CN108257185A - More checkerboard angle point detection process and camera marking method - Google Patents
More checkerboard angle point detection process and camera marking method Download PDFInfo
- Publication number
- CN108257185A CN108257185A CN201810004029.5A CN201810004029A CN108257185A CN 108257185 A CN108257185 A CN 108257185A CN 201810004029 A CN201810004029 A CN 201810004029A CN 108257185 A CN108257185 A CN 108257185A
- Authority
- CN
- China
- Prior art keywords
- angle point
- candidate
- conllinear
- candidate angular
- angular
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/80—Analysis of captured images to determine intrinsic or extrinsic camera parameters, i.e. camera calibration
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of more checkerboard angle point detection process and the scaling methods of video camera, include the following steps：The present invention is by obtaining a tessellated image containing multiple and different angles；Corner Detection is carried out to the image of acquisition and determines candidate angular；The candidate angular is screened further according to angle point growth algorithm, determines X-comers, and identify and be partitioned into each gridiron pattern array, obtain the technical solution of multiple X-comers sequences, so as to enormously simplify calibrating procedure, calibration cost is reduced, has saved the nominal time.
Description
Technical field
The present invention relates to camera calibration technical field more particularly to a kind of checkerboard angle point detection process and video camera
Scaling method.
Background technology
Vision is that the mankind obtain the most important mode of information, and machine vision is as the automation solution party for substituting eyes imaging
Case has played increasing effect, such as robot navigation, vehicle DAS (Driver Assistant System), automatic assembling etc. in multiple neighborhoods,
Video camera is the image device of machine vision, can be quantified using multiple two dimensional images and three-dimensional modeling, but premise are carried out to visual field
It is to need to know the physical characteristic of video camera and geometrical property, camera properties parameter can be obtained by camera calibration.
On the one hand, low cost acquisition reference object image directly determines the cost of production and calibrating camera module, and then
Determine the cost of end product；On the other hand, the characteristic point of high efficiency extraction calibration object can substantially save the nominal time, and then
More possibilities are created for batch production.
In existing patented technology, manpower is substituted to change the pose of scaling board using manipulator, is needed still exist for
From multiple angle shot scaling boards, the time-consuming of entire calibration process not decreased significantly, this is unfavorable for demarcating camera shooting on a large scale
Machine, and design and structure mechanical device need the devices such as motor, electric machine controller, mechanism member and rational control and operation
Method, this adds increased the costs and difficulty for establishing hardware environment, are unfavorable for the application of machine vision.
Invention content
The present invention carrys out the skill of calibrating camera from multiple images of multiple angles aiming at needing to shoot in the prior art
Art problem is, and it is an object of the present invention to provide a kind of new camera marking method, this method only need shooting one to contain multi-angle scaling board
Image.
Present invention firstly provides a kind of more checkerboard angle point detection process, include the following steps：
Step 1: obtain a tessellated image containing multi-angle；
Step 2: carrying out Corner Detection to the image of acquisition determines candidate angular；
Step 3: being screened according to angle point growth algorithm to the candidate angular, X-comers are determined, and identify
Each gridiron pattern array is partitioned into, obtains multiple X-comers sequences.
Corner Detection described in step 2 has SUSAN Corner Detections, Harris Corner Detections, the angle point based on contour code
Detection or the Corner Detection based on wavelet transformation, preferably Harris Corner Detections.
After determining candidate angular to the image progress Corner Detection of acquisition, template matches are further included to detect pair of angle point
Title property, specifically includes：
Candidate angular neighborhood is determined according to default template size；
Calculate the candidate angular neighborhood and the correlation of default template；
When the related sexual satisfaction preset condition, retain the candidate angular, otherwise, abandon the candidate angular.
It after the symmetry of template matches detection angle point, further includes and further detects symmetry using window, with screening
Candidate angular, specially：
The window of one 11 × 11 is established centered on candidate angular, is counted in the window according to the following formula and meets gray scale
The pixel number n of symmetry：
Wherein, (x, y) is the coordinate of candidate angular, and (i, j) is respectively the offset of line direction and column direction, T for this 11
Pixel grey scale average value in × 11 windows as pixel number n ﹥ 50, retains the candidate angular, otherwise, abandons the candidate
Angle point.
Step 3, according to angle point growth algorithm to the candidate angular carry out angle point growth before, first to candidate angular into
Row coordinate setting.
Step 3, the angle point growth algorithm include：
Based on any candidate angular, one is grown in a first direction and arranges the conllinear angle point in the first direction；
Based on the conllinear angle point of the first direction, the conllinear angle of several columns second direction is grown in a second direction
Point.
It is described based on any candidate angular, grow one in a first direction and arrange the conllinear angle point in the first direction and include：
Using any candidate angular as current candidate angle point, a sub-pixel edge of the current candidate angle point is determined
Direction as the first direction, and by the current candidate angle point and with the current candidate angle point in the first direction
Upper adjacent candidate angular is included in as a pair of adjacent known angle point in the conllinear angle point of the first direction；
Respectively along the positive and negative both direction of the first direction, traverse out with adjacent known angle point in the first direction
Upper adjacent next candidate angular detects the prediction error P of angle point to be predicted as angle point to be predicted
Wherein, r, s are adjacent two known angle points respectively, and t is angle point to be predicted, Cr、Cs、CtIt is angle point r, s, t respectively
Image coordinate, | Cr-Cs| the mould for calculating vector st is long；When predicting that error P is less than or equal to threshold value, then by the angle to be predicted
Point is included in the conllinear angle point of first direction and as lower one in adjacent known angle point, continues to traverse in said first direction
Adjacent next candidate angular；Otherwise just (" direction " refers to the positive direction of the first direction herein in this direction for stopping
Or the negative direction of the first direction) traversal；
After all candidate angulars in said first direction have traversed, detect the conllinear angle point quantity of the first direction with
Whether desired value (the angle point quantity of the direction i.e. on gridiron pattern) is consistent, if unanimously, the conllinear angle point of the first direction is chessboard
Lattice angle point；If inconsistent, give up the conllinear angle point of the first direction.
Based on the conllinear angle point of the first direction, the conllinear angle point of several columns second direction is grown in a second direction
Including：
Using any angle point in the conllinear angle point of the first direction as current candidate angle point, the current candidate angle is determined
Point another sub-pixel edge direction as the second direction, and by the current candidate angle point and with the current time
Select the role a little in said first direction adjacent candidate angular as a pair of adjacent known angle point；
Respectively along the positive and negative both direction of the second direction, traverse out with the current candidate angle point described second
Adjacent next candidate angular is as angle point to be predicted on direction, and detects the prediction error P of angle point to be predicted
Wherein, r, s are two adjacent known angle points respectively, and t is angle point to be predicted, Cr、Cs、CtIt is angle point r, s, t respectively
Image coordinate, | Cr-Cs| the mould for calculating vector st is long；When predicting that error P is less than or equal to threshold value, then by the angle to be predicted
Point and the current candidate angle point are included in the conllinear angle point of second direction and as lower to adjacent known angle point, continue traversal described
Adjacent next candidate angular in second direction；Otherwise just stopping the direction, (" direction " refers to the second direction herein
Positive direction or the second direction negative direction) on traversal；
After all candidate angulars in this second direction have traversed, detect the conllinear angle point quantity of the second direction with
Whether desired value (the angle point quantity of the direction i.e. on gridiron pattern) is consistent, if unanimously, the conllinear angle point of the second direction is chessboard
Lattice angle point；If inconsistent, give up the conllinear angle point of the second direction.
Secondly the present invention is to provide a kind of camera marking method, include the following steps：
Step 1: obtain a tessellated image containing multi-angle；
Step 2: carrying out Corner Detection to the image of acquisition determines candidate angular；
Step 3: being screened according to angle point growth algorithm to the candidate angular, X-comers are determined, and identify
Each gridiron pattern array is partitioned into, obtains multiple X-comers sequences；
Step 4: it is demarcated with obtained multiple X-comers sequence pair video cameras.
Corner Detection described in step 2 has SUSAN Corner Detections, Harris Corner Detections, the angle point based on contour code
Detection or the Corner Detection based on wavelet transformation, preferably Harris Corner Detections.
After determining candidate angular to the image progress Corner Detection of acquisition, template matches are further included to detect pair of angle point
Title property and using window symmetry is further detected, to screen candidate angular；
The template matches specifically include to detect the symmetry of angle point：
Candidate angular neighborhood is determined according to default template size；
Calculate the candidate angular neighborhood and the correlation of default template；
When the related sexual satisfaction preset condition, retain the candidate angular, otherwise, abandon the candidate angular.
It is described with window carry out correlation detection screening candidate angular include：
The window of one 11 × 11 is established centered on candidate angular, is counted in the window according to the following formula and meets gray scale
The pixel number n of symmetry：
Wherein, (x, y) is the coordinate of candidate angular, and (i, j) is respectively the offset of line direction and column direction, T for this 11
Pixel grey scale average value in × 11 windows as pixel number n ﹥ 50, retains the candidate angular, otherwise, abandons the candidate
Angle point.
The angle point growth algorithm includes：
Using any candidate angular as current candidate angle point, a sub-pixel edge direction of current candidate angle point is determined
As the first direction, and by the current candidate angle point and with current candidate angle point phase in said first direction
Adjacent candidate angular is included in as a pair of adjacent known angle point in the conllinear angle point of the first direction；
Respectively along the positive and negative both direction of the first direction, traverse out with adjacent known angle point in the first direction
Upper adjacent next candidate angular detects the prediction error P of angle point to be predicted as angle point to be predicted
Wherein, r, s are adjacent two known angle points respectively, and t is angle point to be predicted, Cr、Cs、CtIt is angle point r, s, t respectively
Image coordinate, | Cr- Cs | the mould for calculating vector st is long；When predicting that error P is less than or equal to threshold value, then by the angle to be predicted
Point is included in the conllinear angle point of first direction and as lower one in adjacent known angle point, continues to traverse in said first direction
Adjacent next candidate angular；Otherwise just stop the direction (herein " direction " refer to the first direction positive direction or
The negative direction of first direction described in person) on traversal；
After all candidate angulars in said first direction have traversed, detect the conllinear angle point quantity of the first direction with
Whether desired value (the angle point quantity of the direction i.e. on gridiron pattern) is consistent, if unanimously, the conllinear angle point of the first direction is chessboard
Lattice angle point；If inconsistent, give up the conllinear angle point of the first direction.
Using any angle point in the conllinear angle point of the first direction as current candidate angle point, the current candidate angle is determined
Point another sub-pixel edge direction as the second direction, and by the current candidate angle point and with the current time
Select the role a little in said first direction adjacent candidate angular as a pair of adjacent known angle point；
Respectively along the positive and negative both direction of the second direction, traverse out and exist with the adjacent known current candidate angle point
Next candidate angular adjacent in second direction is as angle point to be predicted described in same straight line, and detects the pre- of angle point to be predicted
Survey error P
Wherein, r, s are two adjacent known angle points respectively, and t is angle point to be predicted, Cr、Cs、CtIt is angle point r, s, t respectively
Image coordinate, | Cr-Cs| the mould for calculating vector st is long；When predicting that error P is less than or equal to threshold value, then by the angle to be predicted
Point and the current candidate angle point are included in the conllinear angle point of second direction and as lower to adjacent known angle point, continue traversal same
Next candidate angular adjacent in second direction described in straight line；Otherwise just (" direction " refers to institute herein in this direction for stopping
State the positive direction of second direction or the negative direction of the second direction) traversal；
After all candidate angulars in this second direction have traversed, detect the conllinear angle point quantity of the second direction with
Whether desired value (the angle point quantity of the direction i.e. on gridiron pattern) is consistent, if unanimously, the conllinear angle point of the second direction is chessboard
Lattice angle point；If inconsistent, give up the conllinear angle point of the second direction.
In the present invention, so-called " known angle point " refers to assume that the candidate angular is the conllinear angle point of first direction or second
One supposed premise of the conllinear angle point in direction, for speculating whether candidate angular adjacent thereto meets prediction error P；It is so-called
" a pair of adjacent known angle point ", refers in a first direction or in second direction, it is assumed that two adjacent candidate angulars be one
To adjacent known angle point as premise, for speculating it is pre- whether candidate angular adjacent on first direction or in second direction meets
Survey error.
In the present invention, during so-called " first direction " " second direction " can be the Corner Detection of preceding step two
The angle point direction calculated, such as can be SUSAN Corner Detections, Harris Corner Detections, the angle point based on contour code
Detection or the edge direction calculated of the Corner Detection based on wavelet transformation, preferably Harris Corner Detections calculate two
A edge direction, when first direction is the one edge direction of angle point, then second direction is exactly the another one edge of angle point
Direction.
The positive effect of the present invention is：The present invention is by obtaining a tessellated figure containing multiple and different angles
Picture；Corner Detection is carried out to the image of acquisition and determines candidate angular；The candidate angular is carried out further according to angle point growth algorithm
Screening determines X-comers, and identifies and be partitioned into each gridiron pattern array, obtains the skill of multiple X-comers sequences
Art scheme so as to enormously simplify calibrating procedure, reduces calibration cost, has saved the nominal time.
Description of the drawings
Fig. 1 is the flow diagram of camera marking method of the present invention；
Fig. 2 is the corners Matching Prototype drawing that one embodiment of the invention provides；
Fig. 3 is the angle point growth schematic diagram that one embodiment of the invention provides；
Fig. 4 is the first situation of use prediction error that one embodiment of the invention provides；
Fig. 5 is the second situation of use prediction error that one embodiment of the invention provides.
Specific embodiment
It is understandable for the above objects, features and advantages of the present invention is enable to become apparent, below in conjunction with the accompanying drawings to the present invention
Specific embodiment be described in detail.
As shown in Fig. 1 the flow diagram of the camera marking method of the embodiment of the present invention.With reference to figure 1, video camera
Scaling method includes the following steps：
Include the following steps：
Step S10 obtains a tessellated image containing multi-angle：
According to demand, tessellated calibration environment more than one is built, by polylith gridiron pattern scaling board according to certain pose
Relationship is placed on the securing means.The angle that every piece of scaling board is placed is different, and does not block between each other, and scaling board is most
Amount is full of entire visual field.
A camera site and angle are selected in the front of visual field environment built, is acquired using camera-control equipment
One image, due to the gridiron pattern scaling board containing multiple and different angle poses in the image, this is equivalent to acquire multiple not
With the scaling board image of pose, and practical shooting time is only the acquisition time of an image, and eliminate frequent movement
The operation of video camera or scaling board.
Step S20 carries out Corner Detection to the image of acquisition and determines candidate angular：
Step S21 extracts the angle point in image
Angle point in extraction image does not limit Corner Detection Algorithm, and all Corner Detection Algorithms are all in the protection of the present invention
In the range of.SUSAN Corner Detections, Harris Corner Detections, the Corner Detection based on contour code or the angle based on wavelet transformation
Point detection etc..Present invention detailed description Harris Corner Detection Algorithms carry out Corner Detection, calculate the second-order partial differential coefficient of angle point,
There are three types of the directions of partial derivative：Horizontal direction, vertical direction, 45 ° of directions, the present invention are used and are obtained often based on one-dimensional convolution algorithm
The partial derivative of a pixel is denoted as horizontal direction second-order partial differential coefficient as dxx, and vertical direction second-order partial differential coefficient is dyy, 45 ° of directions
Second-order partial differential coefficient be dxy.Coefficient matrix corresponding to the offset variation amount of neighborhood is：
It is assumed that two characteristic values of matrix G are e1, e2, corresponding feature vector is respectively v1, v2, when e1 and e2 is ratio
Larger numerical value, the corresponding pixels of matrix G are angle point, and v1 and v2 are the edge directions of angle point.
Harris algorithms can detect all X-comers, but also include the angle point of other strong Gradient Features, such as
When background is more complicated, the crosspoint of lines will be easy to be identified as angle point by Harris algorithms, and these points are not taken the photograph
The point set that camera calibration needs are used.Therefore angle point Harris algorithms calculated is as candidate angular.
Step S22, template matches detect the symmetry of angle point；
X-comers have stronger symmetry, can further screen angle point using symmetry.It is not limited at this pair
Title property detection algorithm, all Symmetry Detection algorithms are all within the scope of the present invention.The present invention is come using template matches
Symmetry is detected, template is as shown in Fig. 2, for the brightness of image difference for coping with illumination and angle difference is brought, using adaptive
Template first determines candidate angular neighborhood, then calculate candidate angular neighborhood and the correlation of template according to template size, correlation tool
Body is the average gray in neighborhood where calculating candidate angular, is made using the difference between the average gray in the neighborhood and 255
For the gray value in template clear zone, the gray value of template dark space is set as 0, during related to the template sexual satisfaction preset condition of candidate angular,
Retain candidate angular, otherwise, abandon candidate angular.
Step S23 further detects angle point symmetry using window, to screen candidate angular：
In previous step detection, in candidate angular is not that tessellated angle point can be still extracted, they are needed
Further it is removed.The window of one 11 × 11 is established centered on candidate angular, counts full in the window according to the following formula
The pixel number n of sufficient gray scale symmetry：
Wherein, (x, y) is the coordinate of candidate angular, and (i, j) is respectively the offset of line direction and column direction, T for this 11
Pixel grey scale average value in × 11 windows as pixel number n ﹥ 50, retains candidate angular, otherwise, abandons candidate angular.
Step S30 screens candidate angular according to angle point growth algorithm, determines X-comers, and identify and divide
Each gridiron pattern array is cut out, obtains multiple X-comers sequences.
Step S31 orients the subpixel coordinates of candidate angular
In order to improve the precision of subsequent algorithm, need that the coordinate of candidate angular is more precisely located out.Ladder is used at this
The degree characteristic vertical with angle point edge direction：The angle point edge direction vector calculated according to harris, can count angle point
The pixel of adjacent edges calculates the gradient information of each pixel using gradient operator, can vertically build equation according to vector, make
The subpixel coordinates of candidate angular can be oriented with least square method.
In true picture, the variation of grey scale pixel value is the process of a gradual change, and angle point possibly is present at two pixels
Between and be not unit pixel point position, can generate error if the angle point is represented with close pixel, reduce video camera mark
Fixed precision.In the present embodiment, by determining the sub-pixel location of X-comers, the precision of Corner Detection is carried by Pixel-level
Up to sub-pixel, can improve the accuracy of detection of angle point, while improve the stated accuracy of video camera.
In the present embodiment, sub-pixel location is determined based on angle point edge and gradient vertical characteristic, illustratively, by following
Formula can determine sub-pixel location coordinate：
Wherein, p is matrix g when the coordinate of the pixel corresponding to forefront, and N is specific where current X-comers p
Neighborhood, g are the gradient matrixs in particular neighborhood N, and c is the sub-pixel location coordinate of current X-comers p.
The technical solution of the present embodiment obtains candidate angular by the gray level image for demarcating scene, is grown and calculated according to angle point
Method grows candidate angular, filters out X-comers, instead of determining chessboard by angle point symmetry in the prior art
The method of lattice angle point solves the problems, such as that the background angle point with symmetry can not reject, and realizes in the case where complexity demarcates background
It carries out accurately X-comers to detect, improves the robustness of X-comers detection, and determine according to X-comers
The precision that X-comers detect is improved by Pixel-level to sub-pixel, improves chess by the sub-pixel location of X-comers
The accuracy of detection of disk lattice angle point improves the stated accuracy of video camera.
Angle point growth algorithm can arrange the coordinate of X-comers in an orderly manner, while can preferable Ground Split gridiron pattern figure
As region.It needs exist for using two picture characteristics：X-comers have preferable array characteristics；Chessboard in certain neighborhood
Lattice angle point is equidistant.Therefore angle point growth algorithm is specific as follows：
Step S32 based on any candidate angular, grows one and arranges the conllinear angle point in the first direction in a first direction：
Using any candidate angular as current candidate angle point, current candidate angle point (the solid black circle in such as Fig. 3 is determined
Point O) a sub-pixel edge direction as first direction, and by current candidate angle point and with current candidate angle point described
Adjacent candidate angular is included in as a pair of adjacent known angle point in the conllinear angle point of the first direction on first direction；
Respectively along the positive and negative both direction (solid arrow in such as Fig. 3) of first direction, traverse out and adjacent known angle
Next candidate angular adjacent in a first direction is put as angle point to be predicted, and detects the prediction error P of angle point to be predicted
(as shown in Figure 4)
Wherein, r, s are adjacent two known angle points respectively, and t is angle point to be predicted, Cr、Cs、CtIt is angle point r, s, t respectively
Image coordinate, | Cr-Cs| the mould for calculating vector st is long；When predicting that error P is less than or equal to threshold value, then by the angle to be predicted
Point is included in the conllinear angle point of first direction and as lower one in adjacent known angle point, continues to traverse adjacent in a first direction
Next candidate angular；Otherwise just stop the direction (direction refer to first direction positive direction or first direction it is anti-
Direction) on traversal；
After all candidate angulars in a first direction have traversed, i.e., the conllinear angle point of first direction, which is all found, finishes, then
It is whether consistent with the angle point quantity of the direction on desired value, that is, gridiron pattern to detect the conllinear angle point quantity of first direction, if unanimously,
The conllinear angle point of the first direction is X-comers；If inconsistent, give up the conllinear angle point of first direction.
Based on the conllinear angle point of first direction, it is conllinear to grow several columns second direction in a second direction by step S33
Angle point：
Using any angle point in the conllinear angle point of first direction as current candidate angle point, the current candidate angle point is determined
Another sub-pixel edge direction as the second direction, and by the current candidate angle point and with the current candidate angle
Candidate angular adjacent in said first direction is put as a pair of adjacent known angle point (the solid black dot in such as Fig. 3)；
Respectively along the positive and negative both direction (dotted arrow in such as Fig. 3) of second direction, traverse out and current candidate angle
Next candidate angular adjacent in a second direction is put as angle point to be predicted (the white black circle in such as Fig. 3), and is examined
Survey the prediction error P (as shown in Figure 5) of angle point to be predicted
Wherein, r, s are two adjacent known angle points respectively, and t is angle point to be predicted, Cr、Cs、CtIt is angle point r, s, t respectively
Image coordinate, | Cr-Cs| the mould for calculating vector st is long；When predicting that error P is less than or equal to threshold value, then by the angle to be predicted
Point and current candidate angle point are included in the conllinear angle point of second direction and as lower to adjacent known angle point, continue traversal in second direction
Upper adjacent next candidate angular；Otherwise just (direction refers to the positive direction or second of second direction in this direction for stopping
The negative direction in direction) traversal；
After all candidate angulars in a second direction have traversed, the conllinear angle point quantity of detection second direction is with desired value
Whether the angle point quantity of the direction is consistent on gridiron pattern, if unanimously, the conllinear angle point of the second direction is X-comers；If no
Unanimously, then give up the conllinear angle point of the second direction.
Prediction error is used for weighing the order of accuarcy with known angle point predicting candidate angle point, when prediction error is more than a threshold
During value, the angle point for illustrating to find not is the angle point in gridiron pattern, is stopped to this direction finding, when all stopping lifes on all directions
After length, entire gridiron pattern is just shaped, and more accurate X-comers can be further filtered out using angle point growing method,
It identifies each gridiron pattern array simultaneously, so as to complete gridiron pattern dividing processing, obtains the X-comers of multiple different angles
Sequence.
Step S40 is demarcated with obtained multiple X-comers sequence pair video cameras：
In camera calibration, what it is as input data is multipair X-comers coordinate set, and each pair of point coordinates includes one
Group object point coordinate and one group of picpointed coordinate.Camera model is established to be fitted these coordinates, can be calibrated using mathematical method
The parameter of video camera.Camera calibration at least needs the image of 2 different angle points, and the present invention provides the gridiron patterns of multiple angles
Angular coordinate data can further improve stated accuracy, and the application for follow-up machine vision provides High Accuracy Parameter.
Although the invention has been described by way of example and in terms of the preferred embodiments, but it is not for limiting the present invention, any this field
Technical staff without departing from the spirit and scope of the present invention, may be by the methods and technical content of the disclosure above to this hair
Bright technical solution makes possible variation and modification, therefore, every content without departing from technical solution of the present invention, and according to the present invention
Any simple modifications, equivalents, and modifications made to above example of technical spirit, belong to technical solution of the present invention
Protection domain.
Claims (13)
1. a kind of more checkerboard angle point detection process, which is characterized in that include the following steps：
Step 1: obtain a tessellated image containing multi-angle；
Step 2: carrying out Corner Detection to the image of acquisition determines candidate angular；
Step 3: being screened according to angle point growth algorithm to the candidate angular, X-comers are determined, and identify segmentation
Go out each gridiron pattern array, obtain multiple X-comers sequences.
2. more checkerboard angle point detection process as described in claim 1, which is characterized in that the Corner Detection described in step 2 has
SUSAN Corner Detections, Harris Corner Detections, the Corner Detection based on contour code or the Corner Detection based on wavelet transformation,
It is preferred that Harris Corner Detections.
3. more checkerboard angle point detection process as claimed in claim 2, which is characterized in that angle point inspection is carried out to the image of acquisition
It surveys after determining candidate angular, further includes template matches to detect the symmetry of angle point, specifically include：
Candidate angular neighborhood is determined according to default template size；
Calculate the candidate angular neighborhood and the correlation of default template；
When the related sexual satisfaction preset condition, retain the candidate angular, otherwise, abandon the candidate angular.
4. more checkerboard angle point detection process as claimed in claim 3, which is characterized in that in pair of template matches detection angle point
After title property, further include and symmetry is further detected using window, to screen candidate angular, specially：
The window of one 11 × 11 is established centered on candidate angular, counts that meet gray scale in the window symmetrical according to the following formula
The pixel number n of property：
Wherein, (x, y) is the coordinate of candidate angular, and (i, j) is respectively the offset of line direction and column direction, T for this 11 × 11
Pixel grey scale average value in window as pixel number n ﹥ 50, retains the candidate angular, otherwise, abandons the candidate angular.
5. more checkerboard angle point detection process as described in claim 1, which is characterized in that step 3 grows according to angle point and calculates
Before method carries out angle point growth to the candidate angular, coordinate setting first is carried out to candidate angular.
6. more checkerboard angle point detection process as described in claim 1 or 5, which is characterized in that the angle point growth algorithm packet
It includes：
Based on any candidate angular, one is grown in a first direction and arranges the conllinear angle point in the first direction；
Based on the conllinear angle point of the first direction, the conllinear angle point of several columns second direction is grown in a second direction.
7. more checkerboard angle point detection process as claimed in claim 6, which is characterized in that described using any candidate angular as base
Plinth, grows one and arranges the conllinear angle point in the first direction and include in a first direction：
Using any candidate angular as current candidate angle point, a sub-pixel edge direction of the current candidate angle point is determined
As the first direction, and by the current candidate angle point and with current candidate angle point phase in said first direction
Adjacent candidate angular is included in as a pair of adjacent known angle point in the conllinear angle point of the first direction；
Respectively along the positive and negative both direction of the first direction, traverse out and adjacent known angle point phase in said first direction
Adjacent next candidate angular detects the prediction error P of angle point to be predicted as angle point to be predicted
Wherein, r, s are adjacent two known angle points respectively, and t is angle point to be predicted, Cr、Cs、CtIt is the figure of angle point r, s, t respectively
Picture coordinate, | Cr-Cs| the mould for calculating vector st is long；When predicting that error P is less than or equal to threshold value, then the angle point to be predicted is received
Enter in the conllinear angle point of the first direction and as one in the adjacent known angle point of lower a pair, continue traversal in the first party
Adjacent next candidate angular upwards；Otherwise just stop traversal in this direction；
After all candidate angulars in said first direction have traversed, the conllinear angle point quantity of the first direction and expection are detected
Whether value is consistent, if unanimously, the conllinear angle point of the first direction is X-comers；If inconsistent, give up the first party
To conllinear angle point.
8. more checkerboard angle point detection process as claimed in claim 6, which is characterized in that with the conllinear angle point of the first direction
Based on, the conllinear angle point of several columns second direction is grown in a second direction to be included：
Using any angle point in the conllinear angle point of the first direction as current candidate angle point, the current candidate angle point is determined
Another sub-pixel edge direction as the second direction, and by the current candidate angle point and with the current candidate angle
Candidate angular adjacent in said first direction is put as a pair of adjacent known angle point；
Respectively along the positive and negative both direction of the second direction, traverse out with the current candidate angle point in the second direction
Upper adjacent next candidate angular detects the prediction error P of angle point to be predicted as angle point to be predicted
Wherein, r, s are two adjacent known angle points respectively, and t is angle point to be predicted, Cr、Cs、CtIt is the figure of angle point r, s, t respectively
Picture coordinate, | Cr-Cs| the mould for calculating vector st is long；When predict error P be less than or equal to threshold value when, then by the angle point to be predicted with
The current candidate angular is included in the conllinear angle point of the second direction and as the adjacent known angle point of lower a pair, continues traversal in institute
State next candidate angular adjacent in second direction；Otherwise just stop traversal in this direction；
After all candidate angulars in this second direction have traversed, the conllinear angle point quantity of the second direction and expection are detected
Whether value is consistent, if unanimously, the conllinear angle point of the second direction is X-comers；If inconsistent, give up the second party
To conllinear angle point.
9. a kind of camera marking method, which is characterized in that include the following steps：
Step 1: obtain a tessellated image containing multi-angle；
Step 2: carrying out Corner Detection to the image of acquisition determines candidate angular；
Step 3: being screened according to angle point growth algorithm to the candidate angular, X-comers are determined, and identify segmentation
Go out each gridiron pattern array, obtain multiple X-comers sequences；
Step 4: it is demarcated with obtained multiple X-comers sequence pair video cameras.
10. camera marking method as claimed in claim 9, which is characterized in that the Corner Detection described in step 2 has SUSAN
Corner Detection, Harris Corner Detections, the Corner Detection based on contour code or the Corner Detection based on wavelet transformation, preferably
Harris Corner Detections.
11. camera marking method as claimed in claim 10, which is characterized in that it is true that Corner Detection is carried out to the image of acquisition
After determining candidate angular, the symmetry that further includes template matches to detect angle point and symmetry is further detected using window, with
Screen candidate angular；
The template matches specifically include to detect the symmetry of angle point：
Candidate angular neighborhood is determined according to default template size；
Calculate the candidate angular neighborhood and the correlation of default template；
When the related sexual satisfaction preset condition, retain the candidate angular, otherwise, abandon the candidate angular.
It is described with window carry out correlation detection screening candidate angular include：
The window of one 11 × 11 is established centered on candidate angular, counts that meet gray scale in the window symmetrical according to the following formula
The pixel number n of property：
Wherein, (x, y) is the coordinate of candidate angular, and (i, j) is respectively the offset of line direction and column direction, T for this 11 × 11
Pixel grey scale average value in window as pixel number n ﹥ 50, retains the candidate angular, otherwise, abandons the candidate angular.
12. camera marking method as claimed in claim 9, which is characterized in that the angle point growth algorithm includes：
Using any candidate angular as current candidate angle point, a sub-pixel edge direction conduct of current candidate angle point is determined
The first direction, and by the current candidate angle point and with the current candidate angle point it is adjacent in said first direction
Candidate angular is included in as a pair of adjacent known angle point in the conllinear angle point of the first direction；
Respectively along the positive and negative both direction of the first direction, traverse out and adjacent known angle point phase in said first direction
Adjacent next candidate angular detects the prediction error P of angle point to be predicted as angle point to be predicted
Wherein, r, s are adjacent two known angle points respectively, and t is angle point to be predicted, Cr、Cs、CtIt is the figure of angle point r, s, t respectively
Picture coordinate, | Cr-Cs| the mould for calculating vector st is long；When predicting that error P is less than or equal to threshold value, then the angle point to be predicted is received
Enter the conllinear angle point of first direction and as lower one in adjacent known angle point, it is adjacent in said first direction to continue traversal
Next candidate angular；Otherwise just stop traversal in this direction；
After all candidate angulars in said first direction have traversed, the conllinear angle point quantity of the first direction and expection are detected
Whether value is consistent, if unanimously, the conllinear angle point of the first direction is X-comers；If inconsistent, give up the first party
To conllinear angle point.
13. more checkerboard angle point detection process as claimed in claim 12, which is characterized in that also continue to include：
Using any angle point in the conllinear angle point of the first direction as current candidate angle point, the current candidate angle point is determined
Another sub-pixel edge direction as the second direction, and by the current candidate angle point and with the current candidate angle
Candidate angular adjacent in said first direction is put as a pair of adjacent known angle point；
Respectively along the positive and negative both direction of the second direction, traverse out with the current candidate angle point in the second direction
Upper adjacent next candidate angular detects the prediction error P of angle point to be predicted as angle point to be predicted
Wherein, r, s are two adjacent known angle points respectively, and t is angle point to be predicted, Cr、Cs、CtIt is the figure of angle point r, s, t respectively
Picture coordinate, | Cr-Cs| the mould for calculating vector st is long；When predict error P be less than or equal to threshold value when, then by the angle point to be predicted with
The current candidate angle point is included in the conllinear angle point of second direction and as lower to adjacent known angle point, continues traversal described second
Adjacent next candidate angular on direction；Otherwise just stop traversal in this direction；
After all candidate angulars in this second direction have traversed, the conllinear angle point quantity of the second direction and expection are detected
Whether value is consistent, if unanimously, the conllinear angle point of the second direction is X-comers；If inconsistent, give up the second party
To conllinear angle point.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810004029.5A CN108257185A (en) | 2018-01-03 | 2018-01-03 | More checkerboard angle point detection process and camera marking method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810004029.5A CN108257185A (en) | 2018-01-03 | 2018-01-03 | More checkerboard angle point detection process and camera marking method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108257185A true CN108257185A (en) | 2018-07-06 |
Family
ID=62725716
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810004029.5A Pending CN108257185A (en) | 2018-01-03 | 2018-01-03 | More checkerboard angle point detection process and camera marking method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108257185A (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109872372A (en) * | 2019-03-07 | 2019-06-11 | 山东大学 | A kind of small-sized quadruped robot overall Vision localization method and system |
| CN111462250A (en) * | 2020-04-03 | 2020-07-28 | 上海兆芯集成电路有限公司 | Correction system and correction method |
| CN111539970A (en) * | 2020-04-29 | 2020-08-14 | 华中科技大学 | Checkerboard angular point detection method suitable for structured light three-dimensional reconstruction |
| CN111640157A (en) * | 2020-05-28 | 2020-09-08 | 华中科技大学 | Checkerboard corner detection method based on neural network and application thereof |
| CN111815685A (en) * | 2020-09-09 | 2020-10-23 | 季华实验室 | Checkerboard angular point positioning method and device and electronic equipment |
| CN111833405A (en) * | 2020-07-27 | 2020-10-27 | 北京大华旺达科技有限公司 | Calibration identification method and device based on machine vision |
| CN112614146A (en) * | 2020-12-21 | 2021-04-06 | 广东奥普特科技股份有限公司 | Method and device for judging chessboard calibration corner points and computer readable storage medium |
| CN112907676A (en) * | 2019-11-19 | 2021-06-04 | 浙江商汤科技开发有限公司 | Calibration method, device and system of sensor, vehicle, equipment and storage medium |
| CN113487594A (en) * | 2021-07-22 | 2021-10-08 | 上海嘉奥信息科技发展有限公司 | Sub-pixel angular point detection method, system and medium based on deep learning |
| CN113496145A (en) * | 2020-03-19 | 2021-10-12 | 武汉四维图新科技有限公司 | Label corner point acquisition method, device and equipment |
| CN113658261A (en) * | 2020-05-12 | 2021-11-16 | 武汉Tcl集团工业研究院有限公司 | Checkerboard angular point position detection method and device, intelligent terminal and storage medium |
| WO2022143283A1 (en) * | 2020-12-30 | 2022-07-07 | 深圳市杉川机器人有限公司 | Camera calibration method and apparatus, and computer device and storage medium |
| CN115294078A (en) * | 2022-08-11 | 2022-11-04 | 创新奇智(合肥)科技有限公司 | Method, device and equipment for identifying asymmetric corner cut of glass and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104331900A (en) * | 2014-11-25 | 2015-02-04 | 湖南科技大学 | Corner sub-pixel positioning method in CCD (charge coupled device) camera calibration |
| CN106846412A (en) * | 2017-01-23 | 2017-06-13 | 上海兴芯微电子科技有限公司 | A kind of checkerboard angle point detection process and device |
| CN106887023A (en) * | 2017-02-21 | 2017-06-23 | 成都通甲优博科技有限责任公司 | For scaling board and its scaling method and calibration system that binocular camera is demarcated |
| US20170374345A1 (en) * | 2016-06-23 | 2017-12-28 | Thomson Licensing | Method and apparatus for creating a pair of stereoscopic images using least one lightfield camera |
-
2018
- 2018-01-03 CN CN201810004029.5A patent/CN108257185A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104331900A (en) * | 2014-11-25 | 2015-02-04 | 湖南科技大学 | Corner sub-pixel positioning method in CCD (charge coupled device) camera calibration |
| US20170374345A1 (en) * | 2016-06-23 | 2017-12-28 | Thomson Licensing | Method and apparatus for creating a pair of stereoscopic images using least one lightfield camera |
| CN106846412A (en) * | 2017-01-23 | 2017-06-13 | 上海兴芯微电子科技有限公司 | A kind of checkerboard angle point detection process and device |
| CN106887023A (en) * | 2017-02-21 | 2017-06-23 | 成都通甲优博科技有限责任公司 | For scaling board and its scaling method and calibration system that binocular camera is demarcated |
Non-Patent Citations (2)
| Title |
|---|
| 章为川等: ""图像的角点检测研究综述"", 《电子学报》 * |
| 计算机视觉LIFE: ""基于生长的棋盘格角点检测方法"", 《HTTP://BLOG.CSDN.NET/ELECTECH6/ARTICLE/DETAILS/52770010》 * |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109872372A (en) * | 2019-03-07 | 2019-06-11 | 山东大学 | A kind of small-sized quadruped robot overall Vision localization method and system |
| CN109872372B (en) * | 2019-03-07 | 2021-04-09 | 山东大学 | Global visual positioning method and system for small quadruped robot |
| CN112907676A (en) * | 2019-11-19 | 2021-06-04 | 浙江商汤科技开发有限公司 | Calibration method, device and system of sensor, vehicle, equipment and storage medium |
| CN113496145A (en) * | 2020-03-19 | 2021-10-12 | 武汉四维图新科技有限公司 | Label corner point acquisition method, device and equipment |
| CN111462250A (en) * | 2020-04-03 | 2020-07-28 | 上海兆芯集成电路有限公司 | Correction system and correction method |
| CN111539970A (en) * | 2020-04-29 | 2020-08-14 | 华中科技大学 | Checkerboard angular point detection method suitable for structured light three-dimensional reconstruction |
| CN111539970B (en) * | 2020-04-29 | 2022-09-16 | 华中科技大学 | Checkerboard angular point detection method suitable for structured light three-dimensional reconstruction |
| CN113658261A (en) * | 2020-05-12 | 2021-11-16 | 武汉Tcl集团工业研究院有限公司 | Checkerboard angular point position detection method and device, intelligent terminal and storage medium |
| CN111640157B (en) * | 2020-05-28 | 2022-09-20 | 华中科技大学 | Checkerboard corner detection method based on neural network and application thereof |
| CN111640157A (en) * | 2020-05-28 | 2020-09-08 | 华中科技大学 | Checkerboard corner detection method based on neural network and application thereof |
| CN111833405A (en) * | 2020-07-27 | 2020-10-27 | 北京大华旺达科技有限公司 | Calibration identification method and device based on machine vision |
| CN111833405B (en) * | 2020-07-27 | 2023-12-08 | 北京大华旺达科技有限公司 | Calibration and identification method and device based on machine vision |
| CN111815685A (en) * | 2020-09-09 | 2020-10-23 | 季华实验室 | Checkerboard angular point positioning method and device and electronic equipment |
| CN112614146A (en) * | 2020-12-21 | 2021-04-06 | 广东奥普特科技股份有限公司 | Method and device for judging chessboard calibration corner points and computer readable storage medium |
| CN112614146B (en) * | 2020-12-21 | 2022-05-13 | 广东奥普特科技股份有限公司 | Method and device for judging chessboard calibration corner points and computer readable storage medium |
| WO2022143283A1 (en) * | 2020-12-30 | 2022-07-07 | 深圳市杉川机器人有限公司 | Camera calibration method and apparatus, and computer device and storage medium |
| CN113487594A (en) * | 2021-07-22 | 2021-10-08 | 上海嘉奥信息科技发展有限公司 | Sub-pixel angular point detection method, system and medium based on deep learning |
| CN113487594B (en) * | 2021-07-22 | 2023-12-01 | 上海嘉奥信息科技发展有限公司 | Sub-pixel corner detection method, system and medium based on deep learning |
| CN115294078A (en) * | 2022-08-11 | 2022-11-04 | 创新奇智(合肥)科技有限公司 | Method, device and equipment for identifying asymmetric corner cut of glass and storage medium |
| CN115294078B (en) * | 2022-08-11 | 2024-01-30 | 创新奇智(合肥)科技有限公司 | Glass asymmetric chamfer identification method, device, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108257185A (en) | More checkerboard angle point detection process and camera marking method | |
| CN109146930B (en) | Infrared and visible light image registration method for electric power machine room equipment | |
| CN108898047B (en) | Pedestrian detection method and system based on blocking and shielding perception | |
| CN109211198B (en) | Intelligent target detection and measurement system and method based on trinocular vision | |
| CN108470356B (en) | Target object rapid ranging method based on binocular vision | |
| CN109344882A (en) | Robot based on convolutional neural networks controls object pose recognition methods | |
| CN112200771B (en) | Height measurement method, device, equipment and medium | |
| CN108898634A (en) | Pinpoint method is carried out to embroidery machine target pinprick based on binocular camera parallax | |
| CN108895959B (en) | Camera calibration plate corner point calculation method based on sub-pixels | |
| CN111932504B (en) | Edge contour information-based sub-pixel positioning method and device | |
| CN108470361A (en) | A kind of angle point automatic identification camera calibration method | |
| CN105096348B (en) | The device and method of tint plate in detection image | |
| CN114494045A (en) | Large-scale straight gear geometric parameter measuring system and method based on machine vision | |
| CN105835507A (en) | Method for attaching mobile phone cover lens to liquid crystal display | |
| CN106846412A (en) | A kind of checkerboard angle point detection process and device | |
| CN115014248B (en) | Laser projection line identification and flatness judgment method | |
| CN110807763A (en) | Method and system for detecting ceramic tile surface bulge | |
| CN107452028A (en) | A kind of method and device for determining target image positional information | |
| CN111242026A (en) | Remote sensing image target detection method based on spatial hierarchy perception module and metric learning | |
| CN114612410A (en) | Novel clothing detects device | |
| CN113012096B (en) | Display screen sub-pixel positioning and brightness extraction method, device and storage medium | |
| CN113313116A (en) | Vision-based accurate detection and positioning method for underwater artificial target | |
| CN115277984B (en) | Method for adjusting exposure time of adaptive camera in glass size detection | |
| CN115830018A (en) | Carbon block detection method and system based on deep learning and binocular vision | |
| CN115108466A (en) | Intelligent positioning method for container spreader |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180706 |