CN102507592B - Fly-simulation visual online detection device and method for surface defects - Google Patents
Fly-simulation visual online detection device and method for surface defects Download PDFInfo
- Publication number
- CN102507592B CN102507592B CN201110338628.9A CN201110338628A CN102507592B CN 102507592 B CN102507592 B CN 102507592B CN 201110338628 A CN201110338628 A CN 201110338628A CN 102507592 B CN102507592 B CN 102507592B
- Authority
- CN
- China
- Prior art keywords
- image
- video
- fusion
- ccd
- 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.)
- Active
Links
Images
Landscapes
- Image Processing (AREA)
Abstract
The invention discloses a fly-simulation visual online detection device and method for surface defects, belonging to the technical field of industrial visual detection and image processing. According to the invention, the surface defects under a complicated background are detected by combining a common CCD (Charge-coupled Device) image sensor with a surface defect detection algorithm. According to the principle of the fly-simulation visual online detection device, the low contrast between the defects and the background needed by the traditional defect detection algorithm is broken through, the obtained scene images can be qualitatively approximate to compound-eye low-resolution images by simulating a fly compound-eye visual information processing method in lower-resolution and low-contrast scene images, and the accuracy is higher. Meanwhile, aiming at the surface defects static in a scene, the fly-simulation visual online detection device also can discover the existence of the surface defects, therefore reliable information is provided for subsequent treatment.
Description
Technical field
The invention belongs to industrial vision and detect and technical field of image processing, particularly relate to a kind of imitative online surface defect detection apparatus of fly vision and detection method.
Background technology
In copper strip manufacturing process, owing to affected by severe site environment complicated and changeable, add the vibrations of reflection, scattering properties and the high-speed cruising generation of copper strips strip surface itself, the online detection that also has sheet metal strip, towards future developments such as large scale, high precision, high-speed, high reliability, makes vision detection technology in applying, face problems.Image quality robustness is poor, signal to noise ratio (S/N ratio) is low, and defect shape feature and textural characteristics are weak, redundant information is deficient, has increased image detail analysis and identification difficulty; Particularly tiny flaw is very easily flooded by clutter and noise, and defect detection rate, to environmental change sensitivity, often occurs undetected and alarmed falsely.The accurate detection that tiny flaw is detected becomes the bottleneck that vision detection technology is given full play to premium properties and further promoted.The relative size of or defect far away apart from observation station when defect hour, often needs to adopt the video camera of high spatial resolution to obtain scene information identification surface imperfection.
At nature, the Target detection and identification ability of insect vision system brilliance, make insect to carry out detection and tracking to high-speed moving object interested in any environment, Compound Eye of Insects vision is in level and the parallel processing process of Information acquisi-tion process, final formation detects and stares the fast accurate of the target in complex scene, and insect vision is good state and suitable fixed to the detection and Identification process of complex scene target in essence.The compound eye of for example fly, it is made up of thousands of ommatidiums, each ommatidium can only be seen the very little of outer scene, because the aperture of every ommatidium is very little, between ommatidium and ommatidium, also exist some to disturb simultaneously, thus the image information that compound eye obtains can be subject to diffraction and interference to affect resolution all lower.However, fly still can rely on such vision system search of food and spouse.
Under the inspiration of insect vision system Information acquisi-tion, people have launched a large amount of research work, and attempt makes up the deficiency of existing equipment by some imaging mechanisms of compound eye.And the method for extracting for surface imperfection at present also has a lot, great majority be employing Digital Image Processing method for example: matched filter, wavelet analysis etc.These artificial vision's methods are all to rely on from background, to isolate the thought of defect, require to exist between defect and background certain motion.And the treatment mechanism of Compound Eye of Insects is told us, complicated processing, too simple concerning fly like this.The problems such as the complicated spatial interaction impact that need to consider when compound eye disposal route can also be avoided artificial treatment simultaneously.
Summary of the invention
In order to address the above problem, on the basis that the present invention analyzes compound eye physiologic character biologist, imitate fly acquisition of vision information and process mechanism, the imitative fly vision on-line measuring device of a kind of surface imperfection and detection method are provided.
Principle of work of the present invention: the present invention is by running on the detection process of the next virtual Compound Eye of Insects effects on surface defect of software processing in embedded image processing unit.The picture frame obtaining in the same time from multiple ccd image sensors is spliced, obtain the scene image larger than single ccd image sensor.In simulation ocular nerve Processing Algorithm on dsp chip, adopt and cut apart virtual several facets of image block, each ommatidium has an a small amount of little array of photoreceptor composition.Photoreceptor array can detect in its observed zonule whether have surface imperfection.The wherein corresponding pixel of each photoreceptor.The less image block that each ommatidium is chosen in image is processed.Because the input aperture of ommatidium is very little, be generally 2 degree, so the resolution of ommatidium acquisition image is lower.And in detecting surface defect, the quality of resolution is little on Effect on Detecting impact.
Stronger than other flat sites to the response of Defect Edge in reference biological vision, and for texture information, the mechanism that eye response meeting weakens gradually along with the continuous repetition of texture.And the fast powering-up of optic nerve response and discharge mechanism at a slow speed, consider and in scene, have surface imperfection, between the brightness value of its edge with respect to scene around, there is larger difference so.For a defect lower than scene brightness, only consider from horizontal direction in general, brightness change from left to right should be: bright-dark-bright.Therefore, judge whether to exist surface imperfection by the fusion of two width figure.The two width figure that are divided into, a width Luminance Distribution, on segmentation threshold, is called ON figure, and the Luminance Distribution of an other width is called off figure under segmentation threshold.Make along the defect body region at surface imperfection edge cutting apart like this or a closed region more bigger than defect area appears at (dark defect) in off figure.Just there are two skip signal in the place that defect occurs like this: brightness increases and reduces.By the conversion of symbol, can will be that negative off figure transfers the square calculating of face after an action of the bowels to originally.These two skip signal suppress to process by central side, can be translated into pulse signal, have increased difference between other adjacent signals simultaneously.Finally the off figure of translation is carried out to multiplication fusion treatment with on figure, obtain the output signal of surface imperfection.If exist two pulses to represent so the existence of surface imperfection simultaneously; If only have an existence, after multiplying each other so, be just zero.In this process, not only can obtain the information whether surface imperfection exists, the size of surface imperfection also can reflect in the time span of off pulse daley.
The present invention mainly for be the surface imperfection in complicated industrial production environment, the size that surface imperfection refers generally to defect is here within several pixel coverages.Simultaneously to those textures than more rich region, the mechanism that the fast powering-up of on passage or off passage discharges at a slow speed can make to fall the monochrome information of swinging fluctuating and smooth to and approach constant.
In order to address the above problem, the technical solution used in the present invention is:
The imitative fly vision on-line measuring device of a kind of surface imperfection, comprise ccd image sensor part, it is characterized in that: also comprise embedded image processing unit, described ccd image sensor part comprises some ccd image sensors that gather for video information, described some ccd image sensors are fan-shaped array, and the axis concurrent of each ccd image sensor, the signal wire of described some ccd image sensors is connected to embedded image processing unit by the concentric cable of video.
Aforesaid a kind of surface imperfection is imitated fly vision on-line measuring device, it is characterized in that: described ccd image sensor part comprises five ccd image sensors, and described five ccd image sensors are distributed in a semicircular installing plate 32.
Aforesaid a kind of surface imperfection is imitated fly vision on-line measuring device, it is characterized in that: described embedded image processing unit consists of the following components:
Audio video synchronization device, for coordinating the input of five ccd image sensor cooperative works and control vision signal;
Video encoder, receiver, video synchronizer obtains analog video, for analog video digitizing and video formats are changed;
Picture frame storer, the digital video after the coding of receiver, video scrambler output, for the quantification of digital video and the storage of the rear information of coding;
General purpose I/O, for the input and output of switching value signal;
CPLD, is connected with FLASH with video encoder, picture frame storer, general purpose I/O respectively, for the realization of bus expansion and system sequence circuit and logical circuit;
CPU (central processing unit), for carrying out Intelligent treatment to digitized video;
Communication interfaces of wireless local network, for the interconnected and exchange message of communicating by letter between each node;
FLASH, for storage system software;
D/A, is connected with CPU (central processing unit), for converting digital video to analog video;
SDRAM, for system when the deal with data, store data;
RS485, for communicating by letter between CPU (central processing unit) and PTZ controller, exchange control command;
PTZ controller module, is respectively used to the control to five ccd image sensors.
Aforesaid a kind of surface imperfection is imitated fly vision on-line measuring device, it is characterized in that: described CPU (central processing unit) adopts TMS320DM642 processor, described SDRAM is 64 bit synchronization dynamic memory interface (DMI)s, storage space 4M*64 position, described FLASH is 8 asynchronous sram interfaces, storage space 4M*8 position.
A detection method for the imitative fly vision of surface imperfection, is characterized in that: comprise following two large flow processs: image registration splicing flow process and bionic compound eyes surface defects detection flow process,
Described image registration splicing flow process comprises the following steps:
(1), obtain multiway images, the image that multiple CCD imageing sensors obtain is after perspective transform projects to detection faces, the coordinate in detection plane is just determined;
(2), Image Mosaics pre-service, to the image Primary Location of obtaining, dwindle matching range, improve matching speed, to image denoising sound, state diagram image brightness, to the splicing part proportion territory method of image to image enhancement processing, interested feature in image is selectively outstanding, and its less important information that decays;
(3), image registration, find out the displacement situation between two width or several superimposed images of alignment, the model of describing transforming relationship between two width images carries out image registration;
(4), image co-registration, the image that different CC imageing sensors photograph, light intensity, color saturation all can be variant, therefore have obvious gap at the boundary of Image Mosaics, use weighted average method many CCD image is carried out to fusion treatment;
Described bionic compound eyes surface defects detection flow process comprises the following steps:
(1) scene image that, visual properties compression module obtains splicing does non-linear compression;
(2) rectification processing and the central side, integrated based on threshold value based on local contrast 3 D defects detection module suppress to process;
(3), defect merge output by horizontal direction and two channel signals in vertical direction respectively by together with the Fusion of Cells of pond, then the result of fusion is merged to the output that just can obtain last defect again.
The detection method of the imitative fly vision of aforesaid a kind of surface imperfection, it is characterized in that: described image registration splicing process step (3): between described two width images, the model of transforming relationship adopts Corresponding matching (Homographic Mapping) model, be that original image is that perspective transform obtains, the motion of CCD imageing sensor is mainly to shake mirror, translation, inclination, Rotation and Zoom around its optical centre
Wherein A represents convergent-divergent and rotation, and B represents translation, and CT represents projection.
The detection method of the imitative fly vision of aforesaid a kind of surface imperfection, it is characterized in that: in described bionic compound eyes surface defects detection process step (1), the scene image that visual properties compression module obtains splicing does non-linear compression, adopts linear compression transformation for mula to be:
In this formula
be the brightness value after compression, in this formula denominator part, Section 2 is the method for the moving average of employing, obtains like this midrange coming and has adaptivity.
The detection method of the imitative fly vision of aforesaid a kind of surface imperfection, is characterized in that: in described bionic compound eyes surface defects detection process step (2), by calculating local contrast
select to exist the image block of surface imperfection to process, local contrast
can weigh and in this scene, whether have the surface imperfection that can be identified, the position larger than degree occurs that the possibility of surface imperfection is larger, selecting this position is the center of detected image block, supposes that the size of surface imperfection is 4 pixel left and right here, and what therefore local contrast adopted is
the local contrast of image block, rather than single pixel, local contrast computing formula is as follows:
The detection method of the imitative fly vision of aforesaid a kind of surface imperfection, it is characterized in that: in described bionic compound eyes surface defects detection process step (3), in analysis level directional image, from left to right then the reduction of brightness is before this increase of brightness, the be separated by distance of several pixels of centre, off figure overlaps in the edge of defect with on passage through translation, multiply each other like this and will obtain the result of non-zero, if and there is no surface imperfection in this image block, the namely edge of neither one closure, in off image or on image, lack an edge so, after multiplying each other, result is zero, in vertical direction, be in like manner, upwards there is an edge in the non-vanishing explanation the party of fusion output in horizontal direction, upwards also there is an edge in the non-vanishing explanation the party of fusion output in vertical direction, adopt the mode of logical and to merge two orthogonal directionss, exist surface imperfection to export so this signal if export non-vanishing explanation.
The detection method of the imitative fly vision of aforesaid a kind of surface imperfection, is characterized in that: described (2) Image Mosaics processing comprises the following steps:
The first step: be that two width image SIFT feature extracting methods are extracted to tie point automatically, then utilize affined transformation to detect the change situation of image, after once the tie point that will use having been detected, use Euclid's nearest neighbor algorithm to carry out characteristic matching, a large amount of incorrect link points can be rejected by RMSE coefficient, and the precision after conversion is described by root-mean-square error:
Wherein N refers to the number of all match points;
Second step: when the RMSE of matching double points is greater than certain particular value, abandon this to unique point, if overall RMSE is excessive, carry out first-order error inspection, whole process iteration is carried out for example, until the RMSE of certain a pair of unique point is less than the number that certain value (0.5 pixel) or tie point are less than setting again;
The 3rd step: after slightly having mated, for local deformation is corrected, need further thin coupling, use Harris angular-point detection method intensive feature point set can be detected in image A, first input picture is carried out to Harris Corner Detection, because the process of Harris Corner Detection is to use a series of wave filter to process image, so its operation complexity is based on image size, in order to accelerate processing procedure, we can be divided into fritter by image, then carry out piecemeal processing, in the large person of Harris value of the feature that final feature selecting detects from each, choose, choose front 10% herein as unique point,
The 4th step: in image B, choose with image A in Harris angle point characteristic of correspondence point, the choosing of character pair point directly has influence on the speed of processing and the precision of final registration, first respectively image A and image B set up to wavelet pyramid; Then use cross-matched technology respectively the pyramidal match point of the every one deck corresponding to particular dimensions to be carried out to hierarchical search; Finally use again RMSE coefficient to remove Mismatching point.
The detection method of the imitative fly vision of aforesaid a kind of surface imperfection, is characterized in that: described (4) image co-registration comprises the following steps:
The first step: source images to be merged is weighted to the fusion of these two kinds of classic methods of multiple dimensioned decomposition of average and tower conversion, after merging, compare objective evaluation coefficient, be root-mean-square error, average error and the setting threshold size between fused image and ideal image, if be less than setting threshold, mixing operation finishes;
Second step: if evaluation coefficient is greater than setting threshold, fused image is merged respectively with source images A and B respectively again, two width images after again merging and the evaluation coefficient between ideal image are compared, select the desirable image of fusion situation, be evaluation coefficient smaller, set corresponding with it source images, A or B are as the image of iteration fusion below;
The 3rd step: the evaluation coefficient of comparatively ideal fused images and setting threshold are compared, if be less than threshold value, finish mixing operation, this width image is as final fusion results; If be greater than threshold value, the image after again this width being merged and before selected source images carry out mixing operation, until the objective evaluation coefficient of the image after fusion is less than the threshold value of setting.
The invention has the beneficial effects as follows: the present invention adopts the surface defect detection apparatus of common ccd image sensor and dsp chip composition, rely on imitative ocular nerve mechanism to survey defect, the low-resolution image that is similar to qualitatively compound eye at the scene image obtaining, accuracy rate is higher.To have cost low for this equipment simultaneously, the feature such as computational complexity is low, and for the surface imperfection being still in scene, this equipment still can be found the existence of surface imperfection, for subsequent treatment provides reliable information.
Accompanying drawing explanation
Fig. 1 is system composition schematic diagram of the present invention.
Fig. 2 is embedded image processing unit schematic diagram of the present invention.
Fig. 3 is image registration splicing process flow diagram of the present invention.
Fig. 4 is the imitative compound eye surface defects detection process flow diagram of the present invention.
Embodiment
Below in conjunction with accompanying drawing, the present invention is described in further detail.
As shown in Figure 1, the imitative online surface defect detection apparatus of fly vision of described fan comprises ccd image sensor part and embedded image processing unit; Wherein, sensor construction part adopts axis concurrent to be many ccd image sensors 31 combination units of fan-shaped array, many ccd image sensors 31 are distributed on a semicircular installing plate 32, and the signal wire of many ccd image sensors 31 is connected to embedded image processing unit 34 by the concentric cable 33 of video.
Be as shown in Figure 2 embedded image processing unit of the present invention with DSP core, it is characterized in that audio video synchronization device [2], for coordinating three video camera cooperative works and controlling the input of vision signal, video encoder [3], for digitizing and the video formats conversion of analog video, picture frame storer [4], for quantizing and the storage of information afterwards of encoding, general purpose I/O[5], for the input and output of switching value signal, CPLD[6], for the realization of bus expansion and system sequence circuit and logical circuit, CPU (central processing unit) [9], for carrying out Intelligent treatment etc. to digitized video, communication interfaces of wireless local network [7], exchanges for information communication, FLASH[8], for storage system software, SDRAM[11], for system when the deal with data, store data, RS485 [12], for communicating by letter between CPU (central processing unit) and PTZ controller, exchange control command, PTZ controller module [13], for the control to video camera, D/A[10], for by digitized video converting analogue video, described video encoder [3] is selected and is adopted PHILIPS coding and decoding video core SAA7114, D/A[10] select PHILIPS video decoding chip SAA7121, CPU (central processing unit) [9] is selected the TMS320DM642 processor of TI company, SDRAM[11] be 64 bit synchronization dynamic memory interface (DMI)s, storage space 4M*64 position, be mapped to the CE0 port of CPU (central processing unit) [9], memory range is orientated 0x80000000 ~ 0x81FFFFFFF as, its work clock is provided by the CLKOUT1 port of CPU (central processing unit) [9], software is configured to cpu clock/4, FLASH[8] be 8 asynchronous sram interfaces, storage space 4M*8 position, be mapped to the CE1 port of CPU (central processing unit) [9], memory range navigates to 0x90000000 ~ 0x9007FFFF.Contrast Fig. 1, between audio video synchronization device of the present invention [2], connect by three concentric cable, the interface of audio video synchronization device [2] one sides is BNC standard interface, support hot plug, audio video synchronization device [2] is connected by I2C bus with CPU (central processing unit) [9] simultaneously, audio video synchronization device [2] receives the decision-making order of CPU (central processing unit) [9], select output 1-3 road analog video to video encoder [3], to the parameter configuration order of video encoder [3] by CPLD[6] realize by I2C bus, digital video after coding inputs to picture frame storer [4] by video encoder [3], CPU (central processing unit) [9] is from picture frame storer [4] reading images frame, described FLASH[8], video encoder [3], picture frame storer [4] and general purpose I/O[5] respectively with CPLD[6] be connected, work required clock signal and logical signal by CPLD[6] produce, D/A[10] be connected individual event intercommunication with CPU (central processing unit) [9], complete the conversion of digital video to analog video, PTZ controller module [13] passes through RS485[12] and CPU (central processing unit) [9] intercommunication, the operation control command of five ccd image sensors of front end is produced and passes through RS485[12 by CPU (central processing unit) [9]] input to PTZ controller module [13].
Image registration splicing flow process of the present invention as shown in Figure 3, the image that multiple CCD imageing sensors obtain is after perspective transform projects to detection faces, coordinate in detection plane is just determined, directly opsition dependent is stitched together, become a complete large view field image, specific implementation step is as follows:
Image Mosaics pre-service
To the image Primary Location of obtaining, dwindle matching range, improve matching speed, to image denoising sound, state diagram image brightness, comprise figure image intensifying and alignment in advance etc., to the splicing part proportion territory method of image to image enhancement processing, interested feature in image is selectively outstanding, and its less important information that decays.Because image exists noise, the lap of adjacent image sample is incomplete same in detail, is filtering image noise, first selects Gaussian filter to carry out smoothing processing and Grads Sharp processing to gray level image.
(2) image registration
Image registration is the core procedure of the process of Image Mosaics, and its target is the displacement situation of finding out between two width or several superimposed images of alignment.The key of accuracy registration will have the model that can well describe transforming relationship between two width images exactly, splicing adopts Corresponding matching (Homographic Mapping) model, be that original image is that perspective transform obtains, the motion of CCD video camera is mainly to shake mirror (pan), translation, inclination, Rotation and Zoom around its optical centre.
（1）
Wherein A represents convergent-divergent and rotation, and B represents translation, and CT represents projection.
(3) fusion of image
The image that different CC imageing sensors photograph, light intensity, color saturation all can be variant, therefore at figure
The boundary of picture splicing has obvious gap, and the present invention uses the fusion treatment of weighted average method to many CCD image.
Detailed bionic compound eyes surface defects detection process flow diagram as shown in Figure 4.Mainly comprise non-linear compression module, merged output module three parts based on local contrast 3 D defects detection module, defect.
The first step, visual properties compression module does non-linear compression to the scene image that obtains of splicing, and the brightness variation range that such processing can compressed image increases the time range of luminance transformation.Here the non-linear compressed transform formula of using is:
In this formula
it is the brightness value after compression.In this formula denominator part, Section 2 is the method for the moving average of employing, obtains like this midrange coming and has adaptivity.
Second step, rectification processing and the central side integrated based on threshold value based on local contrast 3 D defects detection module suppress to process.The key whether defect can be detected is whether the contrast in the local scene that occurs of defect is outstanding.By calculating local contrast
select to exist the image block of surface imperfection to process.Local contrast
can weigh and in this scene, whether have the surface imperfection that can be identified.The position larger than degree occurs that the possibility of surface imperfection is larger, and selecting this position is the center of detected image block.Here suppose that the size of surface imperfection is 4 pixel left and right.What therefore local contrast adopted is
the local contrast of image block, rather than single pixel.Local contrast computing formula is as follows:
The 3rd step, then defect fusion output merge the result of fusion with the output that just can obtain last defect together with horizontal direction is passed through respectively to pond Fusion of Cells with two channel signals in vertical direction again.Here hypothesis is first considered dark defect, and in analysis level directional image, from left to right then the reduction of brightness is before this increase of brightness, the be separated by distance of several pixels of centre.Off figure overlaps in the edge of defect with on passage through translation, multiplies each other like this and will obtain the result of non-zero.And if in this image block, there is no surface imperfection, the namely edge of neither one closure, lacks an edge so in off image or on image, multiply each other after result be zero.In vertical direction, be in like manner.
Upwards there is an edge in the non-vanishing explanation the party of fusion in horizontal direction output, in vertical direction in like manner.Therefore adopt the mode of logical and to merge two orthogonal directionss, exist surface imperfection to export so this signal if export non-vanishing explanation.
More than show and described ultimate principle of the present invention, principal character and advantage.The technician of the industry should understand; the present invention is not restricted to the described embodiments; that in above-described embodiment and instructions, describes just illustrates principle of the present invention; without departing from the spirit and scope of the present invention; the present invention also has various changes and modifications, and these changes and improvements all fall in the claimed scope of the invention.The claimed scope of the present invention is by appending claims and equivalent circle thereof.
Claims (3)
1. the imitative fly vision on-line measuring device of surface imperfection, comprise ccd image sensor part, also comprise embedded image processing unit, described ccd image sensor part comprises some ccd image sensors that gather for video information, it is characterized in that: described some ccd image sensors are fan-shaped array, and the axis concurrent of each ccd image sensor, the signal wire of described some ccd image sensors is connected to embedded image processing unit by the concentric cable of video, described ccd image sensor part comprises five ccd image sensors, described five ccd image sensors are distributed in a semicircular installing plate, described embedded image processing unit consists of the following components:
Audio video synchronization device, for coordinating the input of five ccd image sensor cooperative works and control vision signal;
Video encoder, receiver, video synchronizer obtains analog video, for analog video digitizing and video formats are changed;
Picture frame storer, the digital video after the coding of receiver, video scrambler output, for the quantification of digital video and the storage of the rear information of coding;
General purpose I/O, for the input and output of switching value signal;
CPLD, is connected with FLASH with video encoder, picture frame storer, general purpose I/O respectively, for the realization of bus expansion and system sequence circuit and logical circuit;
CPU (central processing unit), for carrying out Intelligent treatment to digitized video;
Communication interfaces of wireless local network, for the interconnected and exchange message of communicating by letter between each node;
FLASH, for storage system software;
D/A, is connected with CPU (central processing unit), for converting digital video to analog video;
SDRAM, for system when the deal with data, store data;
RS485, for communicating by letter between CPU (central processing unit) and PTZ controller, exchange control command;
PTZ controller module, is respectively used to the control to five ccd image sensors.
2. the imitative fly vision on-line measuring device of a kind of surface imperfection according to claim 1, it is characterized in that: described CPU (central processing unit) adopts TMS320DM642 processor, described SDRAM is 64 bit synchronization dynamic memory interface (DMI)s, storage space 4M*64 position, described FLASH is 8 asynchronous sram interfaces, storage space 4M*8 position.
3. a detection method for the imitative fly vision of surface imperfection, is characterized in that: comprise following two large flow processs: image registration splicing flow process and bionic compound eyes surface defects detection flow process,
Described image registration splicing flow process comprises the following steps:
(1), obtain multiway images, the image that multiple ccd image sensors obtain is after perspective transform projects to detection faces, the coordinate in detection plane is just determined;
(2), Image Mosaics pre-service, to the image Primary Location of obtaining, dwindle matching range, improve matching speed, to image denoising sound, state diagram image brightness, to the splicing part proportion territory method of image to image enhancement processing, interested feature in image is selectively outstanding, and its less important information that decays;
Described (2) Image Mosaics processing comprises the following steps:
The first step: be that two width image SIFT feature extracting methods are extracted to tie point automatically, then utilize affined transformation to detect the change situation of image, after once the tie point that will use having been detected, use Euclid's nearest neighbor algorithm to carry out characteristic matching, a large amount of incorrect link points can be rejected by RMSE coefficient, and the precision after conversion is described by root-mean-square error:
Wherein N refers to the number of all match points, (x
a,n, y
a,n) SIFT unique point in presentation video A; (x
b,n, y
b,n) the SIFT unique point that matches in presentation video B and image A; Δ x, Δ y represents respectively SIFT point mistake matching value; (u, the v) radiation of presentation video conversion;
Second step: when the RMSE of matching double points is greater than certain particular value, abandon this to unique point, if overall RMSE is excessive, carry out again first-order error inspection, whole process iteration is carried out until the RMSE of certain a pair of unique point is less than the number that certain value or tie point are less than setting;
The 3rd step: after slightly having mated, for local deformation is corrected, need further thin coupling, use Harris angular-point detection method intensive feature point set can be detected in image A, first input picture is carried out to Harris Corner Detection, because the process of Harris Corner Detection is to use a series of wave filter to process image, so its operation complexity is based on image size, in order to accelerate processing procedure, image is divided into fritter, then carry out piecemeal processing, in the large person of Harris value of the feature that final feature selecting detects from each, choose, choose front 10% as unique point,
The 4th step: in image B, choose with image A in Harris angle point characteristic of correspondence point, the choosing of character pair point directly has influence on the speed of processing and the precision of final registration, first respectively image A and image B set up to wavelet pyramid; Then use cross-matched technology respectively the pyramidal match point of the every one deck corresponding to particular dimensions to be carried out to hierarchical search; Finally use again RMSE coefficient to remove Mismatching point;
(3), image registration, find out the displacement situation between two width or several superimposed images of alignment, utilize the model of describing transforming relationship between two width images to carry out image registration;
Described image registration splicing process step (3): between described two width images, the model of transforming relationship adopts Corresponding matching (Homographic Mapping) model, be that original image is that perspective transform obtains, the motion of ccd image sensor is shaken mirror, translation, inclination, Rotation and Zoom around its optical centre
Wherein A represents convergent-divergent and rotation, and B represents translation, C
trepresent projection; X, y is the position of pixel, p (x, y) is x, the locational pixel value of y,
Presentation video is at the translational movement of x, y direction,
Presentation video is at the rotation amount of x, y direction;
The Rotation and Zoom of presentation video;
(4), image co-registration, the image that different ccd image sensors photograph, light intensity, color saturation all can be variant, therefore have obvious gap at the boundary of Image Mosaics, use weighted average method many ccd images are carried out to fusion treatment;
Described (4) image co-registration comprises the following steps:
The first step: source images to be merged is weighted to the fusion of these two kinds of classic methods of multiple dimensioned decomposition of average and tower conversion, comparative evaluation coefficient after merging, be root-mean-square error, average error and the setting threshold size between fused image and ideal image, if be less than setting threshold, mixing operation finishes;
Second step: if evaluation coefficient is greater than setting threshold, fused image is merged respectively with source images A and B respectively again, two width images after again merging and the evaluation coefficient between ideal image are compared, select the desirable image of fusion situation, be evaluation coefficient smaller, set corresponding with it source images, A or B are as the image of iteration fusion below;
The 3rd step: the evaluation coefficient of desirable fused images and setting threshold are compared, if be less than threshold value, finish mixing operation, this width image is as final fusion results; If be greater than threshold value, the image after again this width being merged and before selected source images carry out mixing operation, until the evaluation coefficient of the image after fusion is less than the threshold value of setting;
Described bionic compound eyes surface defects detection flow process comprises the following steps:
(1) scene image that, utilizes visual properties compression module to obtain splicing does non-linear compression;
In described bionic compound eyes surface defects detection process step (1), the scene image that visual properties compression module obtains splicing does non-linear compression, adopts non-linear compressed transform formula to be:
I in this formula
c(x, y) is the brightness value after compression, and in this formula denominator part, Section 2 is the method for the moving average of employing, obtains like this midrange coming and has adaptivity;
(2) rectification processing and the central side, integrated based on threshold value based on local contrast 3 D defects detection module suppress to process;
In described bionic compound eyes surface defects detection process step (2), by calculating local contrast C
local(x, y) selects to exist the image block of surface imperfection to process, local contrast C
local(x, y) can weigh in this scene, whether there is the surface imperfection that can be identified, the position that contrast is larger occurs that the possibility of surface imperfection is larger, selecting this position is the center of detected image block, here suppose that the size of surface imperfection is 4 pixel left and right, what therefore local contrast adopted is the local contrast of 2 × 2 image blocks, rather than single pixel, and local contrast computing formula is as follows:
Σ I (x, y) is the brightness value sum of an image block of 2 × 2 take (x, y) as top left corner pixel in image, I
meanfor the image block average brightness of these 6 × 6 sizes centered by image block;
(3), defect merge output by horizontal direction and two channel signals in vertical direction respectively by together with the Fusion of Cells of pond, then the result of fusion is merged to the output that just can obtain last defect again;
In described bionic compound eyes surface defects detection process step (3), in analysis level directional image, from left to right then the reduction of brightness is before this increase of brightness, the be separated by distance of several pixels of centre, off image overlaps in the edge of defect with on image through translation, multiply each other like this and will obtain the result of non-zero, if and there is no surface imperfection in this image block, the namely edge of neither one closure, in off image or on image, lack an edge so, after multiplying each other, result is zero, in vertical direction, be in like manner, upwards there is an edge in the non-vanishing explanation the party of fusion output in horizontal direction, upwards also there is an edge in the non-vanishing explanation the party of fusion output in vertical direction, adopt the mode of logical and to merge two orthogonal directionss, there is surface imperfection if export non-vanishing explanation, export so this signal.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201110338628.9A CN102507592B (en) | 2011-11-01 | 2011-11-01 | Fly-simulation visual online detection device and method for surface defects |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201110338628.9A CN102507592B (en) | 2011-11-01 | 2011-11-01 | Fly-simulation visual online detection device and method for surface defects |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102507592A CN102507592A (en) | 2012-06-20 |
| CN102507592B true CN102507592B (en) | 2014-05-28 |
Family
ID=46219699
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201110338628.9A Active CN102507592B (en) | 2011-11-01 | 2011-11-01 | Fly-simulation visual online detection device and method for surface defects |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102507592B (en) |
Families Citing this family (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103091330A (en) * | 2013-01-11 | 2013-05-08 | 河海大学常州校区 | Copper strip surface defect identifying device for simulating human visual perception mechanism and method thereof |
| CN103293523B (en) * | 2013-06-17 | 2015-04-22 | 河海大学常州校区 | Hyperspectral remote sensing small target detection method based on multiple aperture information processing |
| CN105892501A (en) * | 2013-09-10 | 2016-08-24 | 充梦霞 | Working method of self-adaptive image information acquisition system |
| CN104270576B (en) * | 2014-10-23 | 2017-07-04 | 吉林大学 | A kind of bionic telescopic formula sector compound eye |
| JP6276734B2 (en) * | 2015-07-22 | 2018-02-07 | 矢崎総業株式会社 | Inspection apparatus and inspection method |
| JP6516884B2 (en) * | 2016-02-01 | 2019-05-22 | 三菱電機株式会社 | Connection inspection work support system |
| CN105954292B (en) * | 2016-04-29 | 2018-09-14 | 河海大学常州校区 | Underwater works surface crack detection device based on the bionical vision of compound eye and method |
| CN106097315A (en) * | 2016-06-03 | 2016-11-09 | 河海大学常州校区 | A kind of underwater works crack extract method based on sonar image |
| CN107784660B (en) * | 2017-11-02 | 2020-12-04 | 武汉新芯集成电路制造有限公司 | Image processing method, image processing system and defect detection device |
| CN109949261B (en) * | 2017-12-15 | 2021-03-19 | 中科晶源微电子技术（北京）有限公司 | Method for processing graphics, graphics processing device and electronic equipment |
| CN108416765B (en) * | 2018-01-30 | 2020-08-18 | 华南理工大学 | Method and system for automatically detecting character defects |
| CN108416794A (en) * | 2018-03-21 | 2018-08-17 | 湘潭大学 | A kind of nickel foam surface defect image dividing method |
| CN108537150B (en) * | 2018-03-27 | 2019-01-18 | 长沙英迈智越信息技术有限公司 | Reflective processing system based on image recognition |
| CN108986086A (en) * | 2018-07-05 | 2018-12-11 | 福州大学 | The detection of typographical display panel inkjet printing picture element flaw and classification method and its device |
| CN108982514A (en) * | 2018-07-12 | 2018-12-11 | 常州大学 | A kind of bionical vision detection system of casting surface defect |
| CN109557944B (en) * | 2018-11-30 | 2021-08-20 | 南通大学 | Moving target position detection method |
| CN110310314B (en) * | 2019-03-26 | 2022-01-25 | 上海联影智能医疗科技有限公司 | Image registration method and device, computer equipment and storage medium |
| CN110132327B (en) * | 2019-06-05 | 2021-09-17 | 知恒科技(天津)有限公司 | Photoelectric encoder |
| CN111340715B (en) * | 2019-09-19 | 2024-02-06 | 杭州海康慧影科技有限公司 | Grid pattern weakening method and device of image and electronic equipment |
| CN112547576A (en) * | 2020-12-21 | 2021-03-26 | 山东大舜医药物流有限公司 | Vaccine sorting line ex-warehouse rechecking system |
| CN114140679B (en) * | 2021-10-26 | 2022-07-01 | 中科慧远视觉技术(北京)有限公司 | Defect fusion method, device, recognition system and storage medium |
| CN116124067B (en) * | 2023-04-04 | 2023-07-07 | 山东大学 | Multi-sensor wheel diameter measuring instrument and measuring method |
| CN118115501A (en) * | 2024-04-30 | 2024-05-31 | 成都泰特润博检测技术有限公司 | Data analysis method and system based on industrial CT imaging |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US4750022A (en) * | 1985-12-27 | 1988-06-07 | Canon Kabushiki Kaisha | Optical projection system |
| CN101867828A (en) * | 2009-12-31 | 2010-10-20 | 南京工程学院 | Multi-charge coupled device group self-adaptive imaging instrument of cambered optical structure and method thereof |
| CN102081296A (en) * | 2010-12-01 | 2011-06-01 | 南京航空航天大学 | Device and method for quickly positioning compound-eye vision imitated moving target and synchronously acquiring panoramagram |
-
2011
- 2011-11-01 CN CN201110338628.9A patent/CN102507592B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US4750022A (en) * | 1985-12-27 | 1988-06-07 | Canon Kabushiki Kaisha | Optical projection system |
| CN101867828A (en) * | 2009-12-31 | 2010-10-20 | 南京工程学院 | Multi-charge coupled device group self-adaptive imaging instrument of cambered optical structure and method thereof |
| CN102081296A (en) * | 2010-12-01 | 2011-06-01 | 南京航空航天大学 | Device and method for quickly positioning compound-eye vision imitated moving target and synchronously acquiring panoramagram |
Non-Patent Citations (5)
| Title |
|---|
| 在线超声检测缺陷波信号的仿生采集技术;施克仁等;《无损检测》;19970131;第19卷(第1期);2-4 * |
| 基于昆虫视觉原理的角速度测量系统;陶圣等;《清华大学学报(自然科学版)》;20070831;第47卷(第8期);1308-1311 * |
| 施克仁等.在线超声检测缺陷波信号的仿生采集技术.《无损检测》.1997,第19卷(第1期),2-4. |
| 邢强.仿昆虫复眼结构的运动目标位置检测.《中国优秀硕士学位论文全文数据库（电子期刊）》.2011,参见第二章第2.2节. * |
| 陶圣等.基于昆虫视觉原理的角速度测量系统.《清华大学学报(自然科学版)》.2007,第47卷(第8期),1308-1311. |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102507592A (en) | 2012-06-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102507592B (en) | Fly-simulation visual online detection device and method for surface defects | |
| CN104599258B (en) | A kind of image split-joint method based on anisotropic character descriptor | |
| CN102901489B (en) | Surface gathered water accumulated ice detection method and device | |
| CN109655825A (en) | Data processing method, device and the multiple sensor integrated method of Multi-sensor Fusion | |
| CN110047144A (en) | A kind of complete object real-time three-dimensional method for reconstructing based on Kinectv2 | |
| CN110689562A (en) | Trajectory loop detection optimization method based on generation of countermeasure network | |
| CN109615611A (en) | A kind of insulator self-destruction defect inspection method based on inspection image | |
| CN111291639B (en) | Cross-source ship feature fusion learning and identifying method based on layered variation self-coding | |
| Wei | Converting 2d to 3d: A survey | |
| CN106155299B (en) | A kind of pair of smart machine carries out the method and device of gesture control | |
| CN110378995B (en) | Method for three-dimensional space modeling by using projection characteristics | |
| CN114694011A (en) | Fog penetrating target detection method and device based on multi-sensor fusion | |
| US20150170369A1 (en) | Specularity determination from images | |
| CN111612717A (en) | Water surface image reflection removing and repairing method and device based on countermeasure network | |
| CN109919832A (en) | One kind being used for unpiloted traffic image joining method | |
| CN116091314A (en) | Infrared image stitching method based on multi-scale depth homography | |
| CN116978009A (en) | Dynamic object filtering method based on 4D millimeter wave radar | |
| CN111951339A (en) | Image processing method for performing parallax calculation by using heterogeneous binocular cameras | |
| Wang et al. | Mirrornerf: One-shot neural portrait radiance field from multi-mirror catadioptric imaging | |
| Babu et al. | An efficient image dahazing using Googlenet based convolution neural networks | |
| Zhao et al. | Image dehazing based on haze degree classification | |
| Shit et al. | An encoder‐decoder based CNN architecture using end to end dehaze and detection network for proper image visualization and detection | |
| CN115131459B (en) | Reconstruction method and device for floor plan | |
| CN111126508A (en) | Hopc-based improved heterogeneous image matching method | |
| CN116433580A (en) | Dam crack detection method based on intelligent unmanned aerial vehicle |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20230315 Address after: Room 201-3, Building D, Vitality Business Plaza, 185 Jumao Street, Yuanhe Street, Xiangcheng District, Suzhou City, Jiangsu Province, 215131 Patentee after: Huixinjia (Suzhou) Intelligent Technology Co.,Ltd. Address before: 213022, No. 200, Jinling North Road, Xinbei District, Jiangsu, Changzhou Patentee before: CHANGZHOU CAMPUS OF HOHAI University |
|
| TR01 | Transfer of patent right |