CN107218894A - A kind of subpixel accuracy thickness of detector detection method of fast and stable - Google Patents
A kind of subpixel accuracy thickness of detector detection method of fast and stable Download PDFInfo
- Publication number
- CN107218894A CN107218894A CN201710295849.XA CN201710295849A CN107218894A CN 107218894 A CN107218894 A CN 107218894A CN 201710295849 A CN201710295849 A CN 201710295849A CN 107218894 A CN107218894 A CN 107218894A
- Authority
- CN
- China
- Prior art keywords
- value
- coordinate
- boundary
- curve
- brightness
- 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
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01B—MEASURING LENGTH, THICKNESS OR SIMILAR LINEAR DIMENSIONS; MEASURING ANGLES; MEASURING AREAS; MEASURING IRREGULARITIES OF SURFACES OR CONTOURS
- G01B11/00—Measuring arrangements characterised by the use of optical techniques
- G01B11/02—Measuring arrangements characterised by the use of optical techniques for measuring length, width or thickness
- G01B11/06—Measuring arrangements characterised by the use of optical techniques for measuring length, width or thickness for measuring thickness ; e.g. of sheet material
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
- Length Measuring Devices By Optical Means (AREA)
Abstract
The present invention relates to a kind of subpixel accuracy thickness of detector detection method of fast and stable.Comprise the following steps：Border Primary Location：Two regions of target and air are divided the image into, the line of demarcation in the two regions is then obtained, is used as preliminary boundary alignment result；Sub-pix precise positioning：Using the method for arc tangent curve matching, solved, be fitted for the positioning result of each row, then using Bezier using particle swarm optimization algorithm；Abnormal conditions processing.It is improved for sub-pix boundary alignment algorithm, substantially estimate from first positioning, parameter, in terms of fast search optimal parameter three, on the basis of it ensure that positioning precision meets requirement, the lifting of algorithm speed is furthermore achieved that, so as to improve the application potential of sub-pix boundary alignment algorithm.
Description
Technical field
The invention belongs to automatic detection field, it is related to industrial vision technical field, and in particular to a kind of fast and stable
Subpixel accuracy thickness of detector detection method.
Background technology
With developing rapidly for China's manufacturing industry, the attention to quality becomes more and more important.In the production process of various products,
Quality testing is carried out using automation equipment to various components, it is possible to rate of reducing the number of rejects and seconds, reduce the loss to material, this is
Through as the strong means for improving enterprise profit and brand influence.Industrial vision technology is that automatic detection field is most normal
Technology.
The thickness and size of component are measured, are the most common applications of industrial vision detection.Obtaining first device
After the dimensional parameters of part, it is possible to judge whether device is qualified, so as to be rejected in advance, prevent from making follow-up manufacture link
Into influence.Which by measuring the various sizes parameter of component, then contrasted with standard size, it is possible to first device known
Part is underproof, and underproof ratio and detail.
The Thickness sensitivity of device, is the most common step in automatic detection.Current Thickness sensitivity, be all by
In image, the sub-pix border of device is positioned and realized.But, current common sub-pix boundary alignment algorithm,
Complexity is too high, the redundant computation that there is higher degree, causes the efficiency of automatic checkout equipment and reduces, so that production capacity is reduced,
It has impact on the profit of enterprise.
The content of the invention
It is an object of the present invention to provide a kind of subpixel accuracy thickness of detector detection method of fast and stable, for sub-pix
Boundary alignment algorithm is improved, and is substantially estimated from first positioning, parameter, in terms of fast search optimal parameter three, ensure that
Positioning precision is met on the basis of requirement, furthermore achieved that the lifting of algorithm speed, so as to improve sub-pix boundary alignment calculation
The application potential of method.
In order to realize the above object the technical solution adopted by the present invention is：A kind of subpixel accuracy device of fast and stable
Part thickness detecting method, comprises the following steps：
1) border Primary Location：Two regions of target and air are divided the image into, the boundary in the two regions is then obtained
Line, is used as preliminary boundary alignment result；
2) sub-pix precise positioning：Using the method for arc tangent curve matching, solved using particle swarm optimization algorithm,
It is fitted for the positioning result of each row, then using Bezier；
3) abnormal conditions are handled.
Further, the deviation of the first position location of border Primary Location and final sub-pix scenic spot coordinate is in 1 picture
Within the distance of element.
Further, described border Primary Location comprises the following steps：
A) by the thick border Primary Location to essence
The Primary Location coordinate points of interface are obtained according to following flow：
1a) image down：Image is first reduced, a quarter of original size is narrowed down to；
2a) extract target area reference brightness value, air section reference brightness value：Precalculate in image, target area
With the brightness reference value of air section, using clustering method, to all pixels, according to brightness value, it is divided into two major classes, Ran Houji
The average brightness of each class is calculated, respectively as object brightness and the reference value of air section brightness；
3a) binarization operation：After the reference brightness value of target area and air section is obtained, the equal of the two is utilized
All pixels, as the binary-state threshold of entire image, are carried out binarization operation by value；
4a) extract largest connected region：Extract the largest connected region in binarization operation result；
5a) the preliminary boundary alignment based on symmetry；Utilize the symmetrical of the pixel intensity variation rule curve near border
Property obtains preliminary judgement；
Abnormality processing when 6a) just positioning；
B) positioning at load-bearing interface
The scope at angle of inclination 1b) is set, the angle is traveled through, according to the angle value of setting, bianry image entered
Row floor projection, obtains drop shadow curve；
2b) to the drop shadow curve under various angles, analyzed, statistics projection value rises to picture traverse value N, institute from 0
The line number of experience is needed, the minimum corresponding angle value of situation of line number is selected as the inclination angle at load-bearing interface；
The corresponding drop shadow curve in inclination angle 3b) is found, the difference of two row projection values above and below Difference Calculation, calculating is carried out to it,
The a line for selecting difference value maximum, is used as the position at load-bearing interface；
4b) and then in difference curves, the corresponding value in load-bearing interface is filtered out, maximum of which difference value is searched again for, as
The first position location on top device border；
C) positioning of device right boundary：
1c) the top device boundary position obtained using preceding step, and load-bearing interface position, obtain the two point
Boundary line；
2c) point that the coordinate value in all boundary points is located above line of demarcation is extracted；
3c) according to the point that is extracted, the leftmost side and the rightmost side are found；
4c) in the pixel coverage of 20, left-hand point periphery, upright projection is carried out to binary map；
Difference 5c) is carried out to drop shadow curve, the maximum position of difference value is found, is used as the left margin of device；
6c) in the pixel coverage of 20, right-hand point periphery, upright projection is carried out to binary map, it is poor that drop shadow curve is carried out
Point, the maximum position of difference value is found, the right margin of device is obtained；
D) positioning of top device coordinate fillet
The load-bearing areal coordinate that 1d) is obtained using preceding step, top device coordinate, device right boundary position, by device
Corresponding sub- contours extract comes out；
The sub- profile of device 2d) is directed to, its corresponding convex closure is calculated；
The convex closure 3d) is utilized, new bianry image is generated, convex closure area filling is white；
4d) convex closure area image makes the difference with original device image；
5d) in differential chart, four maximum connected regions of area are navigated to；
The position for obtaining four connected regions 6d) is calculated, device fillet position is obtained.
Further, described sub-pix precise positioning passes through bent to the row pixel near arc tangent curve and boundary position
Line is fitted, and the mathematic(al) representation of arc tangent curve is：Y=p1*atan (p2*x+p3)+p4, is obtained when the brightness in prostatitis is bent
The mathematical models of line, then, are obtained using following formula：X0=-p3/p2, calculates the subpixel coordinates for obtaining each row.
Further, described sub-pix precise positioning comprises the following steps：
A) the direct valuation of two parameters of p1 and p4：The shape of brightness curve is utilized in advance, and two parameters of p1 and p4 are calculated
Out, then the two parameter values for calculating as p1 and p4 initial value：
According to following flow：
1a) since above initial boundary position, it will be gradually added to when the pixel in prostatitis in set, if in set
Pixel intensity standard deviation be less than 5, then continue to increase new pixel, if greater than 5, then stop the new pixel of increase, calculate this
The average value of pixel intensity in Shi Jihe, is used as the mean flow rate on the upside of this；
2a) since below initial boundary position, it will be gradually added to when the pixel in prostatitis in set, if in set
Pixel intensity standard deviation be less than 5, then continue to increase new pixel, if greater than 5, then stop the new pixel of increase, calculate this
The average value of pixel intensity in Shi Jihe, obtains the mean flow rate on the downside of this, the mean flow rate on the upside of this and putting down on the downside of this
The difference of equal brightness is p1；
3a) p4 calculating：Pixel brightness value all on brightness curve is calculated into average value, boundary curve central point is obtained
The ordinate of position；
B) the arc tangent curve based on particle cluster algorithm is solved：
According to following flow：
One group of particle 1b) is set, 20 altogether, each particle includes 4 parameters, and initial value is set at random, represents anyway
Cut a solution of curve；
The curve corresponding to current 20 particles 2b) is calculated, error is negated, made by the error with actual brightness curve
For fitness；
Current most suitable parameter of curve 3b) is found, the direction and position of each particle is then adjusted, continues to calculate；
4b) repeat 2b), 3b) two steps, untill the error of best particle is less than threshold value 10；
5b) calculated according to formula x0=-p3/p2 and obtain accurate sub-pix boundary position；
C) application of previous column result, obtains p2 and p3 initial value：
After a specified row curve matching is completed, by the p2 and p3 value in the parameter of the row, the bright of next column is used as
Write music the parameter of line, p1 and p4 are still estimated by calculating；
D) solution of noise jamming：
It is larger for error to determine when whether the fitting result in prostatitis is correct by judging the error in fit procedure
Situation, the row, without the precise positioning on border, are to be repaired by edge smoothing in the later stage；
E) edge smoothing：
Using Bezier approximating method, the fitting of border point coordinates is completed.
Further, described abnormal conditions processing includes
A) there is no device early warning：Detected by the vertical coordinate analyzed on glass interface, by the coordinate of interface,
Fitting is in alignment, then calculates each point to the distance of straight line, if arriving the distance of straight line a little, both less than 3
Pixel, then without device, for not having the situation that device is present in image, it is necessary to carry out early warning；
B) multiple device early warning：Including：
1b) polylith device is separated from each other：1 time, early warning are above if up number of times and decline number of times；
2b) polylith device occurs overlapping：In the rule using device coordinate raising and lowering, the right boundary of device is obtained
Afterwards, the thickness distribution of the device between right boundary is investigated, between right boundary, if the thickness value of some position,
It is equal to or over the height of double thickness, then explanation there occurs the overlapping of multiple devices, carry out early warning；
C) gap early warning：
1c) the contours extract of device area：During the Primary Location of device boundaries, the left and right of device is had been obtained for
The position on border, according to this position, the corresponding binaryzation subgraph of device is intercepted out, profile is then extracted, follow-up step
Suddenly the detection in gap will on the basis of profile, be carried out；
2c) top device and left and right sides profile are filtered out：Using the load-bearing interface coordinate obtained in first position fixing process, with
And device right boundary coordinate, the lower left corner and the bottom right angular coordinate of device are obtained, then according to the two coordinate values, by device
Top profile, left side profile, right lateral contours are all filtered out, and leave behind the profile point of gap area；
3c) gap area dimensional measurement：By above 1c), 2c) two steps processing, now retain the straight line got off
Section, is the profile corresponding to the gap between device and glass planar, the profile point of analysis now, calculates the border in the region,
If the height in the region is less than 5 pixels, and width is more than 20 pixels, then is gap, it is necessary to early warning；
D) filtering of dust interference：
1d) to sub-pix boundary coordinate, local fitting a straight line is carried out：Increased using side, the mode of side search is exploratory
Boundary point is divided into multiple writings to close；
Distance 2d) is more than to the coordinate points of threshold value, picked out；In each subset, by profile distance between beeline and dot
Coordinate points more than threshold value are picked out；
3d) for the coordinate points for being selected out, it is segmented；The coordinate points for being selected out in analysis subset, according to
Whether connect, be divided into multiple subsegments；
The width of each subsegment 4d) is investigated, width is less than threshold value, and all can be regarded as is dust；
5d) recalculate coordinate points：The corresponding subsegment of dust is filtered out, by the coordinate points in the subsegment, all with corresponding
Coordinate points on the straight line of fitting are substituted；
E) early warning of conditions of streaking
1e) the judgement of top device border vertical coordinate uniformity：In the left and right ends of device, a length is each selected
For 20 region, then in the center of device, one length of selection is 50 region, and the device in these three regions is calculated respectively
The average value of the top profile point vertical coordinate of part, if having a region, its average coordinates and center among left and right ends
Domain average coordinates are compared, and difference is more than 5, then there is conditions of streaking, early warning；
2e) device center region, is analyzed with the brightness uniformity of two end regions of head and the tail：With 1e) equally, in the left and right of device
Two ends, the region that each one length of selection is 20, then in the center of device, one length of selection is 50 region,
The mean flow rate in these three regions is calculated respectively, if there is a region among left and right ends, its brightness is averaged with central area
Brightness, difference is more than 20, then there is conditions of streaking, early warning.
The technical effects of the invention are that：
1st, speed is fast, sub-pix boundary alignment method proposed by the invention, in the links of algorithm, has all carried out most
The optimization of limits, bottom line has been reduced by amount of calculation, and the overall operation speed of system can be improved to greatest extent；
2nd, precision is high, the actual (real) thickness (repeatedly being measured using slide measure) with device, and deviation is within 3 microns；
3rd, various disturbed conditions are adapted to, during the true use of system, the various abnormal conditions being likely encountered are entered
Processing is gone, it is ensured that these abnormal conditions in the case of necessity, can also be carried out without interference with normal detection to abnormal conditions
Early warning.
Brief description of the drawings
Fig. 1 is thickness of detector detecting system general flow chart of the present invention；
Fig. 2 is device boundaries Primary Location flow chart of the present invention；
Fig. 3 is device load-bearing interface positioning flow figure of the present invention；
Fig. 4 is the positioning flow figure of device right boundary of the present invention；
Fig. 5 is device fillet position calculation flow chart of the present invention；
Fig. 6 is sub-pix boundary alignment flow chart of the present invention；
Fig. 7 is parameter p1 and p4 of the present invention estimation flow chart；
Fig. 8 solves flow chart for arc tangent curve of the present invention based on PSO algorithms；
Fig. 9 is the abnormal conditions process chart in thickness of detector detection process of the present invention；
Figure 10 is the overhaul flow chart in the presence of multiple devices of the invention；
Figure 11 is early warning flow chart in device gap of the present invention；
Figure 12 is filtering dust flow chart of the present invention；
Figure 13 is conditions of streaking early warning flow chart of the present invention.
Embodiment
Referring to the drawings, the abnormal conditions that the present invention is handled include：There is no have multiple devices, device in device, the visual field in the visual field
Part occurs to overlap, and there is the dust interference on gap, device between device and glass card, device angles, which are put, is forbidden what is caused
Whipping phenomenon.It is improved for sub-pix boundary alignment algorithm, from first positioning, parameter is substantially estimated, fast search is most preferably joined
Three aspects of number, on the basis of it ensure that positioning precision meets requirement, furthermore achieved that the lifting of algorithm speed, so as to carry
The application potential of high sub-pix boundary alignment algorithm.
1st, the Primary Location on border
Two regions of target and air are divided the image into, the line of demarcation in the two regions is then obtained, is used as preliminary side
Boundary's positioning result.So follow-up sub-pixel is other to be accurately positioned, and is deployed on this basis, so as to save substantial amounts of search
Time.In addition, in the Primary Location stage, also by the coordinate according to boundary point, obtain the horizontal level where device.
2nd, sub-pix precise positioning
The method that the present invention uses arc tangent curve matching.To realize that the subpixel accuracy of boundary point is positioned.In order to carry
The speed of high curve matching, is solved using particle swarm optimization algorithm.Finally, for the positioning result of each row, then use
Bezier is fitted, and make that final edge becomes is smooth.
3rd, abnormal conditions are handled
The step is, it is necessary to solve some abnormal conditions：There is no there are multiple devices in device, the visual field in the visual field, device hair
It is raw overlapping, there is the dust interference on gap, device between device and glass card, device angles put the whipping for being forbidden to cause
Phenomenon.
Specifically, the position of 1. quick positioner parts in the picture：Due to each coordinate points position, subpixel coordinates are carried out
Accurate solution, required amount of calculation is very big, it is therefore necessary to subpixel coordinates solution before, just obtain border just
Position location.The deviation of first position location and final sub-pix scenic spot coordinate is within the distance of 1 pixel.So can be
During follow-up arc tangent curve matching, the number of times of search is reduced, completes to solve as early as possible.
A) by the thick border Primary Location to essence
The Primary Location coordinate points of interface are obtained according to following flow：
1a) image down
In order to improve the precision of coarse positioning, image is first reduced, a quarter of original size is narrowed down to.So calculate
Method can handle as far as possible few pixel, while not interfering with final precision again.
2a) extract target area reference brightness value, air section reference brightness value
, it is necessary to precalculate in image, the brightness references of target area and air section before binarization operation is carried out
Value.Using clustering method, to all pixels, according to brightness value, it is divided into two major classes.Then the brightness for calculating each class is averaged
Value, respectively as object brightness and the reference value of air section brightness.
3a) binarization operation
After the reference brightness value of target area and air section is obtained, using the average of the two, entire image is used as
Binary-state threshold, to all pixels carry out binarization operation.
4a) extract largest connected region
In order to reduce the interference of noise as far as possible, it is necessary to extract the largest connected region in binarization operation result.Because device
Part and weight-bearing surface are adhesions, and they collectively constitute target area, therefore only need to find the connected region of maximum area, so that it may
To filter out the interference of various noises.
5a) the preliminary boundary alignment based on symmetry
After binarization operation, the border of connected domain, may and real boundary bit be equipped with certain deviation because two
Value threshold value, is not necessarily equal to the brightness value of real border position.In order to obtain enough accurately initial boundary positions, reduction
The complexity of subsequent searches, the present invention is tentatively sentenced using the symmetry of the pixel intensity variation rule curve near border
Fixed estimation.
Specific practice is as follows：Using binaryzation border as central point, set length for 21 row to search window；So
This 21 pixels are investigated one by one afterwards, acquisition length is a 11 row pixel in each pixel；Then, this 11 are analyzed
The centre symmetry of pixel；The position for selecting symmetry best, is used as first positioning border.
Abnormality processing when 6a) just positioning
In the image of actual photographed, can be disturbed by various abnormal conditions --- because device between device and glass card
Part fillet and the gap occurred, or the noise that boundary perimeter occurs because of binaryzation.These disturbed conditions can all make side
The Primary Location on boundary runs into challenge.
The boundary alignment under disturbed condition is realized using following method：If row pixel intensity change is normal (only
Change in the presence of at one), then it is normal conditions, does not interfere with；If the row pixel intensity saltus step repeatedly, belongs to disturbed condition.
B) positioning at load-bearing interface
Load-bearing interface is the platform for placing detected device, is generally possible to rotate, its surface may be considered completely
Smooth.In order to obtain the elaborate position of device, it is necessary to obtain the coordinate at load-bearing interface in advance from first positioning boundary coordinate.For
Raising stability, reduces the interference of noise, the present invention using projection method, realizing the positioning at load-bearing interface.Meanwhile, it is
Prevent wiring inclination that may be present of weighing, in addition it is also necessary to image is rotated within the specific limits, then could project.
Specifically：
The possible scope in angle of inclination 1b) is set, the angle is traveled through.According to the angle value of setting, to binary map
As carrying out floor projection, drop shadow curve is obtained；
2b) to the drop shadow curve under various angles, analyzed, statistics projection value rises to picture traverse value N, institute from 0
The line number of experience is needed, the minimum corresponding angle value of situation of line number is selected as the inclination angle at load-bearing interface；
The corresponding drop shadow curve in inclination angle 3b) is found, the difference of two row projection values above and below Difference Calculation, calculating is carried out to it,
The a line for selecting difference value maximum, is used as the position at load-bearing interface；
4b) and then in difference curves, the corresponding value in load-bearing interface is filtered out, maximum of which difference value is searched again for, as
The first position location on top device border.
C) positioning of device right boundary
It is accurately positioned for the ease of follow-up sub-pixel is other, needs also exist for determining the right boundary of device.In order to reduce
The interference of noise, border is determined using upright projection.Comprise the following steps that：
1c) the top device boundary position obtained using preceding step, and load-bearing interface position, obtain the two point
Boundary line (the two vertical coordinate average value)；
2c) point that the coordinate value in all boundary points is located above line of demarcation is extracted；
3c) according to the point that is extracted, the leftmost side and the rightmost side are found；
4c) in the pixel coverage of 20, left-hand point periphery, upright projection is carried out to binary map；
Difference 5c) is carried out to drop shadow curve, the maximum position of difference value is found, is used as the left margin of device；
6c) same method, obtains the right margin of device；
D) positioning of top device coordinate fillet
Because the corner of some devices is fillet, the coordinate value of its boundary point is lower than normal top boundary point.Such as
Fruit can not be accurately positioned to fillet position, then follow-up accurate thickness measure can be impacted.The present invention is using as follows
Method setting circle angular region：
, will 1d) using the load-bearing areal coordinate, top device coordinate, device right boundary position that above several steps are obtained
The corresponding sub- contours extract of device comes out；
The sub- profile of device 2d) is directed to, its corresponding convex closure is calculated；
The convex closure 3d) is utilized, new bianry image is generated, convex closure area filling is white；
4d) convex closure area image makes the difference with original device image；
5d) in differential chart, four maximum connected regions of area are navigated to；
The position for obtaining four connected regions 6d) is calculated, and then reckoning obtains device fillet position.
2. sub-pix boundary alignment
In the image that the industrial camera of high definition is shot, the border of device can form in brightness and a kind of slowly decline
Stairstepping.This shape and the shape of arc tangent curve are closely similar.The reason for this shape occur, is that industrial camera is being caught
When obtaining image, due to data sampling precision and camera lens definition the problems such as, cause the border of device air can not be completely
Separate, so as to there occurs a certain degree of blooming.The usual this border with certain transition effect is, it is necessary to using sub- picture
Plain boundary alignment technology, otherwise can influence the measurement of device size.
Traditional sub-pix boundary alignment technology, gray scale Moment Methods etc., amount of calculation is all larger, have impact on the reality of algorithm
Shi Xingneng.The present invention to arc tangent curve with the row pixel curve near boundary position by being fitted, to realize border
Sub-pixel positioning.The mathematic(al) representation of arc tangent curve is as follows：
Y=p1*atan (p2*x+p3)+p4 (1)
It is assumed that the column direction brightness of image curve near boundary position, is also the shape for meeting arc tangent curve.Therefore
Formula (1) need to be only used, is solved when four parameters of the brightness curve in prostatitis：P1, p2, p3, p4, you can obtain the brightness when prostatitis
The mathematical models of curve.Then, accurately sub-pix boundary position, is exactly the center position of arc tangent curve, can be with
Obtained using following formula：
X0=-p3/p2 (2)
According to formula (2), the subpixel coordinates for obtaining each row will soon be calculated.
In order to improve the speed of solution, the present invention does not recycle such as LM etc iterative method, and directly uses
Particle swarm optimization algorithm (PSO) carries out fast search, it is achieved thereby that the extensive speed-raising of solution procedure.
Sub-pixel positioning process flow is as follows：
A) the direct valuation of two parameters of p1 and p4
A total of four parameters of arc tangent curve, in order to reduce searching times, the shape of brightness curve can be utilized in advance,
Two parameters of p1 and p4 are calculated, so can subsequently the two parameter values calculated be regard as the initial of p1 and p4
Value.
P1 calculating --- the physical quantity of p1 descriptions is exactly the coverage of arc tangent curve in vertical direction in fact.Cause
Then the estimation of this value, can make the difference and obtain by calculating the average brightness of boundaries on either side.But, calculating, both sides are bright
, it is necessary to which luminance transition region is filtered out during degree average value, in order to avoid interfere.Specific practice is as follows：
1a) since above initial boundary position, it will be gradually added to successively when the pixel in prostatitis in set.If collection
Pixel intensity standard deviation in conjunction is less than 5, then can continue to increase new pixel.If greater than 5, then stop the new picture of increase
Element.The average value of the pixel intensity in now gathering is calculated, the mean flow rate on the upside of this is used as.
2a) equally, since below initial boundary position, same operation is also carried out, the mean flow rate of downside is obtained.
The difference of two brightness values, is exactly p1.
3a) p4 calculating --- p4 is exactly the ordinate of boundary curve center position in fact.The calculating of the value is simpler
It is single, it is only necessary to which that pixel brightness value all on brightness curve is calculated into average value.
B) the arc tangent curve based on particle cluster algorithm (PSO) is solved
Arc tangent curve model in formula (1) includes 4 parameters, it is impossible to accurately solved using the mode solved equation,
Therefore conventional method is to utilize the alternative manners, the preferable result of Step wise approximation such as gradient descent method.These methods are often
Searching times are more, and are easily trapped into local optimum.The problem of in order to solve speed and local optimum simultaneously, the present invention is used
Optimized search strategy (PSO algorithms), to carry out the solution of arc tangent curve model.
Particle cluster algorithm (PSO) is a kind of classical Evolutionary Computation.The algorithm simulation rule of birds search of food
Rule.Bevy is in some range searching food, and they have respective position and speed.If a certain bird have found food,
Other birds on periphery can then refer to its direction, and the heading of oneself is adjusted.Meanwhile, each bird also has certain
Memory, up to the present they can remember, once the most place of food, its fly direction, also join to a certain extent
Examine the position.So, after a small number of iteration several times, the overall fitness of flock of birds, it is possible to reach a higher value, from
And obtain a preferably solution.
Specific way using the parameter of PSO Algorithm for Solving arc tangent curves is as follows：
One group of particle 1b) is set, 20 altogether, each particle includes 4 parameters, and initial value is set at random, represents anyway
Cut a solution of curve；
The curve corresponding to current 20 particles 2b) is calculated, error is negated, made by the error with actual brightness curve
For fitness；
Current most suitable parameter of curve 3b) is found, the direction and position of each particle is then adjusted, continues to calculate；
Above-mentioned two step 4b) is repeated, untill the error of best particle is less than threshold value 10；
5b) 4 parameters of now best particle, are exactly the parameter of the arc tangent curve finally needed, are counted according to formula (2)
Calculation obtains accurate sub-pix boundary position.
C) application of previous column result, obtains p2 and p3 initial value
, can be by the p2 in the parameter of the row and p3 value, under after completing to specify the curve matching of a certain row
The parameter of the brightness curve of one row.And p1 and p4 are still estimated by calculating.It can so currently to be listed in and be tied
When fruit is searched for, initial value at the beginning is exactly relatively actual position, so can further reduce volumes of searches.
D) solution of noise jamming problem
The presence of noise, can cause the shape of brightness curve and conventional arc tangent curve, and difference is very big.Now directly cover
Solved with arc tangent curve model, final result and the situation without noise jamming are differed also very big.The present invention passes through
The error in fit procedure is judged, to determine when whether the fitting result in prostatitis is correct.For the situation that error is larger, the row are not
Enter the precise positioning of row bound, but the later stage is repaired by edge smoothing.
E) edge smoothing
To the border subpixel coordinates value tentatively obtained, the coordinate between adjacent column carries out smooth operation, to reduce noise
Interference, or the random deviation in image sampling process, caused border jitter phenomenon.If do not put down to boundary coordinate
It is sliding, then process may be filtered out to follow-up dust and impacted, cause many real dusts can not be detected, so as to influence
Measurement result.
The present invention uses Bezier approximating method, to complete the fitting of border point coordinates.
3. abnormal conditions processing
A) there is no device early warning
Detect with the presence or absence of the tactful fairly simple of device in the visual field, it should be noted that the time of saving, improves efficiency of algorithm.
The present invention is adopted with the following method, whether there is device in the fast verification visual field：
During in the absence of device, glass interface is straight line in image.By analyzing the vertical seat on glass interface
Mark can be detected.By the coordinate of interface, it is fitted in alignment.Then each point is calculated to the distance of straight line.Such as
Fruit arrives a little the distance of straight line, all sufficiently small, then without device.
For there is no the situation of device presence in image, it is necessary to carry out early warning, prevent measurement result from occurring abnormal.
B) multiple device early warning
Due to system mechanics part exception that may be present, causing may be while there are multiple devices in image.This feelings
The necessary early warning of condition, prevents measurement result from occurring abnormal.Handle in two kinds of situation：
1b) polylith device is separated from each other
The number of times of the raising and lowering of boundary coordinate is investigated, is above 1 time if up number of times and decline number of times, then can
To understand early warning.
2b) polylith device occurs overlapping
After the rule using device coordinate raising and lowering, the right boundary for obtaining device, it is necessary to investigate left and right
The thickness distribution of device between border.Because the thickness of device is all fixed, therefore empirical value can be set.On left and right side
Between boundary, if the thickness value of some position, near or above the height of double thickness, then illustrate to there occurs multiple devices
It is overlapping.At this time, it may be necessary to carry out early warning.
C) gap early warning
The larger dust of particle is there may be on glass plate, if now device is located on dust by chance, device and glass
Gap just occurs between glass disk.Because industrial detection permissible accuracy is in micron order, this gap is enough to cause thickness of detector
Detection is inaccurate.Therefore early warning must be carried out to such case.
Adopt the presence with the following method to detect gap：
1c) the contours extract of device area
During the Primary Location of device boundaries, the approximate location of the right boundary of device is had been obtained for.According to this
Individual position, the corresponding binaryzation subgraph of device is intercepted out, profile is then extracted.Subsequent step is by the basis of profile
On, carry out the detection in gap.
2c) top device and left and right sides profile are filtered out
Using the load-bearing interface coordinate obtained in first position fixing process, and device right boundary coordinate, it can substantially obtain
The lower left corner of device and bottom right angular coordinate.Then according to the two coordinate values, by the top profile, left side profile, right side of device
Profile is all filtered out, and leaves behind the profile point of possible gap area.
3c) gap area dimensional measurement
By the processing of above two steps, the straightway got off is now retained, it is possible to be considered that device is put down with glass
The profile corresponding to gap between face.The profile point of analysis now, calculates the border in the region.If the height in the region is small
In 5 pixels, and width is more than 20 pixels, then it is assumed that be gap.Now need early warning.
D) filtering of dust interference
Positioning of the dust to sub-pix border, influences very big, if dust can not accurately be filtered, whole device is thick
Degree detecting system can not obtain correct measurement result.Especially China of today, atmosphere pollution is serious, particulate in air ratio
Very high, interference of the dust to industrial detection system of weight, very generally.
The present invention carries out filtering dust according to following strategy：
1d) to sub-pix boundary coordinate, local fitting a straight line is carried out：Increased using side, the mode of side search is exploratory
Boundary point is divided into multiple writings to close：
First, the profile point of most coordinate is selected, set is added.Then, the new abutment points in right side are added successively and gathered.Often
A point is added, a straight line is just fitted, then arrives the distance of the straight line in set of computations a little.If average distance is more than
Threshold value (1 pixel), then stop increase point, and straight line subsegment fitting is completed.Follow-up profile point is added in new set.
Distance 2d) is more than to the coordinate points of threshold value, picked out；
In each subset, the coordinate points that profile distance between beeline and dot is more than threshold value are picked out.
3d) for the coordinate points for being selected out, it is segmented；
The coordinate points for being selected out in analysis subset, according to whether connection, is divided into multiple subsegments.
The width of each subsegment 4d) is investigated, width is less than threshold value, and all can be regarded as is dust；
Coordinate points 5d) are recalculated, the corresponding subsegment of dust is filtered out, by the coordinate points in the subsegment, all with corresponding
Coordinate points on the straight line of fitting are substituted.
E) early warning of conditions of streaking
Putting position of the device on glass plate, is influenceed by plant equipment, does not ensure that the side of device each time
To vertical all with camera optical axis direction.When device direction and camera optical axis direction are present compared with mitre, due to Jiao of video camera
Away from smaller, the device end away from camera, it is impossible to correct to focus on, and at the end of device image, there is a brightness
Higher gray area.The thickness measure that the region directly results in device is gone wrong, it is therefore necessary to which such case is examined
Survey, and carry out early warning.
Using following flow, to realize the early warning of conditions of streaking：
1e) the judgement of top device border vertical coordinate uniformity
During generation conditions of streaking, because the pixel shared by the vertical direction of device distal end in the picture is less, therefore it is pushed up
Portion's boundary coordinate, the upright position than other top boundary coordinates on device is relatively low.This characteristic can be utilized, to conditions of streaking
Carry out preliminary early warning.Specific practice is as follows：
In the left and right ends of device, the region that each one length of selection is 20.Then in the center of device, selection
One length is 50 region, and the average value of the top profile point vertical coordinate of the device in these three regions is calculated respectively.
If there is a region among left and right ends, its average coordinates is compared with the average coordinates of central area, and difference is more than
5, then it is assumed that there is conditions of streaking, early warning.
2e) device center region, is analyzed with the brightness uniformity of two end regions of head and the tail
In the same manner as above, in the left and right ends of device, the region that each one length of selection is 20.Then in device
Heart position, one length of selection is 50 region, and the mean flow rate in these three regions is calculated respectively.
If there is a region among left and right ends, its brightness and central area mean flow rate, difference are more than 20, then it is assumed that
There is conditions of streaking, early warning.
Claims (9)
1. the subpixel accuracy thickness of detector detection method of a kind of fast and stable, it is characterised in that comprise the following steps：
1) border Primary Location：Two regions of target and air are divided the image into, the line of demarcation in the two regions is then obtained, made
For preliminary boundary alignment result；
2) sub-pix precise positioning：Using the method for arc tangent curve matching, solved using particle swarm optimization algorithm, for
The positioning result of each row, then be fitted using Bezier；
3) abnormal conditions are handled.
2. a kind of subpixel accuracy thickness of detector detection method of fast and stable according to claim 1, its feature exists
In：The deviation of the first position location of border Primary Location and final sub-pix scenic spot coordinate is within the distance of 1 pixel.
3. a kind of subpixel accuracy thickness of detector detection method of fast and stable according to claim 2, its feature exists
In described border Primary Location comprises the following steps：
A) by the thick border Primary Location to essence
The Primary Location coordinate points of interface are obtained according to following flow：
1a) image down：Image is first reduced, a quarter of original size is narrowed down to；
2a) extract target area reference brightness value, air section reference brightness value：Precalculate in image, target area and sky
The brightness reference value in gas region, using clustering method, to all pixels, according to brightness value, is divided into two major classes, then calculates every
The average brightness of one class, respectively as object brightness and the reference value of air section brightness；
3a) binarization operation：After the reference brightness value of target area and air section is obtained, using the average of the two, make
For the binary-state threshold of entire image, binarization operation is carried out to all pixels；
4a) extract largest connected region：Extract the largest connected region in binarization operation result；
5a) the preliminary boundary alignment based on symmetry；Using the symmetry of the pixel intensity variation rule curve near border come
Obtain preliminary judgement；
Abnormality processing when 6a) just positioning；
B) positioning at load-bearing interface
The scope at angle of inclination 1b) is set, the angle is traveled through, according to the angle value of setting, water-filling is entered to bianry image
Flat projection, obtains drop shadow curve；
2b) to the drop shadow curve under various angles, analyzed, statistics projection value rises to picture traverse value N from 0, required
The line number of experience, selects the minimum corresponding angle value of situation of line number as the inclination angle at load-bearing interface；
The corresponding drop shadow curve in inclination angle 3b) is found, the difference of two row projection values above and below Difference Calculation, calculating is carried out to it, is selected
The maximum a line of difference value, is used as the position at load-bearing interface；
4b) and then in difference curves, the corresponding value in load-bearing interface is filtered out, maximum of which difference value is searched again for, is used as device
The first position location of top boundary；
C) positioning of device right boundary：
1c) the top device boundary position obtained using preceding step, and load-bearing interface position, obtain the boundary of the two
Line；
2c) point that the coordinate value in all boundary points is located above line of demarcation is extracted；
3c) according to the point that is extracted, the leftmost side and the rightmost side are found；
4c) in the pixel coverage of 20, left-hand point periphery, upright projection is carried out to binary map；
Difference 5c) is carried out to drop shadow curve, the maximum position of difference value is found, is used as the left margin of device；
6c) in the pixel coverage of 20, right-hand point periphery, upright projection is carried out to binary map, difference is carried out to drop shadow curve, looked for
To the position that difference value is maximum, the right margin of device is obtained；
D) positioning of top device coordinate fillet
The load-bearing areal coordinate that 1d) is obtained using preceding step, top device coordinate, device right boundary position, by device correspondence
Sub- contours extract come out；
The sub- profile of device 2d) is directed to, its corresponding convex closure is calculated；
The convex closure 3d) is utilized, new bianry image is generated, convex closure area filling is white；
4d) convex closure area image makes the difference with original device image；
5d) in differential chart, four maximum connected regions of area are navigated to；
The position for obtaining four connected regions 6d) is calculated, device fillet position is obtained.
4. a kind of subpixel accuracy thickness of detector detection method of fast and stable according to claim 1, its feature exists
In described sub-pix precise positioning to arc tangent curve with the row pixel curve near boundary position by being fitted, instead
The mathematic(al) representation of tangent cutve is：Y=p1*atan (p2*x+p3)+p4, is obtained when the accurate mathematical of the brightness curve in prostatitis
Model, then, is obtained using following formula：X0=-p3/p2, calculates the subpixel coordinates for obtaining each row.
5. a kind of subpixel accuracy thickness of detector detection method of fast and stable according to claim 4, its feature exists
In described sub-pix precise positioning comprises the following steps：
A) the direct valuation of two parameters of p1 and p4：The shape of brightness curve is utilized in advance, and two parameters of p1 and p4 are calculated,
The two parameter values calculated again as p1 and p4 initial value：
According to following flow：
1a) since above initial boundary position, it will be gradually added to when the pixel in prostatitis in set, if the picture in set
Plain luminance standard difference is less than 5, then continues to increase new pixel, if greater than 5, then stops the new pixel of increase, calculating now collects
The average value of pixel intensity in conjunction, is used as the mean flow rate on the upside of this；
2a) since below initial boundary position, it will be gradually added to when the pixel in prostatitis in set, if the picture in set
Plain luminance standard difference is less than 5, then continues to increase new pixel, if greater than 5, then stops the new pixel of increase, calculating now collects
The average value of pixel intensity in conjunction, obtains the mean flow rate on the downside of this, the mean flow rate on the upside of this with it is average bright on the downside of this
The difference of degree is p1；
3a) p4 calculating：Pixel brightness value all on brightness curve is calculated into average value, boundary curve center position is obtained
Ordinate；
B) the arc tangent curve based on particle cluster algorithm is solved：
According to following flow：
One group of particle 1b) is set, 20 altogether, each particle includes 4 parameters, and initial value is set at random, represents arc tangent bent
One solution of line；
The curve corresponding to current 20 particles 2b) is calculated, the error with actual brightness curve negates error, as suitable
Response；
Current most suitable parameter of curve 3b) is found, the direction and position of each particle is then adjusted, continues to calculate；
4b) repeat 2b), 3b) two steps, untill the error of best particle is less than threshold value 10；
5b) calculated according to formula x0=-p3/p2 and obtain accurate sub-pix boundary position；
C) application of previous column result, obtains p2 and p3 initial value：
After a specified row curve matching is completed, by the p2 and p3 value in the parameter of the row, it is used as the brightness of next column bent
The parameter of line, p1 and p4 are still estimated by calculating；
D) solution of noise jamming：
By judging the error in fit procedure, to determine when whether the fitting result in prostatitis is correct, for the feelings that error is larger
Shape, the row, without the precise positioning on border, are to be repaired by edge smoothing in the later stage；
E) edge smoothing：
Using Bezier approximating method, the fitting of border point coordinates is completed.
6. a kind of subpixel accuracy thickness of detector detection method of fast and stable according to claim 1, its feature exists
In described abnormal conditions processing includes
A) there is no device early warning：Detected by the vertical coordinate analyzed on glass interface, by the coordinate of interface, fitting
It is in alignment, each point is then calculated to the distance of straight line, if arriving the distance of straight line a little, both less than 3 pixels,
Then without device, for there is no the situation that device is present in image, it is necessary to carry out early warning；
B) multiple device early warning：Including：
1b) polylith device is separated from each other：1 time, early warning are above if up number of times and decline number of times；
2b) polylith device occurs overlapping：Using device coordinate raising and lowering rule, obtain device right boundary it
Afterwards, the thickness distribution of the device between right boundary is investigated, between right boundary, if the thickness value of some position, etc.
In or more than double thickness height, then explanation there occurs the overlapping of multiple devices, carry out early warning；
C) gap early warning：
1c) the contours extract of device area：During the Primary Location of device boundaries, the right boundary of device is had been obtained for
Position, according to this position, the corresponding binaryzation subgraph of device is intercepted out, then extracts profile, subsequent step will
On the basis of profile, the detection in gap is carried out；
2c) top device and left and right sides profile are filtered out：Utilize the load-bearing interface coordinate obtained in first position fixing process, Yi Jiqi
Part right boundary coordinate, obtains the lower left corner and the bottom right angular coordinate of device, then according to the two coordinate values, by the top of device
Profile, left side profile, right lateral contours are all filtered out, and leave behind the profile point of gap area；
3c) gap area dimensional measurement：By above 1c), 2c) two steps processing, now retain the straightway got off, be
The profile corresponding to gap between device and glass planar, the profile point of analysis now, calculates the border in the region, if should
The height in region is less than 5 pixels, and width is more than 20 pixels, then is gap, it is necessary to early warning；
D) filtering of dust interference：
1d) to sub-pix boundary coordinate, local fitting a straight line is carried out：Increased using side, the mode of side search is exploratory by side
Boundary's point is divided into multiple writings and closed；
Distance 2d) is more than to the coordinate points of threshold value, picked out；In each subset, profile distance between beeline and dot is more than
The coordinate points of threshold value are picked out；
3d) for the coordinate points for being selected out, it is segmented；The coordinate points for being selected out in analysis subset, according to whether
Connection, is divided into multiple subsegments；
The width of each subsegment 4d) is investigated, width is less than threshold value, and all can be regarded as is dust；
5d) recalculate coordinate points：The corresponding subsegment of dust is filtered out, by the coordinate points in the subsegment, all with corresponding fitting
Straight line on coordinate points substitute；
E) early warning of conditions of streaking
1e) the judgement of top device border vertical coordinate uniformity：In the left and right ends of device, each one length of selection is 20
Region, then in the center of device, selection one length be 50 region, the device in these three regions is calculated respectively
The average value of top profile point vertical coordinate, if there is a region among left and right ends, its average coordinates is put down with central area
Equal coordinate is compared, and difference is more than 5, then there is conditions of streaking, early warning；
2e) device center region, is analyzed with the brightness uniformity of two end regions of head and the tail：With 1e) equally, in the left and right two of device
End, the region that each one length of selection is 20, then in the center of device, one length of selection is 50 region, point
The mean flow rate in these three regions is not calculated, if there is a region among left and right ends, its brightness is average bright with central area
Degree, difference is more than 20, then there is conditions of streaking, early warning.
7. a kind of subpixel accuracy thickness of detector detection method of fast and stable according to claim 3, its feature exists
In described step 5a) in using binaryzation border as central point, set length for 21 row to search window；Then
This 21 pixels are investigated one by one, acquisition length is a 11 row pixel in each pixel；Then, this 11 pictures are analyzed
The centre symmetry of element；The position for selecting symmetry best, is used as first positioning border.
8. a kind of subpixel accuracy thickness of detector detection method of fast and stable according to claim 3, its feature exists
In described step 6a) in if row pixel intensity change is normal, then be normal conditions, do not interfere with；If the row pixel
Brightness saltus step repeatedly, then belongs to disturbed condition.
9. a kind of subpixel accuracy thickness of detector detection method of fast and stable according to claim 6, its feature exists
In described step 1d) in exploratory boundary point be divided into multiple writings closed：First, the profile point of most coordinate is selected, is added
Set；Then, the new abutment points in right side are added successively and gathered, often added a point, be just fitted a straight line, then calculate collection
The distance of the straight line is arrived in conjunction a little, if average distance is more than threshold value, stops increase point, the straight line subsegment has been fitted
Into follow-up profile point is added in new set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710295849.XA CN107218894B (en) | 2017-04-28 | 2017-04-28 | Rapid and stable sub-pixel precision device thickness detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710295849.XA CN107218894B (en) | 2017-04-28 | 2017-04-28 | Rapid and stable sub-pixel precision device thickness detection method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107218894A true CN107218894A (en) | 2017-09-29 |
| CN107218894B CN107218894B (en) | 2020-02-18 |
Family
ID=59943866
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710295849.XA Expired - Fee Related CN107218894B (en) | 2017-04-28 | 2017-04-28 | Rapid and stable sub-pixel precision device thickness detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107218894B (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108154506A (en) * | 2017-12-27 | 2018-06-12 | 合肥市雅视智能科技有限公司 | A kind of hex nut automatic testing method |
| CN109115127A (en) * | 2018-08-24 | 2019-01-01 | 合肥工业大学 | A kind of sub-pix peak point extraction algorithm based on Bezier |
| CN111242966A (en) * | 2020-01-15 | 2020-06-05 | 北京大豪科技股份有限公司 | Image boundary correction method, device, electronic equipment and storage medium |
| CN111492200A (en) * | 2020-03-17 | 2020-08-04 | 长江存储科技有限责任公司 | Method and system for semiconductor structure thickness measurement |
| CN112818486A (en) * | 2021-02-02 | 2021-05-18 | 上汽大众汽车有限公司 | Hole position arrangement method for loudspeaker sound outlet holes |
| CN113536978A (en) * | 2021-06-28 | 2021-10-22 | 杭州电子科技大学 | Method for detecting disguised target based on significance |
| CN113804115A (en) * | 2021-08-10 | 2021-12-17 | 天津大学 | Concrete pouring base layer thickness monitoring method based on clustering |
| CN114687079A (en) * | 2020-12-31 | 2022-07-01 | 杰克科技股份有限公司 | System and method for managing device parameters of sewing device and computer storage medium |
| CN115841497A (en) * | 2023-02-21 | 2023-03-24 | 广州鲁邦通物联网科技股份有限公司 | Boundary detection method and escalator area intrusion detection method and system |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP5320216B2 (en) * | 2009-08-26 | 2013-10-23 | パナソニック株式会社 | Image processing apparatus, image processing system, and image processing method |
| CN104268857A (en) * | 2014-09-16 | 2015-01-07 | 湖南大学 | Rapid sub pixel edge detection and locating method based on machine vision |
| CN104359403A (en) * | 2014-11-21 | 2015-02-18 | 天津工业大学 | Plane part size measurement method based on sub-pixel edge algorithm |
| CN104573698A (en) * | 2014-12-31 | 2015-04-29 | 河海大学 | Image texture main direction recognition method oriented to imaging measurement |
| JP2015106095A (en) * | 2013-11-29 | 2015-06-08 | 三星ディスプレイ株式會社Ｓａｍｓｕｎｇ Ｄｉｓｐｌａｙ Ｃｏ．，Ｌｔｄ． | Image processing apparatus, image processing method, and computer program |
| CN104715487A (en) * | 2015-04-01 | 2015-06-17 | 大连理工大学 | Method for sub-pixel edge detection based on pseudo Zernike moments |
-
2017
- 2017-04-28 CN CN201710295849.XA patent/CN107218894B/en not_active Expired - Fee Related
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP5320216B2 (en) * | 2009-08-26 | 2013-10-23 | パナソニック株式会社 | Image processing apparatus, image processing system, and image processing method |
| JP2015106095A (en) * | 2013-11-29 | 2015-06-08 | 三星ディスプレイ株式會社Ｓａｍｓｕｎｇ Ｄｉｓｐｌａｙ Ｃｏ．，Ｌｔｄ． | Image processing apparatus, image processing method, and computer program |
| CN104268857A (en) * | 2014-09-16 | 2015-01-07 | 湖南大学 | Rapid sub pixel edge detection and locating method based on machine vision |
| CN104359403A (en) * | 2014-11-21 | 2015-02-18 | 天津工业大学 | Plane part size measurement method based on sub-pixel edge algorithm |
| CN104573698A (en) * | 2014-12-31 | 2015-04-29 | 河海大学 | Image texture main direction recognition method oriented to imaging measurement |
| CN104715487A (en) * | 2015-04-01 | 2015-06-17 | 大连理工大学 | Method for sub-pixel edge detection based on pseudo Zernike moments |
Non-Patent Citations (2)
| Title |
|---|
| 孙秋成等: "一种亚像素精度的边缘检测方法", 《北京工业大学学报》 * |
| 孙维健: "基于反正切函数拟合的亚像素边缘检测方法", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108154506A (en) * | 2017-12-27 | 2018-06-12 | 合肥市雅视智能科技有限公司 | A kind of hex nut automatic testing method |
| CN109115127A (en) * | 2018-08-24 | 2019-01-01 | 合肥工业大学 | A kind of sub-pix peak point extraction algorithm based on Bezier |
| CN111242966A (en) * | 2020-01-15 | 2020-06-05 | 北京大豪科技股份有限公司 | Image boundary correction method, device, electronic equipment and storage medium |
| TWI747283B (en) * | 2020-03-17 | 2021-11-21 | 大陸商長江存儲科技有限責任公司 | Method and system for semiconductor structure thickness measurement |
| WO2021184175A1 (en) * | 2020-03-17 | 2021-09-23 | Yangtze Memory Technologies Co., Ltd. | Methods and systems for semiconductor structure thickness measurement |
| CN111492200B (en) * | 2020-03-17 | 2021-05-14 | 长江存储科技有限责任公司 | Method and system for semiconductor structure thickness measurement |
| CN111492200A (en) * | 2020-03-17 | 2020-08-04 | 长江存储科技有限责任公司 | Method and system for semiconductor structure thickness measurement |
| US11270428B2 (en) | 2020-03-17 | 2022-03-08 | Yangtze Memory Technologies Co., Ltd. | Methods and systems for semiconductor structure thickness measurement |
| CN114687079A (en) * | 2020-12-31 | 2022-07-01 | 杰克科技股份有限公司 | System and method for managing device parameters of sewing device and computer storage medium |
| WO2022143032A1 (en) * | 2020-12-31 | 2022-07-07 | 杰克科技股份有限公司 | Device parameter management system and method for sewing device, and computer storage medium |
| CN112818486A (en) * | 2021-02-02 | 2021-05-18 | 上汽大众汽车有限公司 | Hole position arrangement method for loudspeaker sound outlet holes |
| CN113536978A (en) * | 2021-06-28 | 2021-10-22 | 杭州电子科技大学 | Method for detecting disguised target based on significance |
| CN113536978B (en) * | 2021-06-28 | 2023-08-18 | 杭州电子科技大学 | Camouflage target detection method based on saliency |
| CN113804115A (en) * | 2021-08-10 | 2021-12-17 | 天津大学 | Concrete pouring base layer thickness monitoring method based on clustering |
| CN113804115B (en) * | 2021-08-10 | 2022-08-30 | 天津大学 | Concrete pouring base layer thickness monitoring method based on clustering |
| CN115841497A (en) * | 2023-02-21 | 2023-03-24 | 广州鲁邦通物联网科技股份有限公司 | Boundary detection method and escalator area intrusion detection method and system |
| CN115841497B (en) * | 2023-02-21 | 2023-04-18 | 广州鲁邦通物联网科技股份有限公司 | Boundary detection method and escalator area intrusion detection method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107218894B (en) | 2020-02-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107218894A (en) | A kind of subpixel accuracy thickness of detector detection method of fast and stable | |
| CN109903312B (en) | Football player running distance statistical method based on video multi-target tracking | |
| CN104981105B (en) | A kind of quickly accurate detection and method for correcting error for obtaining element central and deflection angle | |
| CN110287932B (en) | Road blocking information extraction method based on deep learning image semantic segmentation | |
| CN107145908B (en) | A kind of small target detecting method based on R-FCN | |
| CN110533722A (en) | A kind of the robot fast relocation method and system of view-based access control model dictionary | |
| CN111209892A (en) | Crowd density and quantity estimation method based on convolutional neural network | |
| CN107064170A (en) | One kind detection phone housing profile tolerance defect method | |
| CN106023298A (en) | Point cloud rigid registration method based on local Poisson curved surface reconstruction | |
| Liang et al. | Horizon detection from electro-optical sensors under maritime environment | |
| CN109671174A (en) | A kind of pylon method for inspecting and device | |
| CN109711322A (en) | A kind of people's vehicle separation method based on RFCN | |
| CN113129268B (en) | Quality detection method for riveting pier head of airplane | |
| CN108985337A (en) | A kind of product surface scratch detection method based on picture depth study | |
| CN112053317A (en) | Workpiece surface defect detection method based on cascade neural network | |
| TWI765442B (en) | Method for defect level determination and computer readable storage medium thereof | |
| CN115063579B (en) | Train positioning pin looseness detection method based on two-dimensional image and three-dimensional point cloud projection | |
| CN109345513A (en) | A kind of cigarette package defect inspection method with cigarette packet Attitude Calculation | |
| WO2020093631A1 (en) | Antenna downtilt angle measurement method based on depth instance segmentation network | |
| CN101114337A (en) | Ground buildings recognition positioning method | |
| CN113435282A (en) | Unmanned aerial vehicle image ear recognition method based on deep learning | |
| CN114926407A (en) | Steel surface defect detection system based on deep learning | |
| CN112149495B (en) | Video key frame extraction method based on parallax tracking | |
| CN113313107A (en) | Intelligent detection and identification method for multiple types of diseases on cable surface of cable-stayed bridge | |
| CN113362274B (en) | Rainfall monitoring and calculating method |
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 | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20180112 Address after: No. 10 Xiang Pu Road, Feixi County, Hefei City, Anhui Applicant after: Hefei Yai Intelligent Technology Co., Ltd. Address before: No. 10 Xiang Pu Road, Hefei City, Feixi, Feixi, Anhui Applicant before: Hefei JAS vision Intelligent Technology Co., Ltd. |
|
| TA01 | Transfer of patent application right | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20200218 Termination date: 20210428 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |