliding, stripping, grind
The calculating of the failure area such as heap provides more accurate value reference.
The image segmentation correction accuracy for defining this method is px=P (x)/Po(x), wherein Px represents that manually splitting correction steps on
Of the face image x fault zones pixel point set compared to segmentation curved surface adjustment front viewing angle tread image failure pixel point set position
With degree, P (x) is represented using the correction tread image failure region that method is obtained above and front viewing angle correction tread fault zone
Identical pixel point set, Po(x) the front viewing angle correction tread pixel point set that user's work point segmentation method is obtained is represented.Each visual angle
As illustrated in figs. 25-27, image deflects site analysis result is as shown in table 3 for tread fault picture comparison diagram.Fault picture is examined to exist
Segmentation effect under other detection environment, segmentation precision Px is held in 90% or so, it is possible to achieve to tread accident defect position
The accurate judgement of area, numeric reference is provided for the classification of defect rank.
The image deflects site analysis table of table 3
Methods described marks watershed initial partitioning side view visual angle tread surface chart picture using improved coloured image, improves
The segmentation precision of image, then according to tread surface chart as the different characteristics of horizontal and vertical distortion sequentially carries out plane transverse and longitudinal
Coordinate geometry corrects and completes the secondary splitting and visual angle effect in tread region in the process, finally obtains tread front viewing angle
Correction chart picture, changing traditional detection method can only analyze whether tread has damage and can not determine to damage according to side view multi-view image
Hinder the shortcoming of shapes and sizes, size that is relatively directly perceived and accurately analyzing damage field, is that station inspector fault level judges
There is provided value foundation.Experiments verify that, this method has stronger antijamming capability and achieves preferable accuracy of detection.
Claims (10)
1. a kind of train wheel tread image deformity correction method, it is characterised in that comprise the following steps：
Gather the wheel tread image of railway transport vehicle under motion state；
The RGB information that processing is used to recover tread image is carried out to colored tread image using MSRCR algorithms；
Each connected domain crestal line figure of image is obtained using improved coloured image Based On Method of Labeling Watershed Algorithm, it is special according to collection tread image
Point, finds the most connected domain of pixel in image and carries out initial partitioning to tread image, extract tread curved surface；
According to tread surface chart as the different characteristics of horizontal and vertical distortion sequentially carry out plane horizontal, ordinate geometric correction and
The secondary splitting and visual angle effect in tread region are completed during this；
Export tread front viewing angle correction chart picture.
2. train wheel tread image deformity correction method as claimed in claim 1, it is characterised in that described collection campaign
The method of the wheel tread image of railway transport vehicle is as follows under state：
Several cameras composition camera array is fixed in track external position, and identical quantity is set at equal intervals in the inner side of track
Proximity transducer, every camera one proximity transducer of correspondence, wheel often just triggers corresponding phase by a proximity transducer
Machine carries out once photo taking to wheel, and the same wheel diverse location image of camera array collection, which is included, completely takes turns to image information.
3. train wheel tread image deformity correction method as claimed in claim 1, it is characterised in that described using is improved
Coloured image Based On Method of Labeling Watershed Algorithm obtain each connected domain crestal line figure of image method it is as follows：
The coloured image gradient map of tread is obtained, morphology opening and closing reconstruct is carried out；
Maximum to image RGB color component is taken or computing, is retained the very big value information of each chrominance component and is obtained region
The processing of first opening operation post-etching is carried out after maximum, prospect mark is completed；
Mean operation, OST Threshold segmentations and watershed transform are carried out to image RGB color component, made after obtaining background mark
Expansion process is carried out with square masterplate and completes context marker；
Amended coloured image Grad is obtained by minimum scaling method；
Each connected domain crestal line figure of image is obtained after carrying out watershed transform to amended coloured image Grad；
According to collection tread feature of image, find the most connected domain of pixel in image and initial partitioning carried out to tread image,
Extract tread curved surface.
4. train wheel tread image deformity correction method as claimed in claim 1, it is characterised in that described abscissa is several
What bearing calibration is as follows：
First with the gray scale feature of tread curved surface, often the maximum of first half calculates wheel rim in each row in row non-zero pixels
Side and the point of interface position of tread, wheel rim and tread point of interface position are calculated using the method for averaging；
The tread portions that every row is intercepted out from tread area image using two point of interfaces of image R component, which are removed, to distort seriously
Wheel rim rim region；
The tread region split by R component carries out dividing processing to G components and B component matrix；
The tread region of each row of each color component is sampled into specified width, which width using arest neighbors method for resampling；
The vector reconstruction of each row obtained using each component resampling goes out tread image.
5. train wheel tread image deformity correction method as claimed in claim 1, it is characterised in that described ordinate is several
What bearing calibration is as follows：
Tread image after being corrected using the mathematical modeling calculated to abscissa carries out Nonlinear Mapping reconstruct in y direction；
Bilinear interpolation is carried out to the image ordinate after reconstruct and obtains wide high proportion and actual ratio identical tread image.
6. a kind of train wheel tread image deformity correction system, it is characterised in that including：
Image capture module, the wheel tread image for gathering railway transport vehicle under motion state；
Image pre-processing module：Colored tread image is handled using MSRCR algorithms, for recovering the true of tread image
Colour information；
Image initial partitioning module：For obtaining each connected domain crestal line of image using improved coloured image Based On Method of Labeling Watershed Algorithm
Scheme and according to collection tread feature of image, find the most connected domain of pixel in image and initial partitioning is carried out to tread image,
Extract tread curved surface；
Image rectification and secondary splitting module：For according to tread surface chart as the different characteristics of horizontal and vertical distortion is sequentially entered
Row plane horizontal stroke, ordinate geometric correction and the secondary splitting and visual angle effect that complete tread region in the process；
Image output module：For exporting tread front viewing angle correction chart picture.
7. train wheel tread image deformity correction system as claimed in claim 6, it is characterised in that：Described image gathers mould
Block is included in track external position and fixes several cameras composition camera array, and sets identical number at equal intervals in the inner side of track
Proximity transducer is measured, every camera one proximity transducer of correspondence, wheel often just triggers corresponding by a proximity transducer
Camera carries out once photo taking to wheel, and the same wheel diverse location image of camera array collection is believed image comprising complete wheel
Breath.
8. train wheel tread image deformity correction system as claimed in claim 6, it is characterised in that：Described image is divided for the first time
The coloured image gradient map that module is used to obtain tread is cut, morphology opening and closing reconstruct is carried out；
Maximum to image RGB color component is taken or computing, is retained the very big value information of each chrominance component and is obtained region
The processing of first opening operation post-etching is carried out after maximum, prospect mark is completed；To image RGB color component carry out mean operation,
OST Threshold segmentations and watershed transform, are obtained after background mark using square masterplate progress expansion process completion context marker；
Amended coloured image gradient map is obtained by minimum scaling method；A point water is carried out to amended coloured image gradient map
Each connected domain crestal line figure of image is obtained after the conversion of ridge；According to collection tread feature of image, the company of pixel at most in image is found
Logical domain carries out initial partitioning to tread image, extracts tread curved surface.
9. train wheel tread image deformity correction system as claimed in claim 6, it is characterised in that：Described image correct and
Secondary splitting module includes abscissa geometric correction module and ordinate geometric correction module；The abscissa geometric correction mould
First with the gray scale feature of tread curved surface, often the maximum of first half calculates wheel rim side in each row to block in row non-zero pixels
Face and the point of interface position of tread, wheel rim and tread point of interface position are calculated using the method for averaging；Use image R component two
The tread portions that individual point of interface intercepts out every row from tread area image remove the serious wheel rim rim region of distortion；Pass through R
The tread region of component segmentation carries out dividing processing to G components and B component matrix；By the tread region of each row of each color component
Specified width, which width is sampled into using arest neighbors method for resampling；The vector reconstruction of each row obtained using each component resampling goes out tread
Image.
10. train wheel tread image deformity correction system as claimed in claim 6, it is characterised in that：Described image is corrected
And secondary splitting module includes abscissa geometric correction module and ordinate geometric correction module；The ordinate geometric correction
Tread image after module is corrected using the mathematical modeling calculated to abscissa carries out Nonlinear Mapping reconstruct in y direction；
Bilinear interpolation is carried out to the image ordinate after reconstruct and obtains wide high proportion and actual ratio identical tread image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710349621.4A CN107180416B (en) | 2017-05-17 | 2017-05-17 | Train wheel tread image deformity correction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710349621.4A CN107180416B (en) | 2017-05-17 | 2017-05-17 | Train wheel tread image deformity correction method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107180416A true CN107180416A (en) | 2017-09-19 |
| CN107180416B CN107180416B (en) | 2018-06-05 |
Family
ID=59831786
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710349621.4A Active CN107180416B (en) | 2017-05-17 | 2017-05-17 | Train wheel tread image deformity correction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107180416B (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109668515A (en) * | 2017-10-16 | 2019-04-23 | 苏州华兴致远电子科技有限公司 | Detector for train wheel pair size dynamic detection system and detection method |
| CN109726681A (en) * | 2018-12-29 | 2019-05-07 | 北京航空航天大学 | It is a kind of that location algorithm is identified based on the blind way of machine learning identification and image segmentation |
| CN111223055A (en) * | 2019-11-20 | 2020-06-02 | 南京拓控信息科技股份有限公司 | Train wheel tread image correction method |
| JP2020148744A (en) * | 2019-03-15 | 2020-09-17 | シャープ株式会社 | Image processing device and image processing method |
| CN112801110A (en) * | 2021-02-01 | 2021-05-14 | 中车青岛四方车辆研究所有限公司 | Target detection method and device for image distortion correction of linear array camera of rail train |
| CN112950493A (en) * | 2021-02-01 | 2021-06-11 | 中车青岛四方车辆研究所有限公司 | Method and device for correcting image distortion of linear array camera of rail train |
| CN113534740A (en) * | 2021-09-16 | 2021-10-22 | 沈阳机床(集团)有限责任公司 | Detection data analysis and repair method for rim tread of train wheel |
| CN115311254A (en) * | 2022-09-13 | 2022-11-08 | 万岩铁路装备(成都)有限责任公司 | Steel rail contour matching method based on Harris-SIFT algorithm |
| CN115375693A (en) * | 2022-10-27 | 2022-11-22 | 浙江托普云农科技股份有限公司 | Method, system and device for detecting defects of probe of agricultural information acquisition sensor |
| CN116797479A (en) * | 2023-06-26 | 2023-09-22 | 北京拙河科技有限公司 | Image vertical distortion conversion method |
| CN117289235A (en) * | 2023-11-24 | 2023-12-26 | 成都凌亚科技有限公司 | Antenna radar multichannel signal acquisition processing equipment and method |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100318304A1 (en) * | 2008-02-06 | 2010-12-16 | Ntn Corporation | Bearing for wheel with sensor |
| CN104554344A (en) * | 2014-12-19 | 2015-04-29 | 苏州华兴致远电子科技有限公司 | Tread defect information detection system and method |
-
2017
- 2017-05-17 CN CN201710349621.4A patent/CN107180416B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100318304A1 (en) * | 2008-02-06 | 2010-12-16 | Ntn Corporation | Bearing for wheel with sensor |
| CN104554344A (en) * | 2014-12-19 | 2015-04-29 | 苏州华兴致远电子科技有限公司 | Tread defect information detection system and method |
Non-Patent Citations (4)
| Title |
|---|
| 孙冉: "在线式轮对踏面擦伤成像检测系统", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 张海涛 等: "阈值标记的分水岭彩色图像分割", 《中国图象图形学报》 * |
| 田丽丽 等: "铁路货车车轮踏面伤损检测中剥离与擦伤定位方法", 《铁道学报》 * |
| 胡敏 等: "基于形态学标记连通的分水岭图像分割", 《电子测量与仪器学报》 * |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109668515A (en) * | 2017-10-16 | 2019-04-23 | 苏州华兴致远电子科技有限公司 | Detector for train wheel pair size dynamic detection system and detection method |
| CN109668515B (en) * | 2017-10-16 | 2024-04-05 | 苏州华兴致远电子科技有限公司 | Train wheel set size dynamic detection system and detection method |
| CN109726681A (en) * | 2018-12-29 | 2019-05-07 | 北京航空航天大学 | It is a kind of that location algorithm is identified based on the blind way of machine learning identification and image segmentation |
| JP7237671B2 (en) | 2019-03-15 | 2023-03-13 | シャープ株式会社 | Image processing device and image processing method |
| JP2020148744A (en) * | 2019-03-15 | 2020-09-17 | シャープ株式会社 | Image processing device and image processing method |
| CN111223055A (en) * | 2019-11-20 | 2020-06-02 | 南京拓控信息科技股份有限公司 | Train wheel tread image correction method |
| CN112801110A (en) * | 2021-02-01 | 2021-05-14 | 中车青岛四方车辆研究所有限公司 | Target detection method and device for image distortion correction of linear array camera of rail train |
| CN112801110B (en) * | 2021-02-01 | 2022-11-01 | 中车青岛四方车辆研究所有限公司 | Target detection method and device for image distortion correction of linear array camera of rail train |
| CN112950493A (en) * | 2021-02-01 | 2021-06-11 | 中车青岛四方车辆研究所有限公司 | Method and device for correcting image distortion of linear array camera of rail train |
| CN113534740B (en) * | 2021-09-16 | 2021-11-19 | 沈阳机床(集团)有限责任公司 | Detection data analysis and repair method for rim tread of train wheel |
| CN113534740A (en) * | 2021-09-16 | 2021-10-22 | 沈阳机床(集团)有限责任公司 | Detection data analysis and repair method for rim tread of train wheel |
| CN115311254A (en) * | 2022-09-13 | 2022-11-08 | 万岩铁路装备(成都)有限责任公司 | Steel rail contour matching method based on Harris-SIFT algorithm |
| CN115375693A (en) * | 2022-10-27 | 2022-11-22 | 浙江托普云农科技股份有限公司 | Method, system and device for detecting defects of probe of agricultural information acquisition sensor |
| CN115375693B (en) * | 2022-10-27 | 2023-02-10 | 浙江托普云农科技股份有限公司 | Method, system and device for detecting defects of probe of agricultural information acquisition sensor |
| CN116797479A (en) * | 2023-06-26 | 2023-09-22 | 北京拙河科技有限公司 | Image vertical distortion conversion method |
| CN116797479B (en) * | 2023-06-26 | 2024-02-02 | 北京拙河科技有限公司 | Image vertical distortion conversion method |
| CN117289235A (en) * | 2023-11-24 | 2023-12-26 | 成都凌亚科技有限公司 | Antenna radar multichannel signal acquisition processing equipment and method |
| CN117289235B (en) * | 2023-11-24 | 2024-01-30 | 成都凌亚科技有限公司 | Antenna radar multichannel signal acquisition processing equipment and method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107180416B (en) | 2018-06-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107180416B (en) | Train wheel tread image deformity correction method and system | |
| CN109443480B (en) | Water level scale positioning and water level measuring method based on image processing | |
| CN104050446B (en) | Meter pointer image identification method based on pointer width character | |
| CN108355987B (en) | A kind of screen printing of battery quality determining method based on piecemeal template matching | |
| CN109490316A (en) | A kind of surface defects detection algorithm based on machine vision | |
| CN110211101A (en) | A kind of rail surface defect rapid detection system and method | |
| CN105809138A (en) | Road warning mark detection and recognition method based on block recognition | |
| CN106340044A (en) | Camera external parameter automatic calibration method and calibration device | |
| CN102663760B (en) | Location and segmentation method for windshield area of vehicle in images | |
| CN105488454A (en) | Monocular vision based front vehicle detection and ranging method | |
| CN108985170A (en) | Transmission line of electricity hanger recognition methods based on Three image difference and deep learning | |
| CN105447512A (en) | Coarse-fine optical surface defect detection method and coarse-fine optical surface defect detection device | |
| CN111079734B (en) | Method for detecting foreign matters in triangular holes of railway wagon | |
| CN106643549A (en) | Machine vision-based tile size detection method | |
| CN106815583A (en) | A kind of vehicle at night license plate locating method being combined based on MSER and SWT | |
| CN110189375A (en) | A kind of images steganalysis method based on monocular vision measurement | |
| CN110473199A (en) | A kind of detection of color spot acne and health assessment method based on the segmentation of deep learning example | |
| CN115063579B (en) | Train positioning pin looseness detection method based on two-dimensional image and three-dimensional point cloud projection | |
| CN114972177A (en) | Road disease identification management method and device and intelligent terminal | |
| CN103258218B (en) | Masking-out detects skeleton generating method, device, defect inspection method and device | |
| CN116665126A (en) | Robot inspection part defect detection method and application thereof | |
| CN110348307B (en) | Path edge identification method and system for crane metal structure climbing robot | |
| CN114820474A (en) | Train wheel defect detection method based on three-dimensional information | |
| WO2021109011A1 (en) | Intelligent capacitor internal defect detection method based on ultrasound image | |
| CN112949482B (en) | Non-contact type rail sleeper relative displacement real-time measurement method based on deep learning and visual positioning |
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 |