CN103020638B - One is carried out welding bead based on chromatic information and is known method for distinguishing - Google Patents
One is carried out welding bead based on chromatic information and is known method for distinguishing Download PDFInfo
- Publication number
- CN103020638B CN103020638B CN201210487351.0A CN201210487351A CN103020638B CN 103020638 B CN103020638 B CN 103020638B CN 201210487351 A CN201210487351 A CN 201210487351A CN 103020638 B CN103020638 B CN 103020638B
- Authority
- CN
- China
- Prior art keywords
- welding bead
- image
- sample
- pixel
- mother metal
- 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.)
- Active
Links
Landscapes
- Image Analysis (AREA)
Abstract
The present invention relates to and a kind of carry out welding bead based on chromatic information and know method for distinguishing, comprise the following steps: obtain training image, and gather training data and carry out support vector machine training, obtain Optimal Separating Hyperplane; Read the welding bead image a series of to be identified of shooting continuously; On welding bead image, more than one ROI is set according to prior imformation; Extracting respectively in each ROI of welding bead image and preserve the color information vector of each pixel, and is test data by the color information Definition of Vector of each pixel; Classify to the test data within the scope of each ROI according to step Optimal Separating Hyperplane, specifying test data is belong to welding bead pixel or mother metal pixel, and marks classification results; Successively classification is completed to each frame welding bead image; According to the classification results in each ROI of all welding bead images to be identified, determine welding bead edge.The present invention can be widely used in the groove detection of welding robot or other automation equipment, weld joint tracking, postwelding Non-Destructive Testing tracking, the technology such as appearance of weld quality testing and body surface detection in the automatic identifying of welding bead.
Description
Technical field
The present invention relates to a kind of welding bead and know method for distinguishing, carry out welding bead knowledge method for distinguishing particularly about the Automations of Welding such as a kind of robot automation's of being applicable to welding, weld joint tracking, welding bead identification, postwelding Non-Destructive Testing homing guidance and intellectualized detection based on chromatic information.
Background technology
Welding bead identifies it is gordian technique in Automation of Welding and intelligent development process automatically, can be widely used in the aspects such as the automatic teaching of welding track, weld seam real-time follow-up and the homing guidance of postwelding non-destructive detecting device.Visual sensing technology (vision sensor) is because its obtaining information amount is large, contactless, become the automatic recognition method of the welding bead be most widely used at present by advantages such as electromagnetic interference (EMI) are little, it mainly obtains welding bead image by industrial camera shooting, welding bead image is carried out process and obtain welding bead positional information or welding bead edge position information, and adjusted by the conduct route of feedback control module Butt welding gun or non-destructive control probe, ensure that welding gun or non-destructive control probe can accurately advance along welding bead, guarantee welding or Detection job.
The more ripe weld joint tracking product of current employing is all that structure based light method realizes welding bead identification, mainly utilize groove or the obvious three-dimensional geometrical structure of welding bead, the identification object of image procossing is the distortion that structure striation occurs, but in the unconspicuous situation of welding bead structure, said method is difficult to realize the identification to welding bead and tracking, for the cosmetic welding of multi-pass welding, before cosmetic welding, groove is filled full by which floor welding bead front, structure striation projects welding bead surface can not there is obvious distortion, in addition welding bead or the neighbouring mother metal out-of-flatness that may exist or carry the impact of impurity on striation shape secretly, cause utilizing Structure light method to be difficult to obtain reliable weld edge position, again to remove the shaping weld seam of reinforcement through Milling Process, because welding bead and neighbouring mother metal are almost in same plane equally, structured light can not reflect the position on its border, also cause utilizing Structure light method to be difficult to obtain reliable weld edge position, and for quite a few welding bead, due to material, welding technology, the impact of the complicated reasons such as corrosion situation, possess along the metastable color character of bead direction, and with the color in mother metal region, there is comparatively significantly difference, therefore, color information can be adopted to realize the automatic recognition and tracking of welding bead.
Summary of the invention
For the problems referred to above, the object of this invention is to provide a kind of color character based on welding bead and mother metal, set up classification and criterion of identification by support vector machine, realize carrying out welding bead based on chromatic information and knowing method for distinguishing of self-adaptation welding bead identification.
For achieving the above object, the present invention takes following technical scheme: a kind of carry out welding bead based on chromatic information and know method for distinguishing, comprises the following steps: 1) obtain training image, and gathers training data and carry out support vector machine training, obtains Optimal Separating Hyperplane; 2) the welding bead image a series of to be identified taken continuously through colored industrial camera is read; 3) on welding bead image, set more than one ROI according to prior imformation, make the ROI set can cover the region that welding bead edge may occur; 4) extracting respectively in each ROI of welding bead image and preserve the color information vector of each pixel, and is test data by the color information Definition of Vector of each pixel; 5) classify to the test data within the scope of each ROI according to the Optimal Separating Hyperplane of support vector machine training gained in step 1), specifying test data is belong to welding bead pixel or mother metal pixel, and marks classification results; 6) successively step 3) ~ 5 are repeated to each frame welding bead image) complete classification; 7) according to the classification results in each ROI of all welding bead images to be identified, welding bead edge is determined.
Described step 1) obtains training image, and gather training data and carry out support vector machine training, obtain Optimal Separating Hyperplane, comprise the following steps: 1. adopt colored industrial camera, take a welding bead sample image in welding bead starting point to be welded or to be detected, take the welding bead sample image obtained and comprise welding bead region, welding bead fringe region and a part of mother metal region near welding bead; 2. welding bead sample image shooting obtained, as training image, selectes sample welding bead district and sample mother metal district in training image; 3. the mode of stochastic generation sampled point pixel coordinate is adopted to generate pixel sampling point respectively in sample welding bead district and sample mother metal district, and sample welding bead district and sample mother metal district choose several pixels respectively as sampled point, extract the color information vector of each sampled point in sample welding bead district and sample mother metal district, it can be used as training data to preserve, and category is labeled as " welding bead pixel " or " mother metal pixel " respectively; 4. utilize training data to carry out the training of support vector machine, obtain the color information vector in sample welding bead district and the Optimal Separating Hyperplane of the color information vector in sample mother metal district.
Described sample welding bead district comprises welding bead region large as far as possible, and the mother metal region near welding bead edge is selected in sample mother metal district as far as possible.
Described welding bead image comprises welding bead region, welding bead fringe region and a part of mother metal region near welding bead.
Described color information vector adopts the one in RGB component, HSV component or YCbCr component.
Determine in described step 7) that the method at welding bead edge is by searching out an optimum separatrix, making the classification results accuracy of both sides, separatrix reach the highest, using this separatrix as the welding bead marginal position identifying gained.
The present invention is owing to taking above technical scheme, it has the following advantages: 1, the present invention adopts support vector machine to set up classification and criterion of identification to the color-vector information of welding bead image, when welding bead architectural feature is not obvious cause using Structure light method to obtain positional information, based on image color different in welding bead district and mother metal district, to welding bead Image Segmentation Using, and then obtain the positional information at welding bead edge, therefore can be used for tracker according to identifying that the welding bead marginal position Butt welding gun that obtains or non-destructive control probe position adjust, the accuracy of effective raising work.2, first the present invention adopts the machine learning method based on support vector machine to the welding bead sample image of starting point, the color character in welding bead district and mother metal district is sampled and learnt, the color information vector in automatic generation welding bead district and the Optimal Separating Hyperplane of the color information vector in mother metal district, therefore, it is possible to complete the identification mission of dissimilar welding bead adaptively, adapt to produce actual demand.3, the present invention is after support vector machine completes training, according to the Optimal Separating Hyperplane obtained, each pixel in the area-of-interest in every frame welding bead image is classified, this process is consuming time short, classification accurately, have a clear superiority in than existing Structure light method, there is comparatively strong adaptability, and in accuracy and real-time, reach the requirement of producing reality.The present invention can be widely used in the technology such as the groove detection of welding robot or other automation equipment, weld joint tracking, postwelding Non-Destructive Testing tracking, appearance of weld quality testing and body surface detection in the automatic identifying of welding bead, is specially adapted to fill in multilayer welding bead the identification mission of weldering and cosmetic welding and Milling Process and removes in the identification mission of weld seam after reinforcement.
Accompanying drawing explanation
Fig. 1 is the method flow schematic diagram that the present invention carries out welding bead identification;
Fig. 2 is the effect schematic diagram that the welding bead district of welding bead sample image of the present invention and mother metal district carry out sampling, and wherein, triangle represents mother metal district sampled point, and circle represents welding bead district sampled point;
Fig. 3 is the effect schematic diagram that the embodiment of the present invention sets area-of-interest on welding bead image, wherein, and the area-of-interest that little box indicating is selected;
Fig. 4 is the effect schematic diagram that the present invention classifies to area-of-interest according to support vector machine training result, and wherein, black is mother metal pixel, and white is welding bead pixel;
Fig. 5 is welding bead limb recognition result effect schematic diagram of the present invention, and in area-of-interest, black lines represents welding bead edge.
Embodiment
Below in conjunction with drawings and Examples, the present invention is described in detail.
As shown in Figure 1, of the present invention based on chromatic information carry out welding bead know method for distinguishing comprise the following steps:
1, obtain training image, and gather training data and carry out support vector machine training, obtain Optimal Separating Hyperplane, it comprises the following steps:
1) adopt existing colored industrial camera, take a welding bead sample image in welding bead starting point to be welded or to be detected, take the welding bead sample image obtained and comprise welding bead region, welding bead fringe region and a part of mother metal region near welding bead.
2) welding bead sample image shooting obtained is as training image, selected sample welding bead district and sample mother metal district in training image, wherein, sample welding bead district comprises welding bead region large as far as possible, and the mother metal region near welding bead edge is selected in sample mother metal district as far as possible.
3) mode of stochastic generation sampled point pixel coordinate is adopted to generate pixel sampling point respectively in sample welding bead district and sample mother metal district, and sample welding bead district and sample mother metal district choose 50 pixel sampling points respectively (is not limited thereto as sampled point, the number of sampled point can be selected according to actual needs), extract the color information vector of each sampled point in sample welding bead district and sample mother metal district, it can be used as training data to preserve, and category is labeled as " welding bead pixel " or " mother metal pixel " respectively.As shown in Figure 2, in embodiments of the invention, circle represents that sampled point belongs to welding bead pixel, and triangle represents that sampled point belongs to mother metal pixel.
In addition, the training data in the embodiment of the present invention adopts the RGB component of each sampled point, but be not limited thereto, and can choose the component in the different color spaces such as the HSV component of each sampled point or YCbCr as training data according to actual detection case.
4) utilize training data to carry out the training of support vector machine, obtain the color information vector in sample welding bead district and the Optimal Separating Hyperplane of the color information vector in sample mother metal district.
In embodiments of the invention, support vector machine type is C-SVM, adopts linear function as kernel function, the quantity of training data is every class 50, the optimization method adopted when solving Optimal Separating Hyperplane is Quadratic Programming Solution, and solution procedure is prior art, does not repeat them here.
2, read the welding bead image a series of to be identified taken continuously through colored industrial camera, each frame welding bead image comprises welding bead region, welding bead fringe region and a part of mother metal region near welding bead.The pixel size of the welding bead image absorbed in the present embodiment is 288 × 205, but is not limited thereto, and can arrange the size of welding bead image according to concrete application.
3, as shown in Figure 3, on welding bead image, more than one ROI(area-of-interest is chosen) according to prior imformation, the region making the ROI chosen can cover welding bead edge may to occur, in the testing process of reality, following the tracks of due to welding bead is a continuous print process, and in every frame welding bead image, welding bead edge only may appear at the limited range of skew certain distance near former frame welding bead image.The ROI size of the embodiment of the present invention is that 50 × 50, ROI size is not limited thereto, and the needs that can detect according to reality set.
Wherein, prior imformation is choose the reference information of ROI, not only comprise the actual welding bead position of actual welding bead to be detected and the possible range of weld width, and owing to take continuously during welding bead image, when a rear frame welding bead image chooses ROI, the recognition result of former frame welding bead image also can become the prior imformation choosing ROI.
4, extract respectively in each ROI of welding bead image and preserve each pixel color information vector, be test data by the color information Definition of Vector of each pixel.
5, the Optimal Separating Hyperplane of support vector machine training gained in step 1 is adopted to classify to the test data within the scope of each ROI, specifying test data is welding bead pixel or mother metal pixel, and classification results is marked, as shown in Figure 4, in the classification results of embodiments of the invention, white represents that this pixel is classified as welding bead pixel, and black represents that this pixel is classified as mother metal pixel.
6, successively step 3 ~ 5 are repeated to each frame welding bead image and complete classification.
7, according to the classification results in each ROI of all welding bead images to be identified, determine welding bead edge, that is: an optimum separatrix is searched out, the classification results accuracy of both sides, separatrix is made to reach the highest, using this separatrix as the welding bead marginal position identifying gained, as shown in Figure 5, the straight line of white is welding bead marginal position.
In sum, of the present invention based on chromatic information carry out welding bead know method for distinguishing can C Plus Plus realize, also can adopt other programming languages.Embodiments of the invention CPU frequency be 2.4GHz, in save as testing results in the PC environment of 2G, the processing time of every frame welding bead image is 4 ~ 5ms, adopt the accuracy of identification of method of the present invention to welding bead border to reach welding tracking requirement, meet the requirement of real-time needed for production reality completely.
Above-described embodiment is only for illustration of the present invention; wherein training data obtain manner, color space type, SVM type, lineoid optimization method, ROI size and position, all can change to some extent according to the mode etc. on classification results determination welding bead border; every equivalents of carrying out on the basis of technical solution of the present invention and improvement, all should not get rid of outside protection scope of the present invention.
Claims (9)
1. carry out welding bead based on chromatic information and know a method for distinguishing, comprise the following steps:
1) obtain training image, and gather training data and carry out support vector machine training, obtain Optimal Separating Hyperplane;
2) the welding bead image a series of to be identified taken continuously through colored industrial camera is read;
3) on welding bead image, set more than one ROI according to prior imformation, make the ROI set can cover the region that welding bead edge may occur;
4) extracting respectively in each ROI of welding bead image and preserve the color information vector of each pixel, and is test data by the color information Definition of Vector of each pixel;
5) according to step 1) in support vector machine training gained Optimal Separating Hyperplane the test data within the scope of each ROI is classified, specifying test data is belong to welding bead pixel or mother metal pixel, and marks classification results;
6) successively step 3 is repeated to each frame welding bead image) ~ 5) complete classification;
7) according to the classification results in each ROI of all welding bead images to be identified, welding bead edge is determined.
2. as claimed in claim 1 a kind of carry out welding bead based on chromatic information and know method for distinguishing, it is characterized in that: described step 1) obtain training image, and gather training data and carry out support vector machine training, obtain Optimal Separating Hyperplane, comprise the following steps:
1. adopt colored industrial camera, take a welding bead sample image in welding bead starting point to be welded or to be detected, take the welding bead sample image obtained and comprise welding bead region, welding bead fringe region and a part of mother metal region near welding bead;
2. welding bead sample image shooting obtained, as training image, selectes sample welding bead district and sample mother metal district in training image;
3. the mode of stochastic generation sampled point pixel coordinate is adopted to generate pixel sampling point respectively in sample welding bead district and sample mother metal district, and sample welding bead district and sample mother metal district choose several pixels respectively as sampled point, extract the color information vector of each sampled point in sample welding bead district and sample mother metal district, it can be used as training data to preserve, and category is labeled as " welding bead pixel " or " mother metal pixel " respectively;
4. utilize training data to carry out the training of support vector machine, obtain the color information vector in sample welding bead district and the Optimal Separating Hyperplane of the color information vector in sample mother metal district.
3. one as claimed in claim 2 carries out welding bead knowledge method for distinguishing based on chromatic information, and it is characterized in that: described sample welding bead district comprises welding bead region large as far as possible, the mother metal region near welding bead edge is selected in sample mother metal district as far as possible.
4. one as claimed in claim 1 carries out welding bead knowledge method for distinguishing based on chromatic information, it is characterized in that: described welding bead image comprises welding bead region, welding bead fringe region and a part of mother metal region near welding bead.
5. the one as described in claim 1 or 2 or 3 is carried out welding bead based on chromatic information and is known method for distinguishing, it is characterized in that: described color information vector adopts the one in RGB component, HSV component or YCbCr component.
6. one as claimed in claim 4 carries out welding bead knowledge method for distinguishing based on chromatic information, it is characterized in that: described color information vector adopts the one in RGB component, HSV component or YCbCr component.
7. the one as described in claim 1 or 2 or 3 or 6 is carried out welding bead based on chromatic information and is known method for distinguishing, it is characterized in that: described step 7) in determine that the method at welding bead edge is by searching out an optimum separatrix, the classification results accuracy of both sides, separatrix is made to reach the highest, using this separatrix as the welding bead marginal position identifying gained.
8. one as claimed in claim 4 carries out welding bead knowledge method for distinguishing based on chromatic information, it is characterized in that: described step 7) in determine that the method at welding bead edge is by searching out an optimum separatrix, the classification results accuracy of both sides, separatrix is made to reach the highest, using this separatrix as the welding bead marginal position identifying gained.
9. one as claimed in claim 5 carries out welding bead knowledge method for distinguishing based on chromatic information, it is characterized in that: described step 7) in determine that the method at welding bead edge is by searching out an optimum separatrix, the classification results accuracy of both sides, separatrix is made to reach the highest, using this separatrix as the welding bead marginal position identifying gained.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210487351.0A CN103020638B (en) | 2012-11-26 | 2012-11-26 | One is carried out welding bead based on chromatic information and is known method for distinguishing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210487351.0A CN103020638B (en) | 2012-11-26 | 2012-11-26 | One is carried out welding bead based on chromatic information and is known method for distinguishing |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103020638A CN103020638A (en) | 2013-04-03 |
| CN103020638B true CN103020638B (en) | 2015-09-30 |
Family
ID=47969228
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210487351.0A Active CN103020638B (en) | 2012-11-26 | 2012-11-26 | One is carried out welding bead based on chromatic information and is known method for distinguishing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103020638B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3418952A1 (en) * | 2017-06-20 | 2018-12-26 | Lincoln Global, Inc. | Machine learning for weldment classification and correlation |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106934407B (en) * | 2015-12-29 | 2021-04-02 | 航天信息股份有限公司 | Method and system for measuring height of deposit adapting to illumination change |
| CN106990112B (en) * | 2017-03-14 | 2019-07-26 | 清华大学 | Multi-layer multi-pass welding track detection device and method based on multi-visual information fusion |
| CN111037549B (en) * | 2019-11-29 | 2022-09-09 | 重庆顺泰铁塔制造有限公司 | Welding track processing method and system based on 3D scanning and TensorFlow algorithm |
| CN113744243B (en) * | 2021-09-03 | 2023-08-15 | 上海柏楚电子科技股份有限公司 | Image processing method, device, equipment and medium for weld joint tracking detection |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101122585A (en) * | 2007-09-12 | 2008-02-13 | 天津大学 | Automatic identification method for supersonic phased array for detecting oil gas pipeline girth weld defect type |
-
2012
- 2012-11-26 CN CN201210487351.0A patent/CN103020638B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101122585A (en) * | 2007-09-12 | 2008-02-13 | 天津大学 | Automatic identification method for supersonic phased array for detecting oil gas pipeline girth weld defect type |
Non-Patent Citations (1)
| Title |
|---|
| 基于图像色彩信息的焊缝识别算法;邹怡蓉等;《焊接学报》;37-41;20091031;第30卷(第10期);37-41 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3418952A1 (en) * | 2017-06-20 | 2018-12-26 | Lincoln Global, Inc. | Machine learning for weldment classification and correlation |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103020638A (en) | 2013-04-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109900706B (en) | Weld joint based on deep learning and weld joint defect detection method | |
| CA2950791C (en) | Binocular visual navigation system and method based on power robot | |
| CN103020638B (en) | One is carried out welding bead based on chromatic information and is known method for distinguishing | |
| Li et al. | Robust welding seam tracking and recognition | |
| CN110480127A (en) | A kind of seam tracking system and method based on structured light visual sensing | |
| CN104156726A (en) | Workpiece recognition method based on geometric shape feature and device thereof | |
| CN107424144A (en) | Weld joint tracking image processing algorithm based on laser vision | |
| CN104400265A (en) | Feature extraction method applicable to corner weld of laser vision guided welding robot | |
| CN114140439A (en) | Laser welding seam feature point identification method and device based on deep learning | |
| CN108106617A (en) | A kind of unmanned plane automatic obstacle-avoiding method | |
| Yang et al. | Detection of weld groove edge based on multilayer convolution neural network | |
| CN108582075A (en) | A kind of intelligent robot vision automation grasping system | |
| CN106442533A (en) | Weld information extracting system based on industrial CCD | |
| CN111435429B (en) | Gesture recognition method and system based on binocular stereo data dynamic cognition | |
| Liu et al. | Automatic seam detection of welding robots using deep learning | |
| Yevsieiev et al. | THE CANNY ALGORITHM IMPLEMENTATION FOR OBTAINING THE OBJECT CONTOUR IN A MOBILE ROBOT’S WORKSPACE IN REAL TIME | |
| Lin et al. | Intelligent seam tracking of an ultranarrow gap during K-TIG welding: a hybrid CNN and adaptive ROI operation algorithm | |
| CN113723494A (en) | Laser visual stripe classification and weld joint feature extraction method under uncertain interference source | |
| Jin et al. | A new welding seam recognition methodology based on deep learning model MRCNN | |
| Jiang et al. | Mobile robot gas source localization via top-down visual attention mechanism and shape analysis | |
| CN116243329A (en) | High-precision multi-target non-contact ranging method based on laser radar and camera fusion | |
| Budzan | Fusion of visual and range images for object extraction | |
| Ma et al. | WeldNet: A deep learning based method for weld seam type identification and initial point guidance | |
| Lu et al. | Plate additive, seam-tracking technology based on feature segmentation | |
| Chen et al. | Refining object proposals using structured edge and superpixel contrast in robotic grasping |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |