CN108550168A - The multiple target quick high accuracy of Intelligentized mechanical arm view-based access control model identifies and localization method - Google Patents
The multiple target quick high accuracy of Intelligentized mechanical arm view-based access control model identifies and localization method Download PDFInfo
- Publication number
- CN108550168A CN108550168A CN201810348592.4A CN201810348592A CN108550168A CN 108550168 A CN108550168 A CN 108550168A CN 201810348592 A CN201810348592 A CN 201810348592A CN 108550168 A CN108550168 A CN 108550168A
- Authority
- CN
- China
- Prior art keywords
- source images
- image
- identification
- template image
- images
- 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
- 238000000034 method Methods 0.000 title claims abstract description 25
- 230000004807 localization Effects 0.000 title claims abstract description 13
- 230000000007 visual effect Effects 0.000 claims abstract description 8
- 230000001629 suppression Effects 0.000 claims abstract description 6
- 230000004044 response Effects 0.000 claims description 12
- 238000005070 sampling Methods 0.000 claims description 9
- 230000008569 process Effects 0.000 claims description 3
- 230000010339 dilation Effects 0.000 claims description 2
- 238000009499 grossing Methods 0.000 claims description 2
- 230000006870 function Effects 0.000 description 5
- 230000009471 action Effects 0.000 description 3
- 230000007613 environmental effect Effects 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000009434 installation Methods 0.000 description 2
- 238000000513 principal component analysis Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 230000003930 cognitive ability Effects 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012827 research and development Methods 0.000 description 1
- 238000005316 response function Methods 0.000 description 1
- 230000008685 targeting Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
- G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T1/00—General purpose image data processing
- G06T1/0014—Image feed-back for automatic industrial control, e.g. robot with camera
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/20—Image enhancement or restoration using local operators
- G06T5/30—Erosion or dilatation, e.g. thinning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
- G06V10/751—Comparing pixel values or logical combinations thereof, or feature values having positional relevance, e.g. template matching
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20016—Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Databases & Information Systems (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Robotics (AREA)
- Image Analysis (AREA)
Abstract
Identification and localization method of the Intelligentized mechanical arm to multiple target quick high accuracy that the present invention relates to a kind of for connecting firmly visual unit.It is realized in source images to the identification and positioning of the quick high accuracy of template image by template image and source images in the present invention.With the down-sampled method restored with laplacian pyramid of gaussian pyramid in the present invention, in the high-precision for improving identification with ensure that identification and positioning while locating speed.In addition the present invention proposes a kind of plane multi-targets recognition and localization method being combined with non-maxima suppression based on normalized crosscorrelation matching algorithm, this method may be implemented to identification and positioning while multiple targets in source images, and substantially increase robustness compared to tional identification and location algorithm.
Description
Technical field
The invention belongs to computer vision fields, and in particular to for connecting firmly the Intelligentized mechanical arm of visual unit to more mesh
Mark identification and the localization method of quick high accuracy.
Background technology
With the research and development of field of machine vision, more and more industrial robots are equipped with vision system, make machine
Device people is accurately positioned technical aspect in autonomous and mechanical arm and is obviously improved.At the same time, panoramic vision and
The successful application of image algorithm also considerably enhances the level of intelligence of mechanical arm, and the reality that mechanical arm captures object
The standard that Shi Xing, Stability and veracity and mechanical arm carry out the level of intelligence of intelligent grabbing using vision guide embodies.It wants
The level of intelligence for improving a series of actions made by mechanical arm, makes it have the function of Intelligent mobile and crawl, it is important to make
Robot has environmental goals certain cognitive ability, determines the action of itself by perceiving environmental goals, it is made to be not required to transport
Dynamic contact can realize the actions such as crawl to environmental goals, and therefore, the research of vision system has become raising robot automtion
A more popular direction, even more emphasis studies a question among these for the identification of target and localization method.
Currently, in the research field, have some comparative maturities and the identification for putting into application and location algorithm, but work as
Lower existing method is more for a certain concrete application, and there are larger limitations.Such as " the plane based on region division
Workpiece identification and localization method " is only applicable in the workpiece with outline of straight line, is unable to get just when workpiece is irregular shape
True recognition result;" statistical matrix marked region algorithm " reaches identifying purpose using the geometric properties of target itself, chooses matter
Heart labelling method realizes the positioning of target object, this method some shaped objects in mixed and disorderly object for identification, and is not suitable for
In identical multi-targets recognition and positioning;" feature based template simultaneously carries out template matches using principal component analysis (PCA) method
Algorithm " realizes the identification and positioning of plurality of classes target, has certain robustness and improves efficiency, but is run in algorithm
It is required for longer Offtime to learn template or feature before, cannot be satisfied the variability of engineering duty demand.Always
For knot, existing target identification is primarily present problems with location algorithm:(1) algorithm based on gray scale or characteristic matching is only
It can identify single goal;(2) recognition methods of multiple target needs longer Offtime to learn template or feature, is not suitable with
The changeable actual conditions of engineering duty;(3) most of algorithms can ensure the accuracy of identification positioning, but speed is slower, nothing
Method meets the requirement of real-time.
Invention content
It is an object of the invention to provide a kind of target identification of versatility to the Intelligentized mechanical arm comprising visual unit
With localization method, quick to multiple target, high-accuracy, high robust identification and positioning are realized, be that Intelligentized mechanical arm is related
The crawl of function, including but not limited to target, target movement, target are picked, and information support is provided.
Need to use the Intelligentized mechanical arm comprising visual unit, wherein visual unit (one camera) and machinery in the present invention
Arm connects firmly installation, and mechanical arm can obtain the relative position of mechanical arm and target according to camera to the location information of target, to
Target is implemented the functions such as to capture, move, pick.The relative position that installation refers to visual unit and mechanical arm is connected firmly to fix.It is real
The multiple target quick high accuracy identification for applying view-based access control model needs to use template image and source images, wherein Prototype drawing with localization method
Seem the image for the target that previously known needs identify, source images are to connect firmly visual unit by Intelligentized mechanical arm to shoot
The image arrived.
High-precision quickly identification and positioning are carried out to multiple target in image with the method in the present invention, realized in source images
Quick high accuracy identification and positioning to template image, include the following steps:
The first step, gaussian pyramid are down-sampled
Multiple target quick high accuracy identification in the present invention and location algorithm need template image and the camera unit shooting of target
The source images arrived.Template image and source images are reduced into image size to improve image by the way that gaussian pyramid is down-sampled first
Matching speed, gaussian pyramid obtain a series of down-sampled images by Gaussian smoothing and sub-sampling, and each layer is all in accordance under
To upper serial number, layer Gi+1Indicate that size is less than layer GiAdjacent layer.A given integer variable n and one-dimensional sampling sequence f
(n), then the up-sampling sequence definition that base is 2 is as follows:
The definition of down-sampling that base is 2 operation is:
f2↓(n)=f (2n) (2)
In general, the whole process time can be shortened to one third by carrying out a down-sampling, but it is down-sampled after
The detailed information of some images can be lost.
Second step, normalized crosscorrelation matching
Obtain template image and source images it is down-sampled after, cross-correlation operation is normalized in the two.Normalized crosscorrelation
Matching algorithm is a kind of template matching algorithm of classics, can be true by the cross correlation value of calculation template image and matching image
Fixed matched degree, search window position when cross correlation value maximum determine position of the template image in image to be matched.
NCC algorithms have very high accuracy and adaptability, and have robustness to the linear transformation of gray value of image, not by gray scale
The influence of the linear transformation of value.Mathematical expression forms of the NCC algorithm response function R in the position coordinate (i, j) is shown below:
I (i+m-1, j+n-1) indicates source images in the pixel value of the position coordinate (i+m-1, j+n-1), T (m, n) expression Prototype drawings
As the position coordinate (m, n) pixel value,Indicate template image and source images in corresponding template size region respectively
The average value of interior pixel value, (M, N) are the size of template image.Source images and template image can be calculated by above formula
A cross correlation value, it may be determined that matched degree, search window position when cross correlation value maximum determine that template image is waiting for
With the position in image, but that be only limitted to identification single goal the case where.For the multi-targets recognition problem based on single mode plate, by
In a series of influence of interference such as illumination, noise, need to find multiple targets just need to find it is all more than certain response
Position, and remove and be presented as the pixel being sticked together in response image.
Third walks, non-maxima suppression
In a window, if there is the response of multiple pixels meets threshold condition, then retain response it is maximum that
Pixel excludes other pixels.The pixel for meeting response lag condition is arranged according to response descending first in the present invention
Row, then achieve the purpose that detect maximum using the dilation operation of image, are in addition provided with distance threshold in the process to full
The pixel of sufficient condition is constrained, and to have the function that remove error hiding, and then avoids the uncertainty of single constraint.Hair
The bright middle two pixel (x used1,y1) and (x2,y2) between the expression formula of distance be:
Rule of thumb it is found that when NCC receptance function values are more than 0.9, match condition is good.Expansive working window 18*18 is set,
Distance threshold is chosen for that 0.9* template short sides are long, obtains response position of the template in source images, but the position is only one
Point target will obtain complete target and need to restore it.
4th step, laplacian pyramid restore
Laplacian pyramid is to subtract first to reduce a series of images of the image amplified afterwards and constitute by source images, Ke Yili
Solution is the inverse form of gaussian pyramid, the mathematical definition that i-th layer of laplacian pyramid:
G in formulaiIndicate the i-th tomographic image, UP operations are by pixel-map that position in source images is (x, y) to target image
The position (2x+1,2y+1), symbolIndicate convolution algorithm, g5×5Indicate the Gaussian kernel of 5*5.
Advantages of the present invention:
1. by the down-sampled size for reducing source images and template image of gaussian pyramid in the present invention, to substantially reduce
The matched calculation amount of normalized crosscorrelation, again by laplacian pyramid pair after finally obtaining object recognition and detection result
It detects obtained result to be restored, to improve identification and ensure that the high-precision of identification and positioning while locating speed
Degree.
2. the present invention proposes a kind of more mesh of plane being combined with non-maxima suppression based on normalized crosscorrelation matching algorithm
Mark not may be implemented with localization method, this method to identification and positioning while multiple targets in source images, and compared to
Tional identification and location algorithm substantially increase robustness so that false-alarm, false retrieval, missing inspection probability all substantially reduce.
Specific implementation mode
It needs to obtain template image first in the present invention, and shoots and obtain source images, just by template image and source images
Can through the invention in the identification of multiple target quick high accuracy identified in source images with localization method and orient all moulds
Plate image object.It obtains providing Information base for Intelligentized mechanical arm subsequent operation after targeting information.More mesh in the present invention
Mark quick high accuracy identification is as follows with localization method:
The first step, gaussian pyramid are down-sampled
Obtained template image and source images progress gaussian pyramid is down-sampled, and the specific down-sampled parameter of gaussian pyramid needs
It is selected according to the demand in Practical Project situation;
Second step, normalized crosscorrelation matching
Cross Correlation Matching is normalized in the down-sampled result of template image and source images that the first step obtains, is obtained corresponding
Cross correlation value;
Third walks, non-maxima suppression
For carrying out non-maxima suppression operation by the cross correlation value that normalized crosscorrelation matches in second step, weight is rejected
Reinspection, which is surveyed, improves efficiency of algorithm, while constraints removal error hiding is arranged;
4th step, laplacian pyramid restore
It walks to have obtained the response position of the template image in source images by third, then is answered by laplacian pyramid
Original finally obtains template image identification and positioning result in source images.
Claims (1)
1. the multiple target quick high accuracy of Intelligentized mechanical arm view-based access control model identifies and localization method, realize in source images to mould
The identification and positioning of the quick high accuracy of plate image, wherein template image are the images for the target that previously known needs identify,
Source images are the images that the visual unit that Intelligentized mechanical arm connects firmly is shot, which is characterized in that are included the following steps:
The first step, gaussian pyramid are down-sampled
It is first that template image and source images is down-sampled fast to improve images match to reduce image size by gaussian pyramid
Degree, gaussian pyramid obtain a series of down-sampled images by Gaussian smoothing and sub-sampling, and each layer is all in accordance with from top to bottom
Serial number, layer Gi+1Indicate that size is less than layer GiAdjacent layer.An integer variable n and one-dimensional sampling sequence f (n) are given, then
The up-sampling sequence definition that base is 2 is as follows:
The definition of down-sampling that base is 2 operation is:
f2↓(n)=f (2n) (2)
Second step, normalized crosscorrelation matching
To obtain it is down-sampled after template image and source images be normalized, the template image T (m, n) after being normalized
With source images I (m, n), calculate the cross correlation value of T (m, n) and I (m, n) by cross correlation algorithm, receptance function R coordinate (i,
J) the mathematical expression form of position is shown below:
I (i+m-1, j+n-1) indicates source images in the pixel value of the position coordinate (i+m-1, j+n-1), T (m, n) expression Prototype drawings
As the position coordinate (m, n) pixel value,Indicate template image and source images in corresponding template size region respectively
The average value of interior pixel value, (M, N) are the size of template image;
Third walks, non-maxima suppression
Pixel for meeting response lag condition in the response image that is obtained by cross correlation algorithm is dropped according to response
Sequence arranges, and then achievees the purpose that detect maximum using the dilation operation of image, is in addition provided with distance threshold in the process
Pixel to meeting condition constrains, and to have the function that remove error hiding, and then avoids the uncertain of single constraint
Property, two pixel (x1,y1) and (x2,y2) between the expression formula of distance be:
4th step, laplacian pyramid restore
The response image obtained in being walked to third carries out laplacian pyramid and restores to obtain Prototype drawing in final source images
The testing result of picture, the mathematical definition that i-th layer of laplacian pyramid:
G in formulaiIndicate the i-th tomographic image, UP operations are by (the 2x of pixel-map that position in source images is (x, y) to target image
+ 1,2y+1) position, symbolIndicate convolution algorithm, g5×5Indicate the Gaussian kernel of 5*5.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810348592.4A CN108550168A (en) | 2018-04-18 | 2018-04-18 | The multiple target quick high accuracy of Intelligentized mechanical arm view-based access control model identifies and localization method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810348592.4A CN108550168A (en) | 2018-04-18 | 2018-04-18 | The multiple target quick high accuracy of Intelligentized mechanical arm view-based access control model identifies and localization method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108550168A true CN108550168A (en) | 2018-09-18 |
Family
ID=63515249
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810348592.4A Pending CN108550168A (en) | 2018-04-18 | 2018-04-18 | The multiple target quick high accuracy of Intelligentized mechanical arm view-based access control model identifies and localization method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108550168A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112809672A (en) * | 2020-12-31 | 2021-05-18 | 安徽飞凯电子技术有限公司 | Target positioning system for mechanical arm |
CN113759928A (en) * | 2021-09-18 | 2021-12-07 | 东北大学 | Mobile robot high-precision positioning method for complex large-scale indoor scene |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102654902A (en) * | 2012-01-16 | 2012-09-05 | 江南大学 | Contour vector feature-based embedded real-time image matching method |
US10643101B2 (en) * | 2015-07-09 | 2020-05-05 | Texas Instruments Incorporated | Window grouping and tracking for fast object detection |
-
2018
- 2018-04-18 CN CN201810348592.4A patent/CN108550168A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102654902A (en) * | 2012-01-16 | 2012-09-05 | 江南大学 | Contour vector feature-based embedded real-time image matching method |
US10643101B2 (en) * | 2015-07-09 | 2020-05-05 | Texas Instruments Incorporated | Window grouping and tracking for fast object detection |
Non-Patent Citations (2)
Title |
---|
庞博,杨夏: "基于单模板的平面多目标识别与定位方法", 《豆丁网:HTTPS://WWW.DOCIN.COM/P-1990902723.HTML》 * |
赵小川: "《现代数字图像处理技术提高及应用案例详解 MATLAB版》", 30 April 2012, 北京航空航天大学出版社 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112809672A (en) * | 2020-12-31 | 2021-05-18 | 安徽飞凯电子技术有限公司 | Target positioning system for mechanical arm |
CN113759928A (en) * | 2021-09-18 | 2021-12-07 | 东北大学 | Mobile robot high-precision positioning method for complex large-scale indoor scene |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111738261B (en) | Single-image robot unordered target grabbing method based on pose estimation and correction | |
CN108665508B (en) | Instant positioning and map construction method, device and storage medium | |
US7957557B2 (en) | Tracking apparatus and tracking method | |
CN110211180A (en) | A kind of autonomous grasping means of mechanical arm based on deep learning | |
CN109448326B (en) | Geological disaster intelligent group defense monitoring system based on rapid image recognition | |
CN109045676B (en) | Chinese chess recognition learning algorithm and robot intelligent system and method based on algorithm | |
US9361695B2 (en) | Method of recognizing a position of a workpiece from a photographed image | |
KR101460313B1 (en) | Apparatus and method for robot localization using visual feature and geometric constraints | |
CN115147488B (en) | Workpiece pose estimation method and grabbing system based on dense prediction | |
Li et al. | Aruco marker detection under occlusion using convolutional neural network | |
Foresti et al. | Automatic visual recognition of deformable objects for grasping and manipulation | |
CN108550168A (en) | The multiple target quick high accuracy of Intelligentized mechanical arm view-based access control model identifies and localization method | |
CN108388854A (en) | A kind of localization method based on improvement FAST-SURF algorithms | |
EP3825804A1 (en) | Map construction method, apparatus, storage medium and electronic device | |
Patel et al. | Vision-based object classification using deep learning for inventory tracking in automated warehouse environment | |
Mišeikis et al. | Two-stage transfer learning for heterogeneous robot detection and 3d joint position estimation in a 2d camera image using cnn | |
Sharma et al. | An improved feature matching technique for stereo vision applications with the use of self-organizing map | |
CN117381793A (en) | Material intelligent detection visual system based on deep learning | |
CN115533895B (en) | Two-finger manipulator workpiece grabbing method and system based on vision | |
CN108197540A (en) | A kind of fire image Feature extraction and recognition method based on SURF | |
Shi et al. | A fast workpiece detection method based on multi-feature fused SSD | |
JP6041710B2 (en) | Image recognition method | |
Micheal et al. | Comparative analysis of SIFT and SURF on KLT tracker for UAV applications | |
Patil et al. | Sift based approach: Object recognition and localization for pick-and-place system | |
Stefańczyk et al. | Mixing deep learning with classical vision for object recognition |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20180918 |
|
WD01 | Invention patent application deemed withdrawn after publication |