CN105718929B - The quick round object localization method of high-precision and system under round-the-clock circumstances not known - Google Patents
The quick round object localization method of high-precision and system under round-the-clock circumstances not known Download PDFInfo
- Publication number
- CN105718929B CN105718929B CN201610039698.7A CN201610039698A CN105718929B CN 105718929 B CN105718929 B CN 105718929B CN 201610039698 A CN201610039698 A CN 201610039698A CN 105718929 B CN105718929 B CN 105718929B
- Authority
- CN
- China
- Prior art keywords
- target
- segmental arc
- area
- round
- region
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/22—Image preprocessing by selection of a specific region containing or referencing a pattern; Locating or processing of specific regions to guide the detection or recognition
- G06V10/23—Image preprocessing by selection of a specific region containing or referencing a pattern; Locating or processing of specific regions to guide the detection or recognition based on positionally close patterns or neighbourhood relationships
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/09—Recognition of logos
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to technical field of computer vision, the invention discloses a kind of quick round object localization methods of high-precision under round-the-clock circumstances not known, it specifically includes following step: acquiring image using high resolution camera, pyramid decomposition is carried out to image after acquiring image, initial location is carried out to complete object or pollution target on the low-resolution image on pyramid top；Then again by low-resolution image complete object or pollution target position map to high-definition picture；Complete object is calculated in the target mapping area of high-definition picture or pollutes the three-dimensional coordinate of the central point of target.By carrying out initial location to complete object or pollution target on low-resolution image, the mode then mapped can obtain faster locating speed, facilitate the realization of system.The invention also discloses the high-precision systems that quickly circular target positions under a kind of round-the-clock circumstances not known.
Description
Technical field
The present invention relates to technical field of machine vision, the quick round mesh of high-precision under specifically a kind of round-the-clock circumstances not known
Position method and system is demarcated, circular target identification and three-dimensional localization can be realized under the round-the-clock environment in field.
Background technique
In machine vision, area of pattern recognition, circle marker object is often stuck or is engraved on target surface, as object table
The a part in face.Since circle marker object has more prominent features, object itself, the knowledge to circle marker object are relatively identified
It can not be easier, is more stable.This method is widely used in puma manipulator to static or mobile target crawl field.?
Camera is loaded onto the end of manipulator, the circular target in image recognition scene shot by camera, and calculates in circular target
Three-dimensional coordinate of the heart in global coordinate system automatically grabs object with this value guidance manipulator.In the whole process,
The high-precision of circular target, quickly positioning are crucial.
Circular target identification and the method for positioning mainly have:
(1) template matching method.This method passes through the template for establishing circular target in advance, in the target image search and mould
Plate has the region of maximum similarity.Region using where this region as target.In general, being built to adapt to different environment
There are many vertical template needs.
(2) pass through the method for pattern drill.Using a large amount of circular targets characteristic statistics be distributed, to target identification system into
Row training.Using candidate target image as the input for training classifier, candidate target image is judged according to the output of classifier
It whether is real target.
It is two kinds of common circular target recognition methods above, after recognizing the region where circular target, according to hough
Detection method or centroid method calculate the center of circular target.
The effect defect of template matching method: false recognition rate is higher, and especially when circular target is contaminated, false recognition rate is more
Height, it is time-consuming.
Pass through the method for pattern drill: needs are in advance split target, and false recognition rate is higher, especially when round mesh
False recognition rate is higher when mark is contaminated, time-consuming.
High accuracy positioning is obtained using high-definition picture, due to image resolution ratio height, so that the processing time lengthens, it can not
Reach real-time processing.That is, existing high-precision location technique cannot balancing speed and the two indexs of precision well, often
High positioning accuracy is exchanged for sacrifice speed, or is sacrificed positioning accuracy and exchanged fast speed for.
With respect to circular target under indoor environment, there is the characteristics of circular target under round-the-clock circumstances not known: (1) background may be
Arbitrarily.(2) by illumination, block, pollute, shade, perspective deformation are influenced, circular target pattern form is changed.In view of with
Upper factor when using template matching method, is difficult to give accurate template, because the shape after target is contaminated is arbitrary.Cause
It is higher to will lead to false recognition rate using template matching method for this, and then directly results in positioning mistake.(3) when image resolution ratio compared with
Height, and when target is smaller, target area is searched for from piece image, needs to slide template in entire image, and time-consuming.
And the training of pattern classifier needs a large amount of sample, natural environment round-the-clock for field, it is desirable to obtain complete
Sample be nearly impossible.In training classifier, the different feature of sample, training result is different.It determines effective
Sample characteristics are a difficult points.Training sample usually requires manually to split target from image, when sample size is inclined
When big, labor workload will be larger.
The technological deficiency that high accuracy positioning is obtained using high-definition picture, in existing circular target location technology,
In order to obtain higher positioning accuracy, need using high-resolution camera.During target identification, need to the every of image
A region is identified, when image resolution ratio is excessively high, the region for needing to identify is increased, and time overhead is caused to increase.
Summary of the invention
For the above problem existing for recognition positioning method in the prior art, the invention discloses round-the-clock circumstances not knowns
Lower high-precision quickly round object localization method and system.
Technical scheme is as follows:
It is specific to wrap the invention discloses a kind of high-precision method that quickly circular target positions under round-the-clock circumstances not known
It includes following step: image is acquired using high resolution camera, pyramid decomposition is carried out to image after acquiring image, in pyramid
Initial location is carried out to complete object or pollution target on the low-resolution image on top；It then again will be on low-resolution image
Complete object or pollution target position map to high-definition picture；It is had been calculated in the target mapping area of high-definition picture
The three-dimensional coordinate of the central point of whole target or pollution target.By on low-resolution image to complete object or pollution target
Initial location is carried out, the mode then mapped can obtain faster locating speed, facilitate the realization of system.
Further, the above method further includes verifying to the circular target oriented, verification process include with
Under step: (1) calculate coordinates of original image coordinates system under central point (xc,yc) and its n*n neighborhood (n be space circular target map to
One third of the camera as the minor axis length of planar elliptical) gray average gn；(2) with point (xc,yc) centered on, along with x-axis
Angle is respectively 0 degree, 45 degree, 90 degree, 135 degree, 180 degree, 225 degree, 270 degree, (direction and angle can for 315 degree of eight directions
Flexibly to be set as needed) it scans for, behind the boundary for reaching roundel, 5 are continued searching further along the direction
(value is also possible to other values, such as 6,4 etc.) pixel, centered on the 5th pixel, calculating n*n neighborhood, (n is space circle
Shape target maps to camera as the one third of the minor axis length of planar elliptical) gray average, be denoted as gex1、gex2、gex3、
gex4、gex5、gex6、gex7、gex8；(3) d is calculatedi=gn-gexi, i=1~8；If all diSign symbol it is identical, then calculate
Central coordinate of circle out is not pseudo- round coordinate, otherwise is pseudo- coordinate.By the above method, pseudo- round coordinate is rejected, to improve
The precision of identification.
Further, the above-mentioned process that initial location is carried out to pollution target specifically: (1) obtain non-closing in picture
Curve is closed, curve L is denoted as；(2) to each point on curve, the difference along the direction x and y is calculated, is denoted as dx, dy respectively；From
One point starts, and the identical contiguous pixels of symbol are denoted as segmental arc ak, each akWith the symbol of itself, "+" or "-"；(3)
To each segmental arc akConvexity classify: segmental arc akIts boundary rectangle R will be divided to for upper and lower two parts, is calculated to top
The region area s divideduWith the region area s of downward partdIf su>sd, then segmental arc akBe it is downwardly convex, be denoted as "-"；Otherwise it is
It is upward convex, it is denoted as "+"；(4) quadrant classification, in first quartile, segmental arc side are carried out to segmental arc according to the direction of segmental arc and convexity
To for "-", convexity is "+"；In the second quadrant, segmental arc direction is "+", and convexity is "+"；In third quadrant, segmental arc direction
For "-", convexity is "-"；In fourth quadrant, segmental arc direction is "+", and convexity is "-"；(5) from first segmental arc a1Start to follow
Ring, in its neighborhood search segmental arc ai(i ≠ 1), segmental arc aiAnd a1Positioned at different quadrants, segmental arc ai(i ≠ 1) and segmental arc a1It constitutes and waits
Select elliptical two parts；(6) using the coordinate value of candidate oval each pixel as sample parameter, elliptic equation is substituted into, according to
Least square solves elliptic parameter, after acquiring elliptic parameter, establishes oval analytical expression, then by candidate oval each point
Coordinate value substitutes into elliptic equation, calculates residual error, and removal residual error is greater than 2 pixels (range, which can according to need, to be selected)
Pixel, the number of left point are n1；If n1 number be more than initial number of samples n0 80% (percentage can be according to need
Selected), then remaining point is re-used as new sample and recalculates elliptical parameter, until the calculating of each sample is residual
Difference is both less than 2 (range, which can according to need, to be selected) a pixels.By the above method, contaminated round mesh is identified
Mark, improves identification stationkeeping ability, convenient for users to use.
The invention also discloses the high-precision systems that quickly circular target positions under round-the-clock circumstances not known, specifically include
High resolution camera, picture breakdown module, target locating module, mapping block and center point calculation module；The high-resolution
Camera is for acquiring image；Described image decomposing module is used to carry out pyramid decomposition to acquired image, obtains low resolution
Rate image；The target locating module be used on the low-resolution image on pyramid top to complete object or pollution target into
Row initial location；The mapping block be used for by low-resolution image complete object or pollution target position map to high score
Resolution image；The center point calculation module be used in the target mapping area of high-definition picture calculate complete object or
Pollute the three-dimensional coordinate of the central point of target.
Further, above system further includes pseudo- round authentication module, the round removal of puppet for will identify that.To
Improve the precision of identification.
Further, above system further includes polluting round identification module, and the pollution circle identification module goes out for identification
Non-closed curve in picture carries out quadrant classification to segmental arc according to the direction of segmental arc and convexity, is sentenced according to the quadrant sorted out
It is disconnected whether to constitute an ellipse, to identify contaminated circle.
By using above technical solution, the invention has the benefit that by first being determined in low pixel picture
It (is the image of 1,000,000 pixels with original resolution that position, which re-maps positioning in high-precision picture and can obtain faster locating speed,
It is tested, using the mentioned method of the present invention, compared to handling on the original image, 1/3) speed be can be improved.By to puppet circle
Removal can obtain higher positioning accuracy.The present invention can also detect contaminated circular target.
Detailed description of the invention
Fig. 1 is overall implementation flow chart of the invention.
Fig. 2 is progress target identification flow chart on low-resolution image.
Fig. 3 is circular target identification and centralized positioning flow chart.
The circular target that Fig. 4 is identified by verifying whether be pseudo- target schematic diagram.
Fig. 5 is the identification of pollution circle and centralized positioning flow chart.
Fig. 6 is targeted integration flow chart on low-resolution image.
Fig. 7 maps to high-definition picture schematic diagram by low-resolution image for target.
Specific embodiment
With reference to the accompanying drawings of the specification, the specific embodiment that the present invention will be described in detail.
It is specific to walk the invention discloses a kind of high-precision method that quickly circular target positions under round-the-clock circumstances not known
Rapid is to acquire image with high resolution camera, pyramid decomposition is carried out to image after acquiring image, at low point of pyramid top
Initial location is carried out to complete object or pollution target on resolution image；Then again by low-resolution image complete object or
Pollution target position maps to high-definition picture；In the target mapping area of high-definition picture calculate complete object or
Pollute the center position of target.Complete object is carried out on low-resolution image and pollutes the identification of target；Target area by
Low-resolution image maps to high-definition picture；Accurate positioning of the target on high-definition picture, circular target central point
Three-dimensional computations.Using pyramid decomposition, and in top elder generation coarse localization, it is accurately positioned and calculates after then mapping, it is quick, high-precision
Degree orients target.
Implementation flow chart of the invention is as shown in Figure 1, steps are as follows: the first step, for the image (high resolution graphics of input
Picture) I0Pyramid decomposition is carried out, original image is decomposed into the image of different scale, with lowest resolution tomographic image IsAs under
One step low-resolution image to be treated.IsPicture size is generally high-definition picture I0The 1/4 of size.Second step, low
Image in different resolution IsThe upper general area O for completing circular target and polluting circle targets.If target can be found, third is executed
Step.Step 3: by low-resolution image IsOn general area OsMap to high-definition picture I0On, target is obtained original
Region O where go image0.4th step, in subgraph O0Upper detection and positioning complete object or defect target, obtain subgraph
O0Target's center's coordinate (x under coordinate systemc′,yc′)；5th step, by (xc′,yc') map under coordinates of original image coordinates system, it is denoted as
(xc,yc)。
The specific steps of image pyramid decomposition include: step 1: carrying out Gaussian smoothing to image, defining original image is
I0, then smoothed out image is I,For convolution symbol, f is Gaussian template, is defined as follows.
Step 2: sub-sampling being carried out to image, all even number row and columns is removed, obtains low-resolution image Is.This low resolution
The size of rate image is the 1/4 of original image size.
In low-resolution image IsThe upper process for carrying out target identification is as shown in Figure 2.The image of input has been carried out first
Full circle identification is estimated circular target region and is returned if success (circular target is not contaminated).If unsuccessful (round mesh
Mark is contaminated), it carries out polluting round identification, if pollution circle can be recognized, using the position where pollution circle as priori knowledge, estimate
Target area is counted and returned, otherwise is retracted.
The process of complete circle identification specifically includes following step: (1), such as use Da-Jin algorithm or other methods, it is right
Low-resolution image IsBinaryzation is carried out, image I is obtainedb.(2) to IbCanny edge detection is carried out, image I is obtainede.(3) it counts
Nomogram is as IeIn closed outline vector Counter, Counter (i) indicates i-th profile.(4) number of closed outline is
ncounter, started the cycle over from i=1, until i=ncounter；Judge whether Counter (i) is border circular areas in each circulation.
Judge whether Counter (i) is that border circular areas specifically includes following step: S41 calculates the face of Counter (i)
Product, is denoted as S；The length for calculating Counter (i), is denoted as L；Calculate the circularity m=4 π S/L of Counter (i)2；S42 judgement
Whether Counter (i) is circular target region: when meeting following 2 conditions, Counter (i) is circular target area
Domain, otherwise be not.2 conditions are as follows: Lmin≤L≤Lmax, mmin≤m≤mmax。Lmin, Lmax, mmin, mmaxFor length and circularity
Minimum and maximum value empirical value, the empirical value can according to need carry out adaptability adjustment.
Circular target location of the core includes the following steps: that the coordinate of each member vegetarian refreshments in Counter (i) vector is
(xi,yi), i ∈ [1n], n are number a little.The image of circular target on the image is ellipse, elliptical general expression are as follows:
ax2+bxy+cy2+ dx+ey+f=0.By (xi,yi) oval general expression is substituted into, calculated by least square method (a, b, c,
D, e, f) value, elliptical center calculation formula are as follows: (b2-4ac≠0)。
The above method further includes whether the identified circular target of verifying is pseudo- target.The target circle identified such as Fig. 4 institute
Show.(xc,yc) it is the target circle center of circle calculated, verification process includes the following steps: that (1) calculates (xc,yc) and its n*n (ratio
Such as 7*7) the gray average g of neighborhoodin, n be space circular target map to camera as planar elliptical minor axis length three/
One.(2) with point (xc,yc) centered on, it is respectively 0 degree, 45 degree, 90 degree, 135 degree, 180 degree, 225 degree, 270 along with x-axis angle
Degree, 315 degree of eight directions scan for, and behind the boundary for reaching roundel, continue searching 5 pixels further along the direction, with
(the p in Fig. 4 centered on 5th pixel1-p8), (n is that space circular target maps to camera to the gray average of calculating n*n neighborhood
As the one third of the minor axis length of planar elliptical), it is denoted as gex1、gex2、gex3、gex4、gex5、gex6、gex7、gex8.Calculate di=
gin-gexi, i=1~8；If all diSign symbol it is identical, then calculated central coordinate of circle is not pseudo- round coordinate, on the contrary
It is pseudo- coordinate.
The definition of pollution circle refers in field environment, and due to blocking, shade, the profile of circular target is contaminated, no
It is a circle again；But there are partially visible circular arcs.The identification of pollution circle is exactly to pass through identification circular arc, goes to speculate by circular arc round
Target position.Algorithm flow is as shown in Figure 5.
The process of identification pollution circle includes the following steps: that (1) judges whether current connected curve is closed curve: from institute
Closed curve is removed in some curves, remaining is non-closed curve.(2) segmental arc direction is calculated: if it is non-closed curve,
Note curve is L；To each point on curve, the difference along the direction x and y is calculated, is denoted as dx, dy respectively；It is opened from first point of L
Beginning is counted, and dx (x, y) × identical contiguous pixels of dy (x, y) symbol are denoted as segmental arc ak, each akWith the symbol of itself
Number, "+" or "-".It is assumed that this section of arc has 10 points, the direction dx × dy of the 1-5 point is positive, the direction dx × dy of the 6th point
It is negative, the direction of the 7-10 point is positive；Then the 1-5 point is connected, becomes one section of arc；6th point is also one section of arc,
Only only has an isolated point；The 7-10 point is connected as one section of arc.(3) pseudo-ellipse segmental arc is removed: removal principle:
(1) segmental arc a is calculatedkLength, when length (this value is empirical value, may be configured as other different values) a pixel less than 20,
The segmental arc is removed, to avoid the interference of noise.(2) a is calculatedkBoundary rectangle R, when (this value is warp to R most bond length less than 5
Test value, may be configured as other different values) a pixel when, the segmental arc is removed, to avoid the segmental arc being made of collinear points.(4) right
Segmental arc convexity is classified: to segmental arc akConvexity classify.Segmental arc akIts boundary rectangle R will be divided to for upper and lower two parts,
Calculate the region area s of upward partuWith the region area s of downward partdIf su>sd, then segmental arc akIt is downwardly convex, note
For "-"；Otherwise to be upward convex, it is denoted as "+".(5) carry out quadrant classification to segmental arc: foundation is the direction and convexity according to segmental arc.
In first quartile, segmental arc direction is "-", and convexity is "+"；In the second quadrant, segmental arc direction is "+", and convexity is "+"；?
In third quadrant, segmental arc direction is "-", and convexity is "-"；In fourth quadrant, segmental arc direction is "+", and convexity is "-"；(6) it examines
Astronomical observation choosing is oval: it is assumed that the number of candidate elliptic arc is n.From first segmental arc a1It starts the cycle over, meets such as in its neighborhood search
The segmental arc a of lower conditioni(i ≠ 1), segmental arc aiAnd a1Positioned at different quadrants.Neighborhood is defined as follows: with a1Center centered on,
d1For the rectangular area of side length.d1For empirical value.Meet the segmental arc a of conditioni(i ≠ 1) and segmental arc a1Constitute two of candidate circle
Point.(7) it calculates candidate elliptical center: using the coordinate value of candidate oval each pixel as sample parameter, substituting into ellipse side
Journey solves elliptic parameter according to least square.After acquiring elliptic parameter, oval analytical expression is established, then will be on candidate ellipse
The coordinate value of each point substitutes into elliptic equation, calculates residual error, and removal residual error is greater than the pixel of 2 pixels, the number of left point
For n1.If remaining point is re-used as new sample again more than the 80% of n0 (n0 is initial number of samples) by n1 number
Elliptical parameter is calculated, until the calculating residual error of each sample is both less than 2 pixels.
In low resolution, when only detecting a circular target region, circular target region is border circular areas
Extraneous quadrangle.When detecting multiple circular targets, and when two circular targets are closer (such as concentric circles, two centers of circle
Target essentially coincides), it needs two target areas being integrated into a target area.If each circular target region is by rectangle table
Show C (ri00,ri01,ri10,ri11), i=1...m, ri00,ri01,ri10,ri11Indicate four vertex of rectangle, m indicates target
Number.Adjacent target area is just entirely a target area by the purpose of target area integration.Targeted integration process is such as
Shown in Fig. 6.
Using the 1st target as benchmark target, is started the cycle over from i-th (i=2) a target, judge whether i-th of target needs
Together with the 1st targeted integration.If it is required, then by where datum target region and i-th target region carry out
Integration；Conversely, i-th of target area to be stored in new vector.It is to be recycled to terminate and then each target in new vector is pressed into figure
6 process is integrated.
The coordinate that reference area is enabled in the present invention is C0(r000,r001,r010,r011), the coordinate of i-th of target area is
Ci(ri00,ri01,ri10,ri11).Wherein, r0j,kAnd rij,kJ, k=0,1 respectively indicate four vertex of target area, definition
Are as follows: r000=(x000,y000), r001=(x001,y001), r010=(x010,y010), r011=(x011,y011),ri00=
(xi00,yi00), ri01=(xi01,yi01), ri10=(xi10,yi10), ri11=(xi11,yi11)。
Calculate C0And CiWhether region, which needs, is integrated, and following step is specifically included: (1) calculating C0And CiRegion is public
Region area S0i；S0i=0, from CiFirst element in region starts, and from left to right, is recycled from top to bottom；It is assumed that current
The coordinate of point is (xi,yi), if x000≤xi≤x001, and y000≤yi≤y001；Then S0i=S0i+1；(2) it calculates i-th
Whether target area and reference area need to integrate, and calculate C0Area S0, CiArea Si.IfOrThen indicate that i-th of target area needs and reference area is integrated, the region after integration is known as new benchmark
Region.Updated reference area C0′。
Target is located at different regions on low-resolution image and high-definition picture, as shown in Figure 7.To complete target
Positioning on high-definition picture needs target mapping to high-definition picture from low-resolution image.Fig. 7 target is by low
Image in different resolution maps to high-definition picture schematic diagram.
Target is specifically included into following step: (1) mesh in the area maps where low-resolution image to high-resolution
It is marked on low-resolution image IsOn coordinate be (xs,ys), diameter of the target circle on low-resolution image is set as rs, this value
For empirical value, then rectangular area position of the target on low-resolution image are as follows: (r00′,r01′,r10′,r11′)。
r00'=(xs-rs- 10, ys-rs-10)；
r01'=(xs+rs+ 10, ys-rs-10)；
r10'=(xs-rs- 10, ys+rs+10)；
r11'=(xs+rs+ 10, ys+rs+10)。
(2) target is in high-definition picture IoUpper rectangular area position O0(r00,r01,r10,r11) are as follows:
r00=(2 (xs-rs- 10), 2 (ys-rs-10))；
r01=(2 (xs+rs+ 10), 2 (ys-rs-10))；
r10=(2 (xs-rs- 10), 2 (ys+rs+10))；
r11=(2 (xs+rs+ 10), 2 (ys+rs+10))。
(3) in O0On identify and position target.Target's center's coordinate is (xc′,yc′)；
(4) target's center is by O0Subcoordinate system maps to I0Image coordinate system (xc,yc)。
xc=xc′+2(xs-rs-10)
yc=yc′+2(ys-rs-10)
According to binocular stereo vision principle, the inside and outside parameter of two cameras is first demarcated, is closed further according to the geometry of two cameras
System calculates circular target central three-dimensional point coordinate.
The coefficient and parameter gone out given in the above embodiments, is available to those skilled in the art to realize or use
Invention, invention, which does not limit, only takes aforementioned disclosed numerical value, in the case where not departing from the thought of invention, the technology of this field
Personnel can make various modifications or adjustment to above-described embodiment, thus the protection scope invented is not by above-described embodiment institute
Limit, and should be the maximum magnitude for meeting the inventive features that claims are mentioned.
Claims (8)
1. the quick round object localization method of high-precision, specifically includes following step: using high under round-the-clock circumstances not known
Resolution camera acquires image, carries out pyramid decomposition to image after acquiring image, the low-resolution image on pyramid top
On to complete object or pollution target carry out initial location；Then again by the complete object or pollution target on low-resolution image
Position maps to high-definition picture；Complete object or pollution target are calculated in the target mapping area of high-definition picture
Central point three-dimensional coordinate；
In low resolution, when only detecting a circular target region, circular target region is the external world of border circular areas
Quadrangle；When detecting multiple circular targets, and when two circular targets are closer, two target areas are integrated into one
Target area, if each circular target region indicates C (ri by rectangle00,ri01,ri10,ri11), i=1...m, ri00,ri01,
ri10,ri11Indicate four vertex of rectangle, m indicates the number of target, using the 1st target as benchmark target, from i-th of target
It starts the cycle over, judges whether i-th of target needs together with the 1st targeted integration, if it is desired, then will be where datum target
Region and i-th of target region integrated；Conversely, i-th of target area is stored in new vector, end to be recycled
And then each target in new vector is integrated；The coordinate for enabling reference area is C0(r000,r001,r010,r011), i-th
The coordinate of a target area is Ci(ri00,ri01,ri10,ri11), wherein r0j,kAnd rij,kJ, k=0,1 respectively indicate target area
Four vertex in domain, is defined as: r000=(x000,y000), r001=(x001,y001), r010=(x010,y010), r011=
(x011,y011),ri00=(xi00,yi00), ri01=(xi01,yi01), ri10=(xi10,yi10), ri11=(xi11,yi11)；
Calculate C0And CiWhether region, which needs, is integrated, and following step is specifically included: (1) calculating C0And CiRegion public domain face
Product S0i；S0i=0, from CiFirst element in region starts, and from left to right, is recycled from top to bottom；It is assumed that the seat of current point
It is designated as (xi,yi), if x000≤xi≤x001, and y000≤yi≤y001；Then S0i=S0i+1；(2) i-th of target area is calculated
Whether domain and reference area need to integrate, and calculate C0Area S0, CiArea Si；IfOr
Then indicate that i-th of target area needs and reference area is integrated, the region after integration is known as new reference area；After update
Reference area C0′；
Described pair of pollution target carries out the process of initial location specifically: (1) obtains the non-closed curve in picture, be denoted as curve
L；(2) to each point on curve, the difference along the direction x and y is calculated, is denoted as dx, dy respectively；Since first point, by dx,
The identical contiguous pixels of dy symbol are denoted as segmental arc ak, each akWith the symbol of itself, "+" or "-"；(3) to each segmental arc
akConvexity classify: segmental arc akIts boundary rectangle R will be divided to for upper and lower two parts, calculates the region area of upward part
suWith the region area s of downward partdIf su>sd, then segmental arc akBe it is downwardly convex, be denoted as "-"；Otherwise to be upward convex, it is denoted as
"+"；(4) quadrant classification is carried out to segmental arc according to the direction of segmental arc and convexity, in first quartile, segmental arc direction is "-", convexity
For "+"；In the second quadrant, segmental arc direction is "+", and convexity is "+"；In third quadrant, segmental arc direction is "-", and convexity is
"-"；In fourth quadrant, segmental arc direction is "+", and convexity is "-"；(5) from first segmental arc a1It starts the cycle over, is searched in its neighborhood
Rope segmental arc ai(i ≠ 1), segmental arc aiAnd a1Positioned at different quadrants, segmental arc ai(i ≠ 1) and segmental arc a1Constitute candidate elliptical two
Point；(6) using the coordinate value of candidate oval each pixel as sample parameter, elliptic equation is substituted into, is solved according to least square
Elliptic parameter after acquiring elliptic parameter, establishes oval analytical expression, then the coordinate value substitution of candidate oval each point is ellipse
Equation of a circle calculates residual error, and removal residual error is greater than the pixel of threshold value, and the number of left point is n1；If n1 number is more than initial
Remaining point is then re-used as new sample and recalculates elliptical parameter by the setting ratio of number of samples n0, until each sample
This calculating residual error is both less than threshold value.
2. the quick round object localization method of high-precision under round-the-clock circumstances not known as described in claim 1, it is characterised in that
The method also includes verifying to the circular target oriented, pseudo- round coordinate is rejected.
3. the quick round object localization method of high-precision under round-the-clock circumstances not known as claimed in claim 2, it is characterised in that
The verification process includes the following steps: that (1) calculates central point (x under coordinates of original image coordinates systemc,yc) and its neighborhood gray scale
Mean value gn；(2) with point (xc,yc) centered on, along being that set angle direction scans for x-axis or y-axis angle, reach circle
Behind the boundary of shape object, N number of pixel is continued searching further along the direction, centered on n-th pixel, the gray scale for calculating neighborhood is equal
Value, obtains the gray average g in 8 directionsex；The gray average g in (3) 8 directionsexSubtract gray average gnIf 8 gray scale differences
Sign symbol it is identical, then calculated central coordinate of circle is not pseudo- round coordinate, otherwise is pseudo- coordinate.
4. the quick round object localization method of high-precision under round-the-clock circumstances not known as claimed in claim 3, it is characterised in that
The gray average of the neighborhood is the gray average of n*n neighborhood.
5. the quick round object localization method of high-precision under round-the-clock circumstances not known as claimed in claim 4, it is characterised in that
The n is that space circular target maps to camera as the one third of the minor axis length of planar elliptical.
6. the high-precision system that quickly circular target positions under round-the-clock circumstances not known, it is characterised in that specifically include high-resolution
Camera, picture breakdown module, target locating module, mapping block and center point calculation module；The high resolution camera is used for
Acquire image；Described image decomposing module is used to carry out pyramid decomposition to acquired image, obtains low-resolution image；Institute
It is fixed for carrying out outline to complete object or pollution target on the low-resolution image on pyramid top to state target locating module
Position；The mapping block be used for by low-resolution image complete object or pollution target position map to high resolution graphics
Picture；The center point calculation module is used to calculate complete object or pollution mesh in the target mapping area of high-definition picture
The three-dimensional coordinate of target central point；
In low resolution, when only detecting a circular target region, circular target region is the external world of border circular areas
Quadrangle；When detecting multiple circular targets, and when two circular targets are closer, two target areas are integrated into one
Target area, if each circular target region indicates C (ri by rectangle00,ri01,ri10,ri11), i=1...m, ri00,ri01,
ri10,ri11Indicate four vertex of rectangle, m indicates the number of target, using the 1st target as benchmark target, from i-th of target
It starts the cycle over, judges whether i-th of target needs together with the 1st targeted integration, if it is desired, then will be where datum target
Region and i-th of target region integrated；Conversely, i-th of target area is stored in new vector, end to be recycled
And then each target in new vector is integrated；The coordinate for enabling reference area is C0(r000,r001,r010,r011), i-th
The coordinate of a target area is Ci(ri00,ri01,ri10,ri11), wherein r0j,kAnd rij,kJ, k=0,1 respectively indicate target area
Four vertex in domain, is defined as: r000=(x000,y000), r001=(x001,y001), r010=(x010,y010), r011=
(x011,y011),ri00=(xi00,yi00), ri01=(xi01,yi01), ri10=(xi10,yi10), ri11=(xi11,yi11)；
Calculate C0And CiWhether region, which needs, is integrated, and following step is specifically included: (1) calculating C0And CiRegion public domain face
Product S0i；S0i=0, from CiFirst element in region starts, and from left to right, is recycled from top to bottom；It is assumed that the seat of current point
It is designated as (xi,yi), if x000≤xi≤x001, and y000≤yi≤y001；Then S0i=S0i+1；(2) i-th of target area is calculated
Whether domain and reference area need to integrate, and calculate C0Area S0, CiArea Si；IfOr
Then indicate that i-th of target area needs and reference area is integrated, the region after integration is known as new reference area；After update
Reference area C0′；
Described pair of pollution target carries out the process of initial location specifically: (1) obtains the non-closed curve in picture, be denoted as curve
L；(2) to each point on curve, the difference along the direction x and y is calculated, is denoted as dx, dy respectively；Since first point, by dx,
The identical contiguous pixels of dy symbol are denoted as segmental arc ak, each akWith the symbol of itself, "+" or "-"；(3) to each segmental arc
akConvexity classify: segmental arc akIts boundary rectangle R will be divided to for upper and lower two parts, calculates the region area of upward part
suWith the region area s of downward partdIf su>sd, then segmental arc akBe it is downwardly convex, be denoted as "-"；Otherwise to be upward convex, it is denoted as
"+"；(4) quadrant classification is carried out to segmental arc according to the direction of segmental arc and convexity, in first quartile, segmental arc direction is "-", convexity
For "+"；In the second quadrant, segmental arc direction is "+", and convexity is "+"；In third quadrant, segmental arc direction is "-", and convexity is
"-"；In fourth quadrant, segmental arc direction is "+", and convexity is "-"；(5) from first segmental arc a1It starts the cycle over, is searched in its neighborhood
Rope segmental arc ai(i ≠ 1), segmental arc aiAnd a1Positioned at different quadrants, segmental arc ai(i ≠ 1) and segmental arc a1Constitute candidate elliptical two
Point；(6) using the coordinate value of candidate oval each pixel as sample parameter, elliptic equation is substituted into, is solved according to least square
Elliptic parameter after acquiring elliptic parameter, establishes oval analytical expression, then the coordinate value substitution of candidate oval each point is ellipse
Equation of a circle calculates residual error, and removal residual error is greater than the pixel of threshold value, and the number of left point is n1；If n1 number is more than initial
Remaining point is then re-used as new sample and recalculates elliptical parameter by the setting ratio of number of samples n0, until each sample
This calculating residual error is both less than threshold value.
7. the high-precision system that quickly circular target positions, feature exist under round-the-clock circumstances not known as claimed in claim 6
In the system also includes pseudo- round authentication modules, the puppet for will identify that is round to be removed.
8. the high-precision system that quickly circular target positions, feature exist under round-the-clock circumstances not known as claimed in claim 6
In the system also includes pollute circle identification module, the non-closed song for polluting circle identification module and going out in picture for identification
Line carries out quadrant classification to segmental arc according to the direction of segmental arc and convexity, according to the quadrant sorted out judge multiple segmental arcs whether structure
At a contaminated ellipse, to identify contaminated circle.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610039698.7A CN105718929B (en) | 2016-01-21 | 2016-01-21 | The quick round object localization method of high-precision and system under round-the-clock circumstances not known |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610039698.7A CN105718929B (en) | 2016-01-21 | 2016-01-21 | The quick round object localization method of high-precision and system under round-the-clock circumstances not known |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105718929A CN105718929A (en) | 2016-06-29 |
| CN105718929B true CN105718929B (en) | 2019-04-30 |
Family
ID=56153684
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610039698.7A Active CN105718929B (en) | 2016-01-21 | 2016-01-21 | The quick round object localization method of high-precision and system under round-the-clock circumstances not known |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105718929B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107490346B (en) * | 2017-08-17 | 2021-05-28 | 江苏省质量和标准化研究院 | RFID multi-label network three-dimensional measurement modeling method based on vision |
| CN107590829B (en) * | 2017-09-18 | 2020-06-30 | 西安电子科技大学 | Seed point picking method suitable for multi-view dense point cloud data registration |
| CN110502954B (en) * | 2018-05-17 | 2023-06-16 | 杭州海康威视数字技术股份有限公司 | Video analysis method and device |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102129685A (en) * | 2011-03-24 | 2011-07-20 | 杭州电子科技大学 | Method for detecting irregular circle based on Gauss pyramid decomposition |
-
2016
- 2016-01-21 CN CN201610039698.7A patent/CN105718929B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102129685A (en) * | 2011-03-24 | 2011-07-20 | 杭州电子科技大学 | Method for detecting irregular circle based on Gauss pyramid decomposition |
Non-Patent Citations (3)
| Title |
|---|
| 一种新的基于弧段提取的椭圆检测方法;王春芳等;《计算机测量与控制》;20150228;第587页右栏第2段-第589页右栏第1段，图1-5 |
| 一种简单的灰度图像边缘检测算法;孙亮,李敬文;《兰州交通大学学报》;20130228(第32卷第1期);第111页右栏第2段-112页左栏第4段，图1-2 |
| 复杂背景下的圆形识别技术研究;程鹏;《中国优秀硕士学位论文全文数据库信息科技辑》;20131015;第35页第2段，第40页第1段-第41页第1段 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105718929A (en) | 2016-06-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106803244B (en) | Defect identification method and system | |
| CN106340044B (en) | Join automatic calibration method and caliberating device outside video camera | |
| CN107392929B (en) | Intelligent target detection and size measurement method based on human eye vision model | |
| CN110765992B (en) | Seal identification method, medium, equipment and device | |
| CN106251353A (en) | Weak texture workpiece and the recognition detection method and system of three-dimensional pose thereof | |
| CN110070557A (en) | A kind of target identification and localization method based on edge feature detection | |
| CN105865344A (en) | Workpiece dimension measuring method and device based on machine vision | |
| US9639781B2 (en) | Systems and methods for classification and alignment of highly similar or self-similar patterns | |
| CN109064481B (en) | Machine vision positioning method | |
| CN109145756A (en) | Object detection method based on machine vision and deep learning | |
| CN106446894A (en) | Method for recognizing position of spherical object based on contour | |
| KR102073468B1 (en) | System and method for scoring color candidate poses against a color image in a vision system | |
| CN104460505A (en) | Industrial robot relative pose estimation method | |
| CN105718929B (en) | The quick round object localization method of high-precision and system under round-the-clock circumstances not known | |
| CN103632384B (en) | The rapid extracting method of built-up type mark point and mark dot center | |
| KR101461108B1 (en) | Recognition device, vehicle model recognition apparatus and method | |
| CN104915678A (en) | Method and device for detecting target object in power transmission line | |
| Li et al. | Road markings extraction based on threshold segmentation | |
| CN106815830B (en) | Image defect detection method | |
| CN107388991A (en) | A kind of more fillet axial workpiece radius of corner measuring methods in end face | |
| CN109308714A (en) | Camera and laser radar information method for registering based on classification punishment | |
| CN115546170A (en) | Fan blade defect positioning method and system based on laser ranging | |
| CN117496401A (en) | Full-automatic identification and tracking method for oval target points of video measurement image sequences | |
| CN114926635B (en) | Target segmentation method in multi-focus image combined with deep learning method | |
| CN103955929B (en) | Image local edge pattern and non-edge mode judging method and judgment means |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |