CN1529286A - Spray deposited blank dynamic size visual dtecting method and data acquisition device - Google Patents

Abstract

The invention includes preprocessing method, method for detecting edge, and method for calculating size. The image edge detection method discloses a new edge detection operator of Sobel-Zernike moments. The operator combines edge detection operator in pixel level with edge detection operator in subpixel level. Iron barrel tube and glass are installed between pickup head and deposited butt in data collecting equipment so as to prevent image quality influenced from dust and sprayed dripping. The invented method possesses advantages of high detection precision and operating speed of algorithm.

Description

The visible detection method of injection deposition blank dynamic dimension and data collector

Technical field: the invention belongs to digital picture and detect and process field, particular content relates to a kind of data collector and utilizes image processing techniques to extract the method for injection deposition blank size.

Background technology: in the spray deposition processing, the deposit preform dimensional accuracy is the important indicator that characterizes the reaction-injection moulding macro-effect, and therefore, the dynamic dimension that detects deposit preform becomes important research direction in the reaction-injection moulding field.In order to grasp the dynamic changing process of deposit preform size, requirement is in course of injection, should detect the deposit preform size in real time, because these work are the automatic control of the deposit preform dimensional accuracy that faces the future, this just requires to detect and should satisfy precision and real-time requirement simultaneously.From the accuracy of detection branch, existing edge detection method can be divided into two classes: pixel edge detects operator and sub-pixel edge detects operator, and the advantage of pixel edge operator is that detection speed is fast, but accuracy of detection often is difficult to meet the demands; The advantage that sub-pixel edge detects operator is the accuracy of detection height, but the time of detecting is long, and real-time is difficult to meet the demands.Once introduced too drastic light sensors and go out the deposit preform size in existing documents and materials, its shortcoming is to cost an arm and a leg, and the requirement of the installation site of laser sensor is very strict, and however, the accuracy of detection of deposit preform size still is difficult to guarantee.Injection deposition blank behavioral characteristics visual information acquisition process is to be senser element with common CCD camera, utilizes suitable Processing Algorithm then, detects the deposit preform size quickly and accurately.For this reason, must solve following two problems: 1, vision sensor must " be seen " deposit preform, so just can extract deposit preform for the image processing algorithm of back and carry out the prerequisite preparation; 2, must select suitable image processing algorithm, accuracy of detection and algorithm two indexs working time are met the demands simultaneously.And in practice, directly utilize the CCD camera still to be difficult to collect high-quality image, reason is the existence disturbed in the course of injection, make the image of gathering be subjected to having a strong impact on of noise, exist two noise likes to influence collection image quality in the course of injection: the one, a large amount of metallic dust of disperse in the spray chamber, make deposit preform image blurring unclear, be difficult to tell the deposit preform edge; The 2nd, the splashing of molten drop causes the camera lens surface to paste large quantitative metal liquid and drips, and makes to produce the random noise that is difficult to eliminate in the image, and hence one can see that, eliminates the influence of this two noise like, is the key of improving picture quality.

Summary of the invention: the digital image information that the present invention will utilize data collector to collect, utilize visible detection method to finish the detection of deposit preform behavioral characteristics again, make the detection of deposit preform size satisfy precision, real-time requirement simultaneously.The visible detection method of injection deposition blank dynamic dimension of the present invention comprises first step preprocess method 1, the second step edge detection method 2 and the 3rd step size computing method 3, its the second step edge detection method 2 is: at first be Pixel-level Sobel edge detection operator coarse positioning edge, its process is: calculate the x of each pixel of pretreatment image earlier, the directional derivative S of y direction _x, S _y(2-1); Compute gradient g (x, y) (2-2) then; Set a threshold value t again, the t value is 60, (2-3), judge then g (x, y)＞t? (2-4), be, can judge that then this point is the probable edge point, and the information that will put is recorded in the structural type array point[x of position] in (2-5), not, then return and calculate next pixel gradient g (x, y) (2-2) step; The information of all probable edge points is recorded in the structural type array point[x of position the most at last] in [2-5], the number of all probable edge points is deposited in (2-6) among the variable nNumber; Be that sub-pixel Zernike moments edge detection operator is reorientated the edge then, its process is: the template of the Zernike moments square of at first deriving, comprise a real number template and a plural template (2-7), again from structural type array point[x] take out the probable edge point (2-8) of next coarse positioning, utilize the template of deriving to calculate the Zernike moments square of this probable edge point then, obtain square A ₂₀, A ₁₁(2-9); Then be calculated as follows marginal points information Φ, k, l (2-10):

$\mathrm{\φ}={\tan }^{-1}\left(\frac{\mathrm{Im}\left[A_{11}\right]}{\mathrm{Re}\left[A_{11}\right]}\right);$ $A_{11}^{\′}=A_{11}{e}^{-\mathrm{j\φ}};$ $l=A_{20}/A_{11}^{\′};$ $k=\frac{{3A}_{11}^{\′}}{2{(1-l^2)}^{3/2}};$

Judge marginal point with k, l then, judge whether to satisfy simultaneously $l>\sqrt{2}/12$ And k＞30? (2-11), not, return from structural type array point[x] take out coarse positioning probable edge point (2-8) step, be, then be defined as marginal point, and the pixel coordinate of this marginal point is stored in array pixel_edgepoint[x] in (2-12), be calculated as follows the sub-pixel coordinate of this marginal point then and be stored in array subpixel_edgepoint[x simultaneously] in (2-13): subpoint1.x=point1.x+3.5lcos Φ; Subpoint1.y=point1.y+3.5lsin Φ; Is the number of judging marginal point again more than or equal to nNumber (2-14)? be, binaryzation edge image then, (2-15), not, then return from structural type array point[x] take out coarse positioning probable edge point (2-8) step, up to being recorded in array point[x] in all probable edge points all rejudge and finish; The dimension process that its 3rd step is calculated arbitrary position on the deposit preform is: determine that earlier any point on the deposit preform picture centre is a preliminary sweep point (3-1), search for along the diametric(al) of deposit preform then, on running into, lower limb point point1, point2 (3-2), then the line segment between these two points is represented the diameter of deposit preform, the actual range of each pixel representative is ySize (3-3) on the diametric(al) that is obtained by off-line calibration, then at array pixel_edgepoint[x] in determine point1 respectively, the position of point2 (3-4), then at array subpixe1_edgepoint[x] in find the sub-pix coordinate subpoint1 of respective edges point, subpoint2 (3-5), calculate the pixel number n umber (3-6) between these two some sub-pix coordinates again, number=[(subpoint1.x-subpoint2.x) ²+ (subpoint1.y-subpoint2.y) ²] ^0.5Then the full-size(d) of deposit preform is the number n umber of picture element between subpoint1 and the subpoint2 coordinate and the product (3-7) of ySize: the computing formula of full-size(d) is: diameter=number*ySize.The data collector of injection deposition blank dynamic dimension comprises spray chamber, CCD camera 5, it also comprises glass 6 and work durm 4, glass outer 6-1 is embedded on the spray chamber door 7, outer work durm 4-1 be arranged in the spray chamber and front end and spray chamber door 7 affixed, gland 8 is arranged on the outside and affixed with spray chamber door 7 of spray chamber, the camera lens 2-1 of described CCD camera 5 is arranged in the groove 8-1 of gland 8 centres, the front end of interior work durm 4-2 is affixed with the rear end of outer work durm 4-1, is connected with inner layer glass 6-2 on the barrel of interior work durm 4-2 front end.The application that focuses on new edge detection method of the present invention, by adopting suitable edge detection operator (boundary operator that the sobel boundary operator of Pixel-level and the zernike square of sub-pixel combine), and then the time that whole deposit preform behavioral characteristics is extracted shortens, precision improves, it is fast that the visible detection method of deposit preform dynamic dimension of the present invention has the travelling speed of accuracy of detection height, algorithm, and have the advantage that satisfies precision, real-time requirement simultaneously.CCD camera in the data collector of the present invention can be gathered image, finishes the conversion of optical signalling to electronic signal; Glass outer 6-1 is embedded on the spray chamber door and places the camera front end, is used to protect the cleaning of camera glass sheet; Outer work durm 4-1 places the camera front end, be fixed on the spray chamber door, combine a sealing of formation work durm with glass outer 6-1, be used to eliminate the interference that dust brings in the spray chamber, improve picture quality, harvester of the present invention increases the work durm of certain-length, sealing before camera after, before transforming the light path that is full of dust is shortened, therefore can make target and background border clear; Interior work durm 4-2 can prevent that molten drop is splashed on the inner layer glass 6-2, eliminates metal and splashes to the influence of image quality.

Description of drawings: Fig. 1 is a schematic flow sheet of the present invention, Fig. 2 is the schematic flow sheet of preprocess method 1, Fig. 3 is the schematic flow sheet of edge detection method 2, Fig. 4 is the schematic flow sheet of size computing method 3, Fig. 5 is a plane sub-pixel edge step model structure synoptic diagram, and Fig. 6 is the data collector structural representation.

Embodiment one: present embodiment is followed successively by preprocess method 1, edge detection method 2 and size computing method 3.Preprocess method 1 is to adopt the local random noise that exists in the average smoothing algorithm removal of images, utilize the adaptive smooth algorithm further to improve picture quality then, strengthen edge of image simultaneously, the process of whole Preprocessing Algorithm is: the primary deposit base image of preparation for acquiring, gray scale f (x to each pixel of original image, y) convolution (1-1) in the 3*3 neighborhood, average then (1-2) obtains new gray-scale value f _a(x, y) (1-3).This process can be improved the influence of random noise, but has blured the edge simultaneously; Be the adaptive smooth algorithm then, the purpose of adaptive smooth algorithm application is further to eliminate random noise and strengthen near the contrast in edge, and the process of this algorithm is: at first calculate the x of each pixel, the gradient G of y direction _x(x, y), G _y(x, y) (1-4): $G_x(x,y)=\frac{1}{2}[f(x+1,y)-f(x-1,y)];$ $G_y(x,y)=\frac{1}{2}[f(x,y+1)-f(x,y-1)];$ Utilize the weighting coefficient W (x of each pixel of gradient calculation of rubber vegetarian refreshments then, y) (1-5), in order to save W (x, y) operation time, weighting coefficient determines that by off-line the gray level of considering entire image is 256, therefore the number of weighting coefficient should not surpass 256 yet, for this reason, the weighting coefficient that off-line is determined is stored in the array, and the time spent can directly extract from array.(x y) recomputates each gray values of pixel points at the 3*3 neighborhood, obtains new gray-scale value f to utilize W at last _e(this algorithm can iterate, till meeting the demands for x, y) (1-6).

$f_e(x,y)=\frac{\underset{i=-1}{\overset{+1}{\mathrm{\Σ}}}\underset{j=-1}{\overset{+1}{\mathrm{\Σ}}}f(x+i,y+j)w(x+i,y+i)}{\underset{i=-1}{\overset{+1}{\mathrm{\Σ}}}\underset{j=-1}{\overset{+1}{\mathrm{\Σ}}}w(x+i,y+i)}$

Edge detection method 2 is the edges that detect the deposit preform size, this process is the core of whole deposit preform feature extraction, algorithm working time and edge precision influence whole algorithm, and detailed process is: calculate the x of pretreated each pixel of image, the directional derivative S of y direction _x, S _y(2-1); Compute gradient g (x, y) (2-2) then; Set a threshold value t again, the t value is 60 (2-3), judge then g (x, y)＞t? (2-4), if g (x y)＞t, can judge that then this point is the probable edge point, and the information that will put is recorded in the structural type array point[x of position] in (2-5); If g is (x, y)≤t, determine that this point is non-marginal point, then return and calculate next pixel gradient g (x, y) (2-2) process, up to finding all possible marginal point, the information of all probable edge points is recorded in the structural type array point[x of position the most at last] in (2-5), the number of all probable edge points is deposited among the variable nNumber; Be that sub-pixel Zernike moments edge detection operator is reorientated the edge then, its process is: the template of the Zernike moments square of at first deriving, comprise a real number template and a plural template (2-7), again from structural type array point[x] take out the probable edge point (2-8) of coarse positioning, utilize the template of deriving to calculate the Zernikemoments square of this probable edge point then, obtain square A ₂₀, A ₁₁(2-9); Then be calculated as follows marginal points information Φ, k, l (2-10):

$\mathrm{\φ}={\tan }^{-1}\left(\frac{\mathrm{Im}\left[A_{11}\right]}{\mathrm{Re}\left[A_{11}\right]}\right);$ $A_{11}^{\′}=A_{11}{e}^{-\mathrm{j\φ}};$ $l=A_{20}/A_{11}^{\′};$ $k=\frac{3A_{11}^{\′}}{2{(1-l^2)}^{3/2}};$

Judge marginal point with k, l then, judge whether to satisfy simultaneously $l>\sqrt{2}/12$ And k＞30? (2-11), not, return from structural type array point[x] take out coarse positioning probable edge point (2-8) step, be, then be defined as marginal point, and the pixel coordinate of this marginal point is stored in array pixel_edgepoint[x] in (2-12), be calculated as follows the sub-pixel coordinate of this marginal point then and be stored in array subpixel_edgepoint[x simultaneously] in (2-13): subpoint1.x=point1.x+3.5lcos Φ; Subpoint1.y=point1.y+3.5lsin Φ; Is the number of judging marginal point again more than or equal to nNumber (2-14)? be, binaryzation edge image (2-15) then, not, then return from structural type array point[x] take out coarse positioning probable edge point (2-8) step, up to being recorded in array point[x] in all probable edge points all rejudge and finish; Size computing method 3 is after edge image obtains, utilize the full-size(d) of the calibration result calculating deposit preform of off-line, the dimension process of calculating arbitrary position on the deposit preform is: determine that earlier any point on the deposit preform picture centre is a preliminary sweep point (3-1), search for along the diametric(al) of deposit preform then, on running into, lower limb point point1, point2 (3-2), then the line segment between these two points is represented the diameter of deposit preform, the actual range of each pixel representative is ySize (3-3) on the diametric(al) that is obtained by off-line calibration, then at array pixel_edgepoint[x] in determine point1 respectively, the position of point2 (3-4), then at array subpixel_edgepoint[x] in find the sub-pix coordinate subpoint1 of respective edges point, subpoint2 (3-5), calculate the pixel number n umber (3-6) between these two some sub-pix coordinates again, number=[(subpoint1.x-subpoint2.x) ²+ (subpoint1.y-subpoint2.y) ²] ^0.5Then the full-size(d) of deposit preform is the number n umber of picture element between subpoint1 and the subpoint2 coordinate and the product (3-7) of ySize: the computing formula of full-size(d) is: diameter=number*ySize.

Embodiment three: with reference to Fig. 3, present embodiment is the data collector of injection deposition blank dynamic dimension, it comprises spray chamber, CCD camera 5, it also comprises glass 6 and work durm 4, glass outer 6-1 is embedded on the spray chamber door 7, outer work durm 4-1 be arranged in the spray chamber and front end and spray chamber door 7 affixed, gland 8 is arranged on the outside and affixed with spray chamber door 7 of spray chamber, outer work durm 4-1 of present embodiment and gland 8 all pass through screw 9 and spray chamber door 7 Joints, the camera lens 2-1 of described CCD camera 5 is arranged in the groove 8-1 of gland 8 centres, the front end of interior work durm 4-2 is affixed with the rear end of outer work durm 4-1, is connected with inner layer glass 6-2 on the barrel of interior work durm 4-2 front end.

The data collector principle of work is as follows: sensor becomes parallel angle to lay with deposit preform substantially, and before the work durm 4-2, molten drop can directly be splashed on the glass outer 6-1 in not adding; After adding interior work durm 4-2, drop must experience the process of flying earlier before being splashed to glass outer 6-1 in interior work durm 4-2, because the flow field in the interior work durm 4-2 is disturbed flow condition, make the running orbit non-rectilinear of molten drop, increased the flight time in interior work durm 4-2 accordingly, this makes molten drop that sufficient time cooling can be arranged before being splashed to glass outer 6-1, be solidified as metallic dust, so just can not paste on the glass outer 6-1, reach and eliminate the purpose that molten drop splashes and disturbs.

Claims (4)

1. the visible detection method of an injection deposition blank dynamic dimension, it comprises that first step preprocess method (1), the second step edge detection method (2) and the 3rd step size computing method (3) is characterized in that its second step edge detection method (2) is: at first be Pixel-level Sobel edge detection operator coarse positioning edge, its process is: calculate the x of each pixel of pretreatment image earlier, the directional derivative S of y direction _x, S _y(2-1); Compute gradient g (x, y) (2-2) then; Set a threshold value t again, the t value is 60 (2-3), judge g (x then, y)＞t? (2-4), be to judge that then this point is the probable edge point, and the information that will put is recorded in the structural type array point[x of position] in (2-5),, then do not return and calculate next pixel gradient g (x, y) (2-2) step; Finally, the information of all probable edge points is recorded in the structural type array point[x of position] in (2-5), the number of all probable edge points is deposited in (2-6) among the variable nNumber; Be that sub-pixel Zernike moments edge detection operator is reorientated the edge then, its process is: the template of the Zernike moments square of at first deriving, comprise a real number template and a plural template (2-7), again from structural type array point[x] take out the probable edge point (2-8) of coarse positioning, utilize the template of deriving to calculate the Zernikemoments square of this probable edge point then, obtain square A ₂₀, A ₁₁(2-9); Then be calculated as follows marginal points information Φ, k, l (2-10):

$\mathrm{\φ}={\tan }^{-1}\left(\frac{\mathrm{Im}\left[A_{11}\right]}{\mathrm{Re}\left[A_{11}\right]}\right);$ A ₁₁＝A ₁₁e-jφ； l＝A ₂₀/A ₁₁′； $k=\frac{3A_{11}^{\′}}{2{(1-l^2)}^{3/2}};$

Judge marginal point with k, l then, judge whether to satisfy simultaneously $l>\sqrt{2}/12$ And k＞30? (2-11), not,

Return from structural type array point[x] take out coarse positioning probable edge point (2-8) step, be, then be defined as marginal point, and the pixel coordinate of this marginal point is stored in array pixel_edgepoint[x] in (2-12), be calculated as follows the sub-pixel coordinate of this marginal point then and be stored in array subpixel_edgepoint[x simultaneously] in (2-13): subpoint1.x=point1.x+3.5lcos Φ; Subpoint1.y=point1.y+3.5lsin Φ; Is the number of judging marginal point again more than or equal to nNumber (2-14)? be, binaryzation edge image (2-15) then, not, then return from structural type array point[x] take out coarse positioning probable edge point (2-8) step, up to being recorded in array point[x] in all probable edge points all rejudge and finish; The dimension process that its 3rd step is calculated arbitrary position on the deposit preform is: determine that earlier any point on the deposit preform picture centre is a preliminary sweep point (3-1), search for along the diametric(al) of deposit preform then, on running into, lower limb point point1, point2 (3-2), then the line segment between these two points is represented the diameter of deposit preform, the actual range of each pixel representative is ySize (3-3) on the diametric(al) that is obtained by off-line calibration, then at array pixel_edgepoint[x] in determine point1 respectively, the position of point2 (3-4), then at array subpixel_edgepoint[x] in find the sub-pix coordinate subpoint1 of respective edges point, subpoint2 (3-5), calculate the pixel number n umber (3-6) between these two some sub-pix coordinates again, number=[(subpoint1.x-subpoint2.x) 2+ (subpoint1.y-subpoint2.y) 2] full-size(d) of 0.5 item deposit preform is the number n umber of picture element between subpoint1 and the subpoint2 coordinate and the product (3-7) of ySize: the computing formula of full-size(d) is: diameter=number*ySize.

2, the visible detection method of injection deposition blank dynamic dimension according to claim 1, it is characterized in that first step preprocess method (1) is: at first be average smoothing algorithm, its process is: to the gray scale f (x of each pixel of original image, y) convolution (1-1) in the 3*3 neighborhood, average then (1-2), obtain new gray-scale value f _a(x, y) (1-3); Be the adaptive smooth algorithm then, its process is: calculate the x of each pixel, the gradient G of y direction earlier _x(x, y), G _y(((x y) recomputates each gray values of pixel points at the 3*3 neighborhood, obtains new gray-scale value f to utilize w at last for x, y) (1-5) to utilize the weighting coefficient W of each pixel of gradient calculation of pixel then for x, y) (1-4) _e(this algorithm can iterate, till meeting the demands for x, y) (1-6).

3, a kind of data collector of injection deposition blank dynamic dimension, it comprises spray chamber, CCD camera (5), it is characterized in that it also comprises glass (6) and work durm (4), glass outer (6-1) is embedded on the spray chamber door (7), outer work durm (4-1) be arranged in the spray chamber and front end and spray chamber door (7) affixed, gland (8) is arranged on the outside and affixed with spray chamber door (7) of spray chamber, the camera lens (2-1) of described CCD camera (5) is arranged in the middle groove (8-1) of gland (8), the front end of interior work durm (4-2) is affixed with the rear end of outer work durm (4-1), is connected with inner layer glass (6-2) on the barrel of interior work durm (4-2) front end.

4, the data collector of injection deposition blank dynamic dimension according to claim 3 is characterized in that described outer work durm (4-1) and gland (8) all pass through screw (9) and spray chamber door (7) Joint.

Images