CN109357630B - A kind of polymorphic type batch workpiece vision measurement system and method - Google Patents
A kind of polymorphic type batch workpiece vision measurement system and method Download PDFInfo
- Publication number
- CN109357630B CN109357630B CN201811273893.1A CN201811273893A CN109357630B CN 109357630 B CN109357630 B CN 109357630B CN 201811273893 A CN201811273893 A CN 201811273893A CN 109357630 B CN109357630 B CN 109357630B
- Authority
- CN
- China
- Prior art keywords
- workpiece
- image
- coordinate
- motion platform
- measurement
- 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
- G01—MEASURING; TESTING
- G01B—MEASURING LENGTH, THICKNESS OR SIMILAR LINEAR DIMENSIONS; MEASURING ANGLES; MEASURING AREAS; MEASURING IRREGULARITIES OF SURFACES OR CONTOURS
- G01B11/00—Measuring arrangements characterised by the use of optical techniques
- G01B11/24—Measuring arrangements characterised by the use of optical techniques for measuring contours or curvatures
Abstract
The present invention provides a kind of polymorphic type batch workpiece vision measurement system and methods, belong to industrial detection automatic field.Measuring system of the invention is by including that the part such as data processing terminal, workpiece sensing unit, image acquisition unit and control unit forms.Measurement method of the invention includes judging whether workpiece for measurement is put into detection platform, obtains partial-depth image, obtains global image, identifies workpiece type and quantity, dimensional parameters measurement required by completion measurement task and etc..The present invention can simultaneously detect multiple types, multi-quantity workpiece simultaneously, compared to conventional measuring methods, measurement workpiece type, quantity, in terms of be all significantly increased.
Description
Technical field
The present invention relates to the technologies such as image procossing, machine vision, motor control and sensor, belong to Machine Vision Detection survey
Amount field.It more specifically, is one kind using image procossing as core, the vision measurement system based on pattern-recognition, specially one
Kind polymorphic type workpiece vision measurement system and method.
Background technique
With the arrival in Flexible Production epoch, manufacture gradually to customization, multiple types direction transformation, not only into one
The diversity for walking larger production shape, to measuring system rapidity and versatility, more stringent requirements are proposed.It is applied to work at present
There are mainly two types of the modes of industry measurement: contact and contactless.It is most widely used to be in contact measurement system equipment
Three coordinate measuring machine, but the equipment is relatively high to environment and personnel requirement, detection speed is slow, easily scuffing measured surface.?
Contactless measuring system, it is most widely used in field of industry detection by the detection device of core of vision, such as Digital Photogrammetric System,
Three-D scanning measuring system etc., but most of for a certain workpiece sensing, simultaneously to multi-panel assembly or polymorphic type workpiece
Detection can not adapt to, and need human assistance to detect, detection efficiency is lower.
Main problem existing for current existing molding visual detection equipment: (1) mainly for some small parts and
The dimensional measurement of sheet-like workpiece needs Multiple-Scan to be even unable to measure when measuring large-size workpiece, such as Chinese patent Shen
Please numbers 201510111403.8, denomination of invention are as follows: a kind of metal plate vision inspection apparatus, the program are equipped with bracket, sliding groove, servo
Motor, source of parallel light, video camera, transparency glass plate, image processor, work stage and light source, wherein video camera is set to work stage
Underface and direction is upward, transparency glass plate are placed in work stage top, and metal plate is put on transparency glass plate and is located at video camera just
Top.The program can only be smaller by dimensions, and the sheet metal work of lighter weight is placed on transparent glass, video camera glass just
Lower section shooting obtains image, but glass plate can not bear heavier work pieces, inconvenient vulnerable to abrasion follow-up maintenance, through glass plate
Shooting easily causes pattern distortion to influence detection accuracy, is not easy to expand glass plate and single camera when measuring large-size workpiece
Coverage is limited；(2) for the measurement of single kind workpiece, vision measurement efficiency is higher, but simultaneously for multiple types workpiece
Measurement, the efficiency of vision-based detection are also unable to reach the beat of on-line checking, such as Chinese Patent Application No.: 201310496405.4,
Denomination of invention are as follows: batch part automatic detection system and method based on machine vision, method are as follows: geometry first is carried out to workpiece
Amount detection；Using acquired image information as template, the related measurement operation processing information of part is stored in system database, for
Subsequent parts are called automatically when measuring；Workpiece is placed on transparent operation platform, makes the alignment lens workpiece, sentences after workpiece imaging
Break the similitude of itself and template part information；It finally carries out automatic measurement and judges qualification, output measurement report.Program skill
Art carries out template matching by pre-stored workpiece Template Information and identifies workpiece for measurement type, and single detection can only identify one
Kind workpiece is detected, and is not able to satisfy multiple types workpiece while being detected, for a large amount of work piece match, needs to establish huge mould
Plate library, matching process consume more times, and matching precision is difficult to ensure, reduces measurement efficiency, and flexibility is not high enough；(3)
The multi-panel measurement of certain specific shape workpiece can only be met, application range has certain limitation, it is sometimes desirable to manually adjust inspection
Survey face is lacking in intelligence degree, such as Chinese Patent Application No.: 201710328284.0, denomination of invention are as follows: multiaxis connection
Dynamic vision, laser combined type non-contact measurement device for measuring and measurement method, structure are that workpiece is arranged on two axis mobile platforms
Turntable is placed, the elevating mechanism moved up and down, visual detection equipment and non-contact distance are set on the top that workpiece places turntable
Measuring device is arranged on elevating mechanism, and visual detection equipment position is fixed in the program, can only by adjusting measurement angle and
The side-information of revolving-turret position acquisition workpiece, but circle contained for side, the workpiece of the shape features such as triangle, by
In certain deformation can be caused to image information in tilt angle scanning, workpiece size cannot be accurately detected.
Summary of the invention
1. technical problems to be solved by the inivention
Do not have versatility in view of existing molding visual detection equipment, to the work of different sizes, shape and complexity
The problem of part is unable to automatic identification and is unable to batch detection, the present invention propose a kind of polymorphic type batch workpiece vision measurement system
And method, it is a kind of high-freedom degree measuring system that can satisfy polymorphic type workpiece while measurement, which is able to carry out single
The measure batch of type and polymorphic type workpiece.
2. technical solution
In order to achieve the above objectives, technical solution provided by the invention are as follows:
A kind of polymorphic type workpiece vision measuring method, comprising the following steps:
Step 1, activation system judge whether workpiece for measurement is put into workpiece calibration platform and determines workpiece in workpiece calibration
The position of platform, if detecting, workpiece is put into workpiece calibration platform and thens follow the steps two, and otherwise whole system is in standby shape
State；
Step 2, is scanned to workpiece for measurement by image acquisition unit and obtains the topography of each section, if sweeping
The topography for retouching acquisition does not obtain sufficient depth information, then scans again after adjusting the position of image acquisition unit, finally
Obtain partial-depth image；
Step 3, system carry out splicing to the partial-depth image of acquisition, obtain whole scan image；
Step 4 handles whole scan image, identifies that workpiece type and quantity, system are combined according to recognition result
The required parameter measured of parameter task list to be measured, calls picture size processing routine to the workpiece image information of acquisition in system
It is handled, calculates corresponding dimensional parameters, completed dimensional parameters required by measurement task and measure；
Step 5, the workpiece parameter and standard parameter information obtained by above-mentioned steps compare, and qualified workpiece is incoming in next step
Process, underproof workpiece are rejected or are marked by mechanical arm handgrip.
As prioritization scheme, in step 3, according to the number of image acquisition unit by the detection zone on workpiece calibration platform
Domain is divided into several sub-regions, is separated between the subregion with markings, the scanning field of view weight of adjacent image acquiring unit
Folded part covers markings, in image mosaic processing, multiple attribute blocks is put on markings, extract the feature of attribute block
Information carries out image mosaic using plane right-angle coordinate coordinate transformation model.
As prioritization scheme, the characteristic information for extracting attribute block is to extract circle marker using Hough transform algorithm
The central coordinate of circle and radius of block, specific steps are as follows:
(1) the attribute block image of acquisition denoise and binaryzation, the denoising are rolled up using Gaussian filter
Product is to image denoising；
(2) attribute block edge contour is extracted using Canny edge detection algorithm, determines the marginal point of Hough transform；
(3) to each non-zero marginal point of edge contour, the Sobel first derivative on the direction x, y is calculated in its neighborhood
Obtain gradient:
If a certain non-zero marginal point pixel value is I (x, y), derivation is carried out using Sobel operator:
Change in the x direction, may be expressed as:
Change in y-direction, may be expressed as:
It usesCalculate gradient magnitude and direction；
(4) multiple marginal points, the search radius R of setting search center location are randomly selectedS, then in each marginal point ladder
It spends on direction apart from marginal point distance RsBoth sides respectively throw a bit, drawn in Hough two-dimensional space all centered on marginal point
Institute's cultellation be endpoint gradient line segment, the point that line segment is passed through chooses the maximum point of accumulator value and sits in accumulator number
Mark is central coordinate of circle, according to central coordinate of circle C (xc,yc) and marginal point coordinate I (x, y) find out radius of circle.
As prioritization scheme, described image splicing realizes that steps are as follows using plane right-angle coordinate coordinate transformation model:
(1) central coordinate of circle and radius of the circle marker block obtained according to the method for above-mentioned Hough transform is straight using plane
Angular coordinate system coordinate transformation model carries out Unitary coordinate, the coordinate transformation model to several scanning figures are as follows:
In formula, △ x, △ y is translation parameters, and m is scale factor, and α is rotation parameter, and (x, y), (x', y') are same point
Point coordinate under Different Plane rectangular coordinate system, Yao Shixian coordinate points must find out △ x in different coordinates conversion, △ y, m,
This four parameters of α, wherein △ x, △ y are easy to acquire, therefore at least containing there are two common points to be convenient for above-mentioned seat between every two width figure
Mark transformation model parametric solution；
(2) according to the coordinate transformation model of solution, next picture successively carries out the coordinate system that coordinate is transformed into a width figure
In, under finally unifying each scanning information to global coordinate system and carry out image co-registration elimination splicing gap, obtaining whole scan
Image.
As prioritization scheme, in step 3, described image splicing uses following steps: described image acquiring unit is laser
Scanner, several laser scanner progressive scans, the collected every data line of every laser scanner distinguish Coutinuous store
In memory address piecemeal, memory address and data source are corresponded, it is known that every data line of every laser scanner acquisition
First address obtains the offset that data store according to field of view width is overlapped, connects since the first address of specified each row of data
It resumes studies and takes image data, find the data storage location of repetition pixel and refused to read, thus in image data read loop
Section carries out image mosaic, obtains whole scan image.
As prioritization scheme, the concrete operations of step 4 are as follows:
(1) global image is pre-processed, target area segmentation, obtains various workpiece target regions；
(2) edge detection successively is carried out to each workpiece target region, obtain clearly edge contour and intercepted, to workpiece wheel
Wide extracted region outer contoured features identify that the affiliated major class of workpiece is completed workpiece and known by the template matching method based on feature
Then other rough sort calls trained BP neural network category identification model according to affiliated major class, extract workpiece profile
Minutia is classified again by the BP neural network category identification model of affiliated major class, determines workpiece specific category, complete
At the disaggregated classification of workpiece identification；It identifies workpiece for measurement type, counts various types of quantity, count the corresponding image of each workpiece
Processing routine number；
(3) system intercepts each workpiece target region, and the number of the processing routine according to corresponding to workpiece integrates library from processing routine
It is middle to call corresponding workpiece image processing routine；
(4) system obtains parameter task list to be measured by remote service end；
(5) system carries out workpiece targets all in global image region corresponding simultaneously by the way of parallel computation
Image processing program obtains items dimensional parameters information required by task list in conjunction with parameter task list to be measured.
It is described as follows to the pretreated process of global image progress as prioritization scheme:
(1) using the method for gaussian filtering to whole scan image global de-noising；
(2) binary conversion treatment is carried out using the method for fixed threshold again, makes workpiece area and background separation；
(3) first the workpiece image Jing Guo binary conversion treatment is expanded, is corroded using the method for morphological operation, smoothly
Image border；
The edge detection obtains the edge contour of workpiece using the method for Canny edge detection, is convenient for subsequent feature
It extracts；
The contour feature selects Hu moment characteristics, and the treatment process based on Hu moment characteristics template matching is as follows:
(1) image to be identified is pre-processed, the profile Hu square for obtaining workpiece image according to feature extraction algorithm is special
Sign；
(2) template image in image to be identified and template library is calculated to the similarity measure of feature, selection is suitably sentenced
Threshold value is determined, therefore, it is determined that the affiliated major class of workpiece out；
Disaggregated classification is completed using workpiece profile minutia, minutia selects small echo moment characteristics, and extraction process is as follows:
(1) workpiece image in each workpiece target region is normalized: sets the workpiece image that f (x, y) is two-dimensional discrete,
The center-of-mass coordinate of image is determined first:Secondly zoom factor α is defined, evenIt takes α=N/ (4* γ), the normalization of workpiece image can be converted to by f (x, y)It carries out；
(2) image Polar coordinates: for continuous function f (x, y), corresponding polar coordinates function representation is f (r, θ),
Middle x=rcos θ, y=rsin θ, r are polar diameter, and θ is polar angle, since two-dimensional digital image is discrete, so being turned by rectangular co-ordinate
When being changed to polar coordinates, equal part need to be carried out to angle, take Δ θ=2 π/N to carry out discretization, (p+q) rank moment characteristics definition of image is such as
Formula: Fpq=∫ ∫ f (r, θ) gp(r) ejq θ drd θ, wherein gp(r) it is function about radius r, p, q are numeric parameter；
(3) FFT transform: (p+q) rank moment characteristics definition of image can be further represented as Fpq=∫ Sq(r)gp(r) rdr,
Again because being the one-dimensional sequence about variable r, and what is actually carried out is discrete transformation, therefore need to be integral discretization: being chosen suitable
When angle interval delta θ=2 π/N, then angle integrate are as follows:N indicates transformation points, m
The constant for indicating [0, N-1], is realized using FFT；
(4) feature is extracted in radial zone { 0≤r≤1 } using wavelet function to the integral result in above-mentioned steps, leads to
Cross gp(p) wavelet structure square, wavelet function will axially be defined as ψj,k=2j/2 ψ (2jr-k)；Select different j and k just
The global characteristics and local feature of available image；The definition of wavelet moment invariants is | | Fj,k,q| |=| | ∫ Sq(r)ψj,k(r)
Rdr | |, wherein q=0 in formula, 1,2 ..., for the r of a certain fixation, Sq(r)=∫ f (r, θ) ejq θ d θ, represent image f (r,
θ) q-th of characteristic quantity in phase space [0,2 π]；Utilize different scale factor j, shift factor k, it will be able to provide | | Fj,k,q|
| the feature in different scale level, noteDj is high frequency
Minutia, Sj is low-frequency information feature, with Dj, SjThe amount of being characterized is identified, D is selected4Value and S when taking k=1,2,3,45
Take k=1, value when 2, totally six values are used as characteristic quantity；
Classification for the wavelet moment characteristic value of workpiece image constructs disaggregated model using single hidden layer configuration BP neural network,
Including input layer, hidden layer, output layer, wherein 6 nodes of input layer, 6 characteristic values of corresponding workpiece image wavelet moment are corresponding
Workpiece type is the desired output of network, and the workpiece classification number that output layer number of nodes is distinguished by the network determines, the node of hidden layer
Number is determined by empirical equation:Wherein k is the number of hidden nodes, and n is input layer number, and m is output node layer
Number, a are the regulating constant between 1~10, and BP neural network model is by the small echo moment characteristics of training sample as BP nerve net
The input of network carries out network model training, obtains workpiece disaggregated model；
The training step of BP neural network model is as follows:
(1) workpiece to be identified is chosen as sample workpiece, by each workpiece respectively every 15 degree, i.e., at positive and negative seven degree
In the range of carry out Image Acquisition；
(2) feature calculation is carried out to several images that Image Acquisition obtains, then counted in different illumination and apart from item
Mean value under part establishes training sample database for BP neural network model；
(3) BP neural network model is trained, each weight is adjusted using the method for gradient decline；
(4) trained BP neural network model is imported into application program, feature is extracted to the testing image comprising workpiece
It carries out Classification and Identification and records various types of quantity.
As prioritization scheme, in step 4, picture size processing routine obtains dimensional parameters, and specific step is as follows:
(1) edge detection is carried out to workpiece target image, extracts the shape edges such as straight line, circle, rectangle, ellipse in workpiece
Profile；
(2) to each shape edges contour fitting, calculate the distance between edge pixel, in conjunction with image pixel distance and reality away from
From relationship, calculate each dimensional parameters.
As prioritization scheme, including data processing terminal, workpiece sensing unit, image acquisition unit, control unit；
The data processing terminal is for storing and processing data information；
The workpiece sensing unit is for placing workpiece for measurement and starting detection system；
Described image acquiring unit is for obtaining partial-depth image；
Described control unit includes motion platform control unit and handgrip control unit；
The motion platform control unit includes motion control card and motor driver, industry control built in the motion control card
At mainboard pci card slot, for issuing different frequency, different number pulse to driver, the motor driver passes through signal
Line and motion control cartoon letters are for receiving pulse to control the speed of motor movement, direction, position；
The handgrip control unit is used to control the movement of grabbing workpiece；
The workpiece sensing unit, image acquisition unit, motion control unit and motor driver respectively with data processing
Terminal called connection.
As prioritization scheme, the workpiece sensing unit includes work measurement platform (3) and several diffuse and establish by cable
Close (2), workpiece calibration platform (3) surface sets detection zone, described several diffuse electric switch (2) it is equidistant, at
One row is arranged in workpiece calibration platform (3) edge；
Described image acquiring unit is laser scanner (1)；
The motion platform control unit is Three-degree-of-freedom motion platform, and the handgrip control unit includes mechanical arm handgrip
(4) it is fixed on manipulator motion track (5) with manipulator motion track (5), the mechanical arm handgrip (3)；
The Three-degree-of-freedom motion platform includes X-axis motion platform track (6), Y-axis motion platform track (7) and Z axis fortune
Moving platform track (8)；The X-axis motion platform track (6) and manipulator motion track (5) are arranged in parallel in the horizontal direction respectively
In the two sides of the work measurement platform (3), the Z axis motion platform track (8) is arranged in a vertical direction to be moved in the X-axis
On platform track (6), the Y-axis motion platform track (7) is arranged along the direction perpendicular to the X-axis motion platform track (6)
On the Z axis motion platform track (8)；The spaced apart linear motor of several segments is set on the Y-axis motion platform track (7)
Motion guide rail, several described laser scanners (1) are arranged on each section of linear motor motion guide rail, described unrestrained anti-
The direction in a row for penetrating optoelectronic switch (2) is vertical with the length direction of the X-axis motion platform track (6).
3. beneficial effect
Using technical solution provided by the invention, compared with existing well-known technique, there is following remarkable result:
(1) present invention can start measuring system by sensor technology automatically, by the depth information of surface sweeping workpiece from
Dynamic adjustment surface sweeping height, reduces manual intervention measurement links, has saved human cost.The present invention is by obtaining partial depth map
As obtaining whole scan image, by the workpiece type and quantity of whole scan image automatic identification test object, eventually by
Corresponding image processing program is called to complete measurement.The present invention can simultaneously detect multiple types, multi-quantity workpiece simultaneously, compare
In conventional measuring methods, measure workpiece type, quantity, in terms of be all significantly increased.
(2) present invention extracts the central coordinate of circle of multiple circle marker blocks by Hough transform algorithm, and Calculation Plane right angle is sat
Translation parameters △ x, △ y, scale factor m, rotation parameter α in mark system coordinate transformation model complete each topography's coordinate system
One, to obtain the global image of large-size workpiece, compared to the image mosaic technology by a large amount of Feature Points Matchings, has and calculate
Measure the advantages that small and splicing precision is high.
(3) present invention cannot identify simultaneously for existing vision detection system and detect multiple types workpiece and batch workpiece
The problem of, by extracting outer contoured features, first carries out feature templates and match the determining affiliated major class of workpiece, then by extracting work
Part image wavelet invariant moment features determine each workpiece specific category using BP neural network disaggregated model.The method overcome biographies
Excessively huge based on appearance profile template library in system scheme, time-consuming for workpiece identification, has difference work to exterior contour similar interior
Detection speed can be improved by the method for secondary classification in the disadvantages of identification accuracy of part is not high, the present invention, reduces to having
The BP for only needing the affiliated major class of re -training is added in new workpiece for the False Rate of identical appearance contoured interior details difference workpiece
Neural network model.
(4) the invention also provides a kind of polymorphic type batch workpiece vision measurement systems, for existing workpiece vision-based detection
Equipment single shot or the case where being limited in scope, need repeatedly acquisition for large-size workpiece or batch workpiece of scanning, pass through Y
The structure of the multiple laser scanners of axis motion platform track carry increases single sweep operation width, can disposably complete a wide range of
The workpiece of detection zone scans, to improve detection efficiency and can adjust laser scanner number according to on-site test task
Amount has certain expansibility, and in addition the present invention is that system can be according to workpiece size amount using Three-degree-of-freedom motion platform
Grade and effective viewing field's adjust automatically of laser scanner detect height, further promote the adaptation range of detection system.
(5) present invention installs multiple electric switches that diffuse by detection platform edge, according to each diffusing reflection switch
It blocks signal and differentiates whether workpiece is put into detection platform, sentence compared to by artificial judgment or by the image shot by camera of auxiliary
It is disconnected, save human cost and fund cost certain；The scanner carry device that the present invention uses, can be according to demand and scene
The various types of image acquisition equipments of environment carry, system structure have preferable expansion.The present invention executes machine using motor
Structure carry image acquisition equipment, mode identification technology carry out identification classification to workpiece, and image processing techniques obtains workpiece size, structure
The high flexibility measuring system built can measure the multiple workpiece of multiple types simultaneously, increase while measuring piece count, promote measurement work
Part specific diversity.
Detailed description of the invention
Fig. 1 is measuring method flow chart of the present invention；
Fig. 2 is measuring system structure chart of the present invention；
Fig. 3 is workpiece sensing unit control structure figure；
Fig. 4 is detection platform functional structure chart；
Fig. 5 is detection system hardware configuration main view；
Wherein: 1- laser scanner, 2- diffuse electric switch, 3- workpiece calibration platform, 4- mechanical arm handgrip, and 5- is mechanical
Arm tracks, 6-X axis motion platform track, 7-Y axis motion platform track, 8-Z axis motion platform track, the fixed Y-axis of 9- are oblique
Tie rod, 10- markings.
Specific embodiment
Embodiment 1
As shown in Figures 4 and 5, a kind of polymorphic type batch workpiece vision measurement system, including data processing terminal, workpiece sensing
Unit, image acquisition unit, control unit；
Data processing terminal is for storing and processing data information；
Workpiece sensing unit is for placing workpiece for measurement and starting detection system；
Image acquisition unit is for obtaining partial-depth image；
Control unit includes motion platform control unit and handgrip control unit；
Motion platform control unit includes motion control card and motor driver, industrial control computer mainboard built in motion control card
At pci card slot, for issuing different frequency, different number pulse to driver, the motor driver passes through signal wire and fortune
Dynamic control cartoon letters are for receiving pulse to control the speed of motor movement, direction, position；
Handgrip control unit is used to control the movement of grabbing workpiece；
Workpiece sensing unit, image acquisition unit, motion control unit and motor driver respectively with data processing terminal
Communication connection, as shown in Figure 2.
Workpiece sensing unit includes work measurement platform 3 and the electric switch 2 that diffuses, and 3 surface of workpiece calibration platform sets inspection
Region is surveyed, several electric switches 2 that diffuse are equidistant, in a row are arranged in 3 edge of workpiece calibration platform；Image obtains
Taking unit is laser scanner 1；Detection zone includes multiple subregions, using markings 10 as the separation sign in adjacent subarea domain
Will.
Motion platform control unit is Three-degree-of-freedom motion platform, and handgrip control unit includes mechanical arm handgrip 4 and machinery
Arm tracks 5, mechanical arm handgrip 3 are fixed on manipulator motion track 5；
Three-degree-of-freedom motion platform includes X-axis motion platform track 6, Y-axis motion platform track 7 and Z axis motion platform rail
Road 8；X-axis motion platform track 6 and manipulator motion track 5 are arranged in parallel in work measurement platform 3 in the horizontal direction respectively
Two sides, Z axis motion platform track 8 are arranged in a vertical direction on X-axis motion platform track 6, and Y-axis motion platform track 7 is along vertical
Directly it is arranged on Z axis motion platform track 8 in the direction of X-axis motion platform track 6；Several segments are set on Y-axis motion platform track 7
Spaced apart linear motor motion guide rail, several laser scanners 1 are arranged on each section of linear motor motion guide rail,
Several laser scanners ensure in same level, and wherein scanner carry device is one to be connected to motor slide rails mobile
The pinboard of pedestal, laser scanner is connected to data processing terminal by data line, and data transfer rate is handled
Terminal.Diffuse electric switch 2 direction in a row it is vertical with the length direction of X-axis motion platform track 6.Motor driver respectively drives
Mobile interface is connected by power line with the Mobile Slide in every section guide rail.
Three-degree-of-freedom motion platform is preferably Three Degree Of Freedom single-cantilever motion platform, relative to gantry structure motion platform,
The open batch workpiece convenient for each dimension scale in three face of single-cantilever structure in single-cantilever motion platform, which is put and is easy to expand, to be prolonged
Long detection range, Three Degree Of Freedom single-cantilever motion platform can dimension scale according to examined workpiece and scanner effective viewing field,
Measurement height and measurement range are freely adjusted, relative to general two axle movement platform, scanner distance is detected object height distance
The fixed measuring system that limits is applicable in workpiece size range.Three Degree Of Freedom single-cantilever motion platform is by X, Y, Z triaxial movement platform
Track composition, wherein Y-axis motion platform track is carry scanner cantilever, has multistage and the movement of disjunct linear motor is led
Rail, several laser scanners are distinguished in the every section guide rails of carry, and every laser scanner can flexibly move linearly, according to required precision
Each scanner spacing is freely adjusted with the maximum sweep width of laser scanner；Produced by eliminating cantilever during the motion
Shake, between Z axis motion platform track and Y-axis motion platform track install a cantilever mounted batter brace 9, constitute triangle
Rock-steady structure.
Data processing terminal stores and processs Various types of data for integrating each image processing module and workpiece identification module
Information controls motor movement, and laser scanner, mechanical arm handgrip etc., data processing terminal includes: workpiece template library module, figure
Picture processing module, motor and mechanical arm handgrip control module, data information memory and transmission module and display module.
It can move along a straight line along the x axis on Y-axis motion platform track, X, Y, Z axis are driven using linear motor；Several are swashed
Photoscanner 1 can be four, and four laser scanners 1 pass through four mechanical equidistant carries of pinboard on cantilever, each
Pinboard is connect with four linear motor rotor seats, and Y-axis, which is equidistantly divided between four section guide rails and guide rail, to be not attached to, in order to Z
The spacing that each scanner is adjusted after axis height change guarantees there is overlapped fov between each scanner；Cantilever can be above and below Z-direction
Linear movement adjustment scanning height, adjusts field range convenient for laser scanner, to detect higher workpiece.X, Y, Z axis movement
Limit switch is installed at the both ends of platform track respectively, beyond stroke range when preventing mobile.X-axis and Z axis installation grating scale and Y-axis
Every section guide rail all installs grating scale, in order to real-time detection three-axis moving speed, position.
Measuring table smooth surface is wear-resisting, does not deform vulnerable to the variation of temperature and humidity.At platform edges
One row is installed equidistantly to diffuse electric switch, quantity and spacing can be adjusted according to minimum workpiece size, and diffuse electric switch
Effect can not only detect whether workpiece is put into platform but also can determine that the placement position of workpiece.
A kind of polymorphic type workpiece vision measuring method, as shown in Figure 1, comprising the following steps:
S1: activation system, the electric switch 2 that each diffuses transmitting laser beam can be blocked when workpiece is put into measuring table
One or more laser beams so that causing one or more optoelectronic switches issues low and high level signal, and are passed by connecting terminal
Enter single-chip microcontroller, single-chip microcontroller carries out coded treatment to incoming low and high level signal and is passed to data processing terminal, obtains the position of workpiece
It sets, detection system is started with this, as shown in Figure 3.
S2: judging whether workpiece for measurement is put into workpiece calibration platform and determining workpiece in the position of workpiece calibration platform, if
Detect that workpiece is put into workpiece calibration platform and thens follow the steps two, otherwise whole system is in standby；
Each trigger signal for diffusing electric switch 2 is numbered in data processing terminal, when workpiece puts workpiece
The electric switch 2 that diffuses accordingly can be triggered on measuring table, then system determines that workpiece is swept in corresponding laser according to output signal
It retouches in instrument detection path, starts corresponding laser scanner 1.
Each detection subregion has the laser beam for the electric switch sending that diffuses accordingly, and workpiece is put into detection platform
Afterwards, a certain number of laser beams can be blocked, photoelectric sensor number can determine whether which sub-district workpiece is placed according to being triggered
Domain, the corresponding laser scanner of each sub-regions, then only open corresponding laser scanner when detecting when starting, can be with
The image information for reducing redundancy, quickly handles data convenient for data processing terminal.
S3: if there is workpiece to be put into measuring table, it is passed to the location information of workpiece into system, starts three-degree-of-freedom motion
Platform and laser scanner, carry measure highly according to default to workpiece calibration in the laser scanner of Y-axis motion platform track
Platform is scanned and obtains partial-depth image, if the partial-depth image that scanning obtains does not obtain sufficient depth information,
Then the scanning height of adjust automatically Y-axis motion platform track scans again；Determine whether to obtain the foundation of sufficient picture depth:
Laser scanner has specific field range, obtains a cloud number if workpiece for measurement height is beyond surface sweeping instrument vertical field of view range
According to be empty, the height of measurement Y-axis motion platform track is increased automatically as principle system；
S4: the partial-depth image that system will acquire carries out splicing, obtains whole scan image；
The detection zone on workpiece calibration platform is divided into several sub-regions, institute according to the number of image acquisition unit
It states and is separated between subregion with markings, the part of the scanning field of view overlapping of adjacent image acquiring unit covers markings, is scheming
As being put into multiple attribute blocks on markings, the characteristic information of attribute block being extracted, using plane right-angle coordinate in splicing
Coordinate transformation model carries out image mosaic；Attribute block uses circle marker, and circle marker block is easy to extract edge contour and then intends
Close out central coordinate of circle information；
The characteristic information for extracting attribute block is the central coordinate of circle and half that circle marker block is extracted using Hough transform algorithm
Diameter, specific steps are as follows:
(1) the attribute block image of acquisition denoise and binaryzation, the denoising are rolled up using Gaussian filter
Product is to image denoising；
(2) attribute block edge contour is extracted using Canny edge detection algorithm, determines the marginal point of Hough transform；
(3) to each non-zero marginal point of edge contour, the Sobel first derivative on the direction x, y is calculated in its neighborhood
Obtain gradient:
If a certain non-zero marginal point pixel value is I (x, y), derivation is carried out using Sobel operator:
Change in the x direction, may be expressed as:
Change in y-direction, may be expressed as:
It usesCalculate gradient magnitude and direction, the i.e. normal of circular arc where the point
Direction and slope.
(4) multiple edge contour points, the search radius RS of setting search center location, then at each edge are randomly selected
Both sides on profile point gradient direction apart from edge contour point distance Rs respectively throw a bit, Hough two-dimensional space draw it is all with
Institute's cultellation centered on marginal point is the gradient line segment of endpoint, and the point that line segment is passed through chooses accumulator in accumulator number
It is worth maximum coordinate i.e. central coordinate of circle, radius of circle is found out according to central coordinate of circle C (xc, yc) and marginal point coordinate I (x, y).
Image mosaic realizes that steps are as follows using plane right-angle coordinate coordinate transformation model:
(1) central coordinate of circle and radius of the circle marker block obtained according to the method for above-mentioned Hough transform is straight using plane
Angular coordinate system coordinate transformation model carries out Unitary coordinate, the coordinate transformation model to several scanning figures are as follows:
In formula, △ x, △ y is translation parameters, and m is scale factor, and α is rotation parameter, (x, y),For same point
Point coordinate under Different Plane rectangular coordinate system, Yao Shixian coordinate points must find out △ x in different coordinates conversion, △ y, m,
This four parameters of α, wherein △ x, △ y are easy to acquire, therefore at least common point is solved containing there are two between every two width figure；
(2) according to the transformation model of solution, next picture successively carries out coordinate and is transformed into the coordinate system of a width figure, most
Under each scanning information is unified to global coordinate system at last and carry out image co-registration elimination splicing gap, obtaining whole scan image.
S5: handling whole scan image, identifies workpiece type and quantity.
It identifies that the detailed process of workpiece type completes rough sort including the use of outer workpiece contour feature, recycles workpiece wheel
Wide minutia completes disaggregated classification；Excessively huge based on appearance profile template library in traditional scheme, the workpiece identification consumption in order to overcome
Duration has the shortcomings that the identification accuracy of difference workpiece is not high to exterior contour similar interior, utilizes outer workpiece contour feature
Rough sort is completed, workpiece profile minutia is recycled to complete disaggregated classification, it is according to needs of production that workpiece foundation is outer first
Profile similarity degree is divided into multiple major class, establishes the contour feature library of each major class, passes through outer profile feature quick of workpiece
With major class belonging to workpiece, workpiece profile minutia is recycled to carry out network instruction as the input of affiliated major class BP neural network
Practice, establishes the specific category identification model based on BP neural network.Detection can be improved by the method for secondary classification in the present invention
Speed reduces to the False Rate with identical appearance contoured interior details difference workpiece, weight is only needed when new workpiece is added
The BP neural network model of major class belonging to new training.
Detailed process are as follows:
(1) global image is pre-processed, target area segmentation, obtains various workpiece target regions；
(2) edge detection successively is carried out to each workpiece target region, obtain clearly edge contour and intercepted, to workpiece wheel
Wide extracted region outer contoured features identify that the affiliated major class of workpiece is completed workpiece and known by the template matching method based on feature
Then other rough sort calls trained BP neural network category identification model according to affiliated major class, extract workpiece profile
Minutia is classified again by the BP neural network category identification model of affiliated major class, determines workpiece specific category, complete
At the disaggregated classification of workpiece identification；It identifies workpiece for measurement type, counts various types of quantity, count the corresponding image of each workpiece
Processing routine number；
It is described as follows to the pretreated process of global image progress:
(1) using the method for gaussian filtering to whole scan image global de-noising；
(2) binary conversion treatment is carried out using the method for fixed threshold again, makes workpiece area and background separation；
(3) first the workpiece image Jing Guo binary conversion treatment is expanded, is corroded using the method for morphological operation, smoothly
Image border avoids workpiece image edge sawtoothization and reduces some unnecessary details；
Edge detection obtains the edge contour of workpiece using the method for Canny edge detection, is convenient for subsequent feature extraction；
Contour feature selects Hu moment characteristics, and Hu invariant moment features are upper in terms of rotation, translation, ratio to have stronger stabilization
Property, there is higher accuracy rate in the identification biggish workpiece of exterior contour difference, it can be with when identifying the similar workpiece of exterior contour
Ignore tiny difference, the present invention matches major class belonging to determining workpiece for measurement using rough segmentation is completed the characteristics of hu moment characteristics.It is based on
The matched treatment process of Hu feature templates is as follows:
(1) image to be identified is pre-processed, the profile Hu square for obtaining workpiece image according to feature extraction algorithm is special
Sign；
(2) template image in image to be identified and template library is calculated to the similarity measure of feature, selection is suitably sentenced
Threshold value is determined, therefore, it is determined that the affiliated major class of workpiece out.
Disaggregated classification is completed using workpiece profile minutia, minutia selects small echo moment characteristics, and extraction process is as follows:
(1) workpiece image is normalized: sets the workpiece image that f (x, y) is two-dimensional discrete, it is first determined the matter of image
Heart coordinate:Secondly zoom factor α is defined, general zoom factor defines method with anti-noise
Property difference disadvantage, adopted alternatively to promote noise immunity, evenTake α=
The normalization of N/ (4* γ), workpiece image can be converted to by f (x, y)It carries out；
(2) image Polar coordinates: for continuous function f (x, y), its corresponding polar coordinates function representation is f (r, θ), wherein
X=rcos θ, y=rsin θ, r are polar diameter, and θ is polar angle, since two-dimensional digital image is discrete, so being converted by rectangular co-ordinate
When for polar coordinates, equal part need to be carried out to angle, take Δ θ=2 π/N to carry out discretization, (p+q) rank moment characteristics definition of image is such as
Formula: Fpq=∫ ∫ f (r, θ) gp(r)ejqθDrd θ, wherein gp(r) it is function about radius r, p, q are numeric parameter；
(3) FFT transform: (p+q) rank moment characteristics definition of image can be further represented as Fpq=∫ Sq(r)gp(r) rdr,
Again because being the one-dimensional sequence about variable r, and what is actually carried out is discrete transformation, therefore need to be integral discretization: being chosen suitable
When angle interval delta θ=2 π/N, then angle integrate are as follows:N indicates transformation points, m
FFT realization can be used in the constant for indicating [0, N-1]；
(4) feature is extracted in radial zone { 0≤r≤1 } using wavelet function to the integral result in above-mentioned steps, leads to
Cross gp(p) wavelet structure square, wavelet function will axially be defined as ψj,k=2j/2ψ(2jr-k).Select different j and k can
To obtain the global characteristics and local feature of image.The definition of wavelet moment invariants is | | Fj,k,q| |=| | ∫ Sq(r)ψj,k(r)
Rdr | |, wherein q=0 in formula, 1,2 ..., for the r of a certain fixation, Sq(r)=∫ f (r, θ) ejqθD θ represents image f (r, θ)
Q-th of characteristic quantity in phase space [0,2 π].Utilize different scale factor j, shift factor k, it will be able to provide | | Fj,k,q||
Feature in different scale level, noteDj is that high frequency is thin
Save feature, SjFor low-frequency information feature, with Dj, SjThe amount of being characterized is identified, D is selected4Value and S when taking k=1,2,3,45Take k
Value when=1,2, totally six values are used as characteristic quantity；
Classification for the wavelet moment characteristic value of workpiece image constructs disaggregated model using single hidden layer configuration BP neural network,
Including input layer, hidden layer, output layer, wherein 6 nodes of input layer, 6 characteristic values of corresponding workpiece image wavelet moment are corresponding
Workpiece type is the desired output of network, and the workpiece classification number that output layer number of nodes is distinguished by the network determines, the node of hidden layer
Number is determined by empirical equation:Wherein k is the number of hidden nodes, and n is input layer number, and m is output node layer
Number, a are the regulating constant between 1~10, and BP neural network is by the small echo moment characteristics of training sample as BP neural network
Input carries out network model training, obtains workpiece disaggregated model.
The training step of BP neural network model is as follows:
(1) workpiece to be identified is chosen as sample workpiece, respectively by each workpiece (workpiece which step is handled)
Every 15 degree, i.e., Image Acquisition is carried out in the range of positive and negative seven degree；
(2) feature calculation is carried out to several images that Image Acquisition obtains, then counted in different illumination and apart from item
Mean value under part establishes training sample database for BP neural network model；
(3) training sample (previous step does not mention training sample) will be obtained to be trained BP neural network model,
Each weight is adjusted using the method for gradient decline；
(4) trained BP neural network model is imported into application program, feature is extracted to the testing image comprising workpiece
It carries out Classification and Identification and records various types of quantity.S6: being numbered according to the workpiece type of identification, from workpiece size processing routine
Different image processing programs is called in library.
S7: calculation processing is carried out to each workpiece image simultaneously, and obtains task parameters table from remote service end.
S8: according to Dimension Types to be measured each in task parameters table, from all size results for scheming to export in this processing routine
It is screened, obtains workpiece size parameter required by Detection task；
S9: the workpiece parameter and standard parameter information obtained by above-mentioned steps compares, and qualified workpiece is passed to the next step,
Underproof workpiece is rejected or is marked by mechanical arm handgrip.
Embodiment 2
Embodiment 2 and the scheme of embodiment 1 are essentially identical, and difference is only that, image mosaic uses following steps: the figure
Picture acquiring unit is scanner, and several scanners progressive scans, the collected every data line of every scanner is distinguished continuous
It is stored in memory address piecemeal, memory address and data source correspond, it is known that every data line of every scanner acquisition
First address obtains the offset that data store according to field of view width is overlapped, connects since the first address of specified each row of data
It resumes studies and takes image data, find the data storage location of repetition pixel and refused to read, thus in image data read loop
Section carries out image mosaic, obtains whole scan image.
Claims (9)
1. a kind of polymorphic type workpiece vision measuring method, which comprises the following steps:
Step 1, activation system judge whether workpiece for measurement is put into workpiece calibration platform and determines workpiece in workpiece calibration platform
Position, if detecting, workpiece is put into workpiece calibration platform and thens follow the steps two, and otherwise whole system is in standby；
Step 2 is scanned to workpiece for measurement by image acquisition unit and obtains the topography of each section, if scanning obtains
The topography obtained does not obtain sufficient depth information, then scans again after adjusting the position of image acquisition unit, final to obtain office
Portion's depth image；
Step 3, system carry out splicing to the partial-depth image of acquisition, obtain whole scan image；
The detection zone on workpiece calibration platform is divided into several sub-regions, the son according to the number of image acquisition unit
Separated between region with markings, the part of the scanning field of view overlapping of adjacent image acquiring unit covers markings, spells in image
It connects in processing, multiple attribute blocks is put on markings, extract the characteristic information of attribute block, using plane right-angle coordinate coordinate
Transformation model carries out image mosaic；
Step 4 handles whole scan image, identifies workpiece type and quantity, system is according to recognition result combination system
In the required parameter measured of parameter task list to be measured, call picture size processing routine to carry out the workpiece image information of acquisition
Processing calculates corresponding dimensional parameters, completes dimensional parameters required by measurement task and measures；
Step 5, the workpiece parameter and standard parameter information obtained by above-mentioned steps compare, and qualified workpiece is passed to the next step,
Underproof workpiece is rejected or is marked by mechanical arm handgrip.
2. polymorphic type workpiece vision measuring method according to claim 1, which is characterized in that
The characteristic information for extracting attribute block is the central coordinate of circle and half that circle marker block is extracted using Hough transform algorithm
Diameter, specific steps are as follows:
(1) the attribute block image of acquisition denoise and binaryzation, the denoising are using Gaussian filter convolution pair
Image denoising；
(2) attribute block edge contour is extracted using Canny edge detection algorithm, determines the marginal point of Hough transform；
(3) to each non-zero marginal point of edge contour, the Sobel first derivative on the direction x, y is calculated in its neighborhood and is obtained
Gradient:
If a certain non-zero marginal point pixel value is I (x, y), derivation is carried out using Sobel operator:
Change in the x direction, may be expressed as:
Change in y-direction, may be expressed as:
It usesCalculate gradient magnitude and direction；
(4) multiple marginal points, the search radius R of setting search center location are randomly selectedS, then in each marginal point gradient side
Upwards apart from marginal point distance RsBoth sides respectively throw a bit, draw all institutes centered on marginal point in Hough two-dimensional space
Cultellation is the gradient line segment of endpoint, and in accumulator number, choose maximum coordinate of accumulator value is the point that line segment is passed through
Central coordinate of circle, according to central coordinate of circle C (xc,yc) and marginal point coordinate I (x, y) find out radius of circle.
3. polymorphic type workpiece vision measuring method according to claim 2, which is characterized in that described image splicing is using flat
Face rectangular coordinate system coordinate transformation model realizes that steps are as follows:
(1) central coordinate of circle and radius of the circle marker block obtained according to the method for above-mentioned Hough transform, is sat using flat square
Mark system coordinate transformation model carries out Unitary coordinate, the coordinate transformation model to several scanning figures are as follows:
In formula, △ x, △ y is translation parameters, and m is scale factor, and α is rotation parameter, and (x, y), (x', y') are same point not
Point coordinate under coplanar rectangular coordinate system, Yao Shixian coordinate points must find out △ x in different coordinates conversion, △ y, m, α this
Four parameters, wherein △ x, △ y are easy to acquire, therefore at least containing there are two common points convenient for the change of above-mentioned coordinate between every two width figure
Shape parameter is changed the mold to solve；
(2) according to the coordinate transformation model of solution, next picture successively carries out coordinate and is transformed into the coordinate system of a width figure, most
Under each scanning information is unified to global coordinate system at last and carry out image co-registration elimination splicing gap, obtaining whole scan image.
4. polymorphic type workpiece vision measuring method according to claim 1, which is characterized in that in step 3, described image
Splicing uses following steps: described image acquiring unit is laser scanner, and several laser scanner progressive scans, every is swashed
For the collected every data line difference Coutinuous store of photoscanner in memory address piecemeal, memory address and data source one are a pair of
It answers, it is known that every data line first address of every laser scanner acquisition obtains data storage according to field of view width is overlapped
Offset, the continuously reading image data since the first address of specified each row of data, find repetition pixel data storage position
It sets and is refused to read, so that reading link in image data carries out image mosaic, obtain whole scan image.
5. polymorphic type workpiece vision measuring method according to claim 1, which is characterized in that the concrete operations of step 4 are such as
Under:
(1) global image is pre-processed, target area segmentation, obtains various workpiece target regions；
(2) edge detection successively is carried out to each workpiece target region, obtain clearly edge contour and intercepted, to workpiece profile area
Outer contoured features are extracted in domain, identify that the affiliated major class of workpiece completes workpiece identification by the template matching method based on feature
Then rough sort calls trained BP neural network category identification model according to affiliated major class, extract workpiece profile details
Feature is classified again by the BP neural network category identification model of affiliated major class, determines workpiece specific category, completes work
The disaggregated classification of part identification；It identifies workpiece for measurement type, counts various types of quantity, count the corresponding image procossing of each workpiece
Program number；
(3) system intercepts each workpiece target region, and the number of the processing routine according to corresponding to workpiece is integrated in library from processing routine and adjusted
With corresponding workpiece image processing routine；
(4) system obtains parameter task list to be measured by remote service end；
(5) system carries out corresponding image to workpiece targets all in global image region by the way of parallel computation simultaneously
Processing routine obtains items dimensional parameters information required by task list in conjunction with parameter task list to be measured.
6. polymorphic type workpiece vision measuring method according to claim 5, which is characterized in that
It is described as follows to the pretreated process of global image progress:
(1) using the method for gaussian filtering to whole scan image global de-noising；
(2) binary conversion treatment is carried out using the method for fixed threshold again, makes workpiece area and background separation；
(3) first the workpiece image Jing Guo binary conversion treatment is expanded, is corroded using the method for morphological operation, smoothed image
Edge；
The edge detection obtains the edge contour of workpiece using the method for Canny edge detection, is convenient for subsequent feature extraction；
The contour feature selects Hu moment characteristics, and the treatment process based on Hu moment characteristics template matching is as follows:
(1) image to be identified is pre-processed, the profile Hu moment characteristics of workpiece image is obtained according to feature extraction algorithm；
(2) similarity measure that the template image in image to be identified and template library is calculated to feature, chooses suitable decision threshold
Value, therefore, it is determined that the affiliated major class of workpiece out；
Disaggregated classification is completed using workpiece profile minutia, minutia selects small echo moment characteristics, and extraction process is as follows:
(1) workpiece image in each workpiece target region is normalized: sets the workpiece image that f (x, y) is two-dimensional discrete, first
Determine the center-of-mass coordinate of image:Secondly zoom factor α is defined, evenIt takes α=N/ (4* γ), the normalization of workpiece image can be converted to by f (x, y)It carries out；
(2) image Polar coordinates: for continuous function f (x, y), corresponding polar coordinates function representation is f (r, θ), wherein x=
Rcos θ, y=rsin θ, r are polar diameter, and θ is polar angle, since two-dimensional digital image is discrete, so being converted to by rectangular co-ordinate
When polar coordinates, equal part need to be carried out to angle, Δ θ=2 π/N is taken to carry out discretization, (p+q) rank moment characteristics of image are defined such as formula:
Fpq=∫ ∫ f (r, θ) gp(r)ejqθDrd θ, wherein gp(r) it is function about radius r, p, q are numeric parameter；
(3) FFT transform: (p+q) rank moment characteristics definition of image can be further represented as Fpq=∫ Sq(r)gp(r) rdr, and because
It is the one-dimensional sequence about variable r, and what is actually carried out is discrete transformation, therefore need to be integral discretization: chooses appropriate
Angle interval delta θ=2 π/N, then angle integrates are as follows:N indicates transformation points, and m is indicated
The constant of [0, N-1], is realized using FFT；
(4) feature is extracted in radial zone { 0≤r≤1 } using wavelet function to the integral result in above-mentioned steps, passes through gp
(p) wavelet structure square, wavelet function will axially be defined as ψj,k=2j/2ψ(2jr-k)；Select different j and k that can obtain
To the global characteristics and local feature of image；The definition of wavelet moment invariants is | | Fj,k,q| |=| | ∫ Sq(r)ψj,k(r) rdr | |,
Wherein q=0 in formula, 1,2 ..., for the r of a certain fixation, Sq(r)=∫ f (r, θ) ejqθD θ represents image f (r, θ) phase sky
Between q-th of characteristic quantity in [0,2 π]；Utilize different scale factor j, shift factor k, it will be able to provide | | Fj,k,q| | in difference
Feature on scale level, noteDjFor high frequency detail spy
Sign, SjFor low-frequency information feature, with Dj, SjThe amount of being characterized is identified, D is selected4Value and S when taking k=1,2,3,45K=1 is taken,
Value when 2, totally six values are used as characteristic quantity；
Classification for the wavelet moment characteristic value of workpiece image constructs disaggregated model using single hidden layer configuration BP neural network, including
Input layer, hidden layer, output layer, wherein 6 nodes of input layer, correspond to 6 characteristic values of workpiece image wavelet moment, corresponding workpiece
Type is the desired output of network, and the workpiece classification number that output layer number of nodes is distinguished by the network determines, the number of nodes of hidden layer by
Empirical equation determines:Wherein k is the number of hidden nodes, and n is input layer number, and m is output layer number of nodes,
A is the regulating constant between 1~10, and BP neural network model is by the small echo moment characteristics of training sample as BP neural network
Input carries out network model training, obtains workpiece disaggregated model；
The training step of BP neural network model is as follows:
(1) workpiece to be identified is chosen as sample workpiece, by each workpiece respectively every 15 degree, i.e., in positive and negative seven degree of model
Enclose interior carry out Image Acquisition；
(2) feature calculation is carried out to several images that Image Acquisition obtains, then counted under different illumination and distance condition
Mean value, establish training sample database for BP neural network model；
(3) BP neural network model is trained, each weight is adjusted using the method for gradient decline；
(4) trained BP neural network model is imported into application program, feature is extracted to the testing image comprising workpiece and is carried out
Classification and Identification simultaneously records various types of quantity.
7. polymorphic type workpiece vision measuring method according to claim 5, which is characterized in that in step 4,
Picture size processing routine obtains dimensional parameters, and specific step is as follows:
(1) edge detection is carried out to workpiece target image, extracts straight line, circle, rectangle, elliptical shape edge contour in workpiece；
(2) to each shape edges contour fitting, the distance between edge pixel is calculated, in conjunction with image pixel distance and actual range
Relationship calculates each dimensional parameters.
8. a kind of polymorphic type batch workpiece vision measurement system, which is characterized in that including data processing terminal, workpiece sensing list
Member, image acquisition unit, control unit；
The data processing terminal is for storing and processing data information；
The workpiece sensing unit is for placing workpiece for measurement and starting detection system；
Described image acquiring unit is for obtaining partial-depth image；
Described control unit includes motion platform control unit and handgrip control unit；
The motion platform control unit includes motion control card and motor driver, industry control owner built in the motion control card
At plate pci card slot, for driver issue different frequency, different number pulse, the motor driver by signal wire with
Motion control cartoon letters are for receiving pulse to control the speed of motor movement, direction, position；
The handgrip control unit is used to control the movement of grabbing workpiece；
The workpiece sensing unit, image acquisition unit, motion platform control unit and motor driver respectively with data processing
Terminal called connection；
System carries out splicing to the partial-depth image of acquisition, whole scan image is obtained, according to image acquisition unit
Detection zone on workpiece calibration platform is divided into several sub-regions by number, is separated between the subregion with markings,
The part of the scanning field of view overlapping of adjacent image acquiring unit covers markings, in image mosaic processing, puts on markings
Enter multiple attribute blocks, extract the characteristic information of attribute block, image mosaic is carried out using plane right-angle coordinate coordinate transformation model.
9. polymorphic type batch workpiece vision measurement system according to claim 8, which is characterized in that
The workpiece sensing unit includes workpiece calibration platform (3) and several electric switches that diffuse (2), the workpiece calibration
Platform (3) surface sets detection zone, and described several diffuse, and electric switch (2) is equidistant, in a row is arranged in the workpiece
Measuring table (3) edge；
Described image acquiring unit is laser scanner (1)；
The motion platform control unit is Three-degree-of-freedom motion platform, and the handgrip control unit includes mechanical arm handgrip (4)
With manipulator motion track (5), the mechanical arm handgrip (4) is fixed on manipulator motion track (5)；
The Three-degree-of-freedom motion platform includes that X-axis motion platform track (6), Y-axis motion platform track (7) and Z axis movement are flat
Platform track (8)；The X-axis motion platform track (6) and manipulator motion track (5) are arranged in parallel in institute in the horizontal direction respectively
The two sides of workpiece calibration platform (3) are stated, the Z axis motion platform track (8) is arranged in a vertical direction in the X-axis motion platform
On track (6), the Y-axis motion platform track (7) is arranged in institute along the direction perpendicular to the X-axis motion platform track (6)
It states on Z axis motion platform track (8)；The spaced apart linear motor movement of several segments is set on the Y-axis motion platform track (7)
Guide rail, several described laser scanners (1) are arranged on each section of linear motor motion guide rail, described to diffuse
The direction in a row of electric switch (2) is vertical with the length direction of the X-axis motion platform track (6).
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811273893.1A CN109357630B (en) | 2018-10-30 | 2018-10-30 | A kind of polymorphic type batch workpiece vision measurement system and method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811273893.1A CN109357630B (en) | 2018-10-30 | 2018-10-30 | A kind of polymorphic type batch workpiece vision measurement system and method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109357630A CN109357630A (en) | 2019-02-19 |
| CN109357630B true CN109357630B (en) | 2019-10-18 |
Family
ID=65347394
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811273893.1A Active CN109357630B (en) | 2018-10-30 | 2018-10-30 | A kind of polymorphic type batch workpiece vision measurement system and method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109357630B (en) |
Families Citing this family (34)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109977471B (en) * | 2019-02-25 | 2023-04-07 | 中车工业研究院有限公司 | Adjusting and repairing calibration system and adjusting and repairing calibration method for rail transit parts |
| CN109781739A (en) * | 2019-03-04 | 2019-05-21 | 杭州晶耐科光电技术有限公司 | Automobile finish surface appearance defects automatic detection system and method |
| CN110097540A (en) * | 2019-04-20 | 2019-08-06 | 东莞中科蓝海智能视觉科技有限公司 | The visible detection method and device of polygon workpeace |
| CN110132166B (en) * | 2019-05-05 | 2021-07-23 | 广州佳帆计算机有限公司 | Product image detection method capable of automatically distributing light and comparison device |
| CN111896539A (en) * | 2019-05-06 | 2020-11-06 | 迪密机电(杭州)有限公司 | High-precision workpiece quality detection equipment and detection method thereof |
| CN110102906A (en) * | 2019-05-10 | 2019-08-09 | 定颖电子(黄石)有限公司 | A kind of radium carves stamp system and radium carves stamp technique |
| CN110216082B (en) * | 2019-05-23 | 2020-07-14 | 上海交通大学 | Fluorescence labeling seed dynamic identification method and system |
| CN110132154B (en) * | 2019-06-05 | 2021-02-05 | 江南大学 | Reed quality detection method based on machine vision |
| CN112183707A (en) * | 2019-07-02 | 2021-01-05 | 深圳市玄羽科技有限公司 | Three-dimensional intelligent counting method and system |
| WO2021012268A1 (en) * | 2019-07-25 | 2021-01-28 | 浙江汉朔电子科技有限公司 | Shelf edge positioning method and device |
| CN111753609B (en) * | 2019-08-02 | 2023-12-26 | 杭州海康威视数字技术股份有限公司 | Target identification method and device and camera |
| CN110598637B (en) * | 2019-09-12 | 2023-02-24 | 齐鲁工业大学 | Unmanned system and method based on vision and deep learning |
| CN110712202B (en) * | 2019-09-24 | 2021-07-16 | 鲁班嫡系机器人（深圳）有限公司 | Special-shaped component grabbing method, device and system, control device and storage medium |
| CN111047588A (en) * | 2019-12-26 | 2020-04-21 | 电子科技大学 | Imaging measurement method for size of shaft type small part |
| CN111104542B (en) * | 2020-01-07 | 2023-04-18 | 成都睿琪科技有限责任公司 | Part identification management method and device |
| CN111238370A (en) * | 2020-02-20 | 2020-06-05 | 中国科学院声学研究所东海研究站 | Intelligent detection method and device for KIT board |
| TWI777173B (en) * | 2020-06-01 | 2022-09-11 | 汎銓科技股份有限公司 | Artificial intelligence identified measuring method for a semiconductor image |
| CN112070708B (en) * | 2020-08-21 | 2024-03-08 | 杭州睿琪软件有限公司 | Image processing method, image processing apparatus, electronic device, and storage medium |
| CN112084964A (en) * | 2020-09-11 | 2020-12-15 | 浙江水晶光电科技股份有限公司 | Product identification apparatus, method and storage medium |
| CN112232376A (en) * | 2020-09-21 | 2021-01-15 | 埃夫特智能装备股份有限公司 | Hu-moment-based sheet metal part identification method |
| CN113631339B (en) * | 2020-10-19 | 2023-03-24 | 日本省力机械株式会社 | Processing device |
| CN112560748A (en) * | 2020-12-23 | 2021-03-26 | 安徽高哲信息技术有限公司 | Crop shape analysis subsystem and method |
| CN113212889B (en) * | 2021-02-26 | 2023-11-03 | 深圳市云鼠科技开发有限公司 | Package visual leakage detection device, method, computer equipment and storage medium |
| CN113074633B (en) * | 2021-03-22 | 2023-01-31 | 西安工业大学 | Automatic detection system and detection method for overall dimension of material |
| CN113639630A (en) * | 2021-04-01 | 2021-11-12 | 浙江大学台州研究院 | Dimension measuring instrument system based on multi-template matching and automatic focusing functions |
| CN113269743B (en) * | 2021-05-20 | 2022-09-30 | 北京理工大学重庆创新中心 | Chip quantity detection method based on iterative translation verification |
| CN113499094B (en) * | 2021-07-08 | 2023-07-25 | 中山大学 | Heart color Doppler ultrasound examination device and method guided by vision and force feedback |
| CN113406093B (en) * | 2021-08-19 | 2021-11-30 | 苏州维嘉科技股份有限公司 | Optical detection equipment and method and device for measuring object attribute thereof |
| CN113701821B (en) * | 2021-09-01 | 2024-03-05 | 西南科技大学 | Building aluminum template identification structure based on local features and method thereof |
| CN113781481A (en) * | 2021-11-11 | 2021-12-10 | 滨州学院 | Method and device for non-contact measurement of shape and size of object and electronic equipment |
| CN114111571A (en) * | 2021-11-15 | 2022-03-01 | 天津大学 | Visual precision detection device for measuring special-shaped workpiece |
| CN114399507B (en) * | 2022-03-25 | 2022-06-17 | 季华实验室 | Mobile phone appearance quality detection method and device, electronic equipment and storage medium |
| CN115345006A (en) * | 2022-08-16 | 2022-11-15 | 杭州宏深科技有限公司 | Three-dimensional digital flexible online detection method based on characteristic template |
| CN116230610B (en) * | 2023-05-08 | 2023-09-26 | 上海隐冠半导体技术有限公司 | Workpiece position adjustment method and system |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103063159B (en) * | 2012-12-31 | 2015-06-17 | 南京信息工程大学 | Part size measurement method based on charge coupled device (CCD) |
| CN108489394A (en) * | 2018-04-17 | 2018-09-04 | 沈阳建筑大学 | A kind of large-scale sheet metal works almost T-stable automatic detection device and method |
-
2018
- 2018-10-30 CN CN201811273893.1A patent/CN109357630B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN109357630A (en) | 2019-02-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109357630B (en) | A kind of polymorphic type batch workpiece vision measurement system and method | |
| US11937020B2 (en) | Object inspection system and method for inspecting an object | |
| CN107229930B (en) | Intelligent identification method for numerical value of pointer instrument | |
| ES2910068T3 (en) | Mobile and automated device for the detection and classification of damage to the bodywork of a vehicle | |
| CN108875911A (en) | One kind is parked position detecting method | |
| CN107389701A (en) | A kind of PCB visual defects automatic checkout system and method based on image | |
| CN104112269B (en) | A kind of solar battery laser groove parameter detection method and system based on machine vision | |
| CN110344621A (en) | A kind of wheel points cloud detection method of optic towards intelligent garage | |
| CN111612737B (en) | Artificial board surface flaw detection device and detection method | |
| CN106248686A (en) | Glass surface defects based on machine vision detection device and method | |
| CN109255787A (en) | Silk ingot scratch detection system and method based on deep learning and image processing techniques | |
| CN104360501A (en) | Visual detection method and device for defects of liquid crystal display screen | |
| MXPA02001474A (en) | Item recognition method and apparatus. | |
| JPH07325161A (en) | Method and equipment for package inspection | |
| CN108416765B (en) | Method and system for automatically detecting character defects | |
| CN108007388A (en) | A kind of turntable angle high precision online measuring method based on machine vision | |
| CN102374993A (en) | Unit and method for detecting appearance of three dimensional pin of integrated circuit package chip | |
| CN104483320A (en) | Digitized defect detection device and detection method of industrial denitration catalyst | |
| CN109840900A (en) | A kind of line detection system for failure and detection method applied to intelligence manufacture workshop | |
| CN109978940A (en) | A kind of SAB air bag size vision measuring method | |
| CN110084842A (en) | A kind of secondary alignment methods of machine user tripod head servo and device | |
| Yang et al. | An automatic aperture detection system for LED cup based on machine vision | |
| CN113504238B (en) | Glass surface defect acquisition device and detection method | |
| CN105956619A (en) | Container lockhole coarse positioning and tracking method | |
| CN112284258B (en) | System for measuring cable structure size parameters based on machine vision algorithm |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |