JPS63247605A - Component recognizing method - Google Patents

Abstract

PURPOSE:To find the position and inclination of a component without contacting by approximating the contour of the component by plural straight lines and finding corner data. CONSTITUTION:An image of the component is picked up and digitized to generate an image pattern 1, and a contour point PO of the pattern and contour points and connecting directions of one round of the pattern are detected from a point PS in the pattern. Then the angle of deviation of the straight line connecting two points which are at a distance of the constant number of points are calculated and sections where the difference between adjacent angles of deviation is zero continuously are detected as a straight line to obtain straight lines 3-10. Then one straight line 11 is restored by using straight lines 6-8 which have the same angle of deviation and adjoin to one another among the respective straight lines 3-10. Further, straight lines 12 and 13 which are shorter than the specific shortest length among remaining straight lines are regarded as noises and removed. The remaining straight lines 14, 15, 16, and 17 after said processing are used to obtain a polygon approximating the contour of the pattern 1, and respective corner data are calculated and matched with respective data which are taught previously to recognize intersections 18-21 as reference positions at the time of calculating position and inclination.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to a component recognition method, and particularly to a chip-type electronic component (
The present invention relates to a component recognition method that can be suitably applied to detecting the position of electronic components, which is necessary when mounting IC components (hereinafter referred to as chip components) or short leads on a circuit board.

Conventional technology Position correction methods for mounting chip components or IC components with short leads (hereinafter simply referred to as components) on a circuit board include a mechanical method using alignment claws and an electrical method using image processing. There is.

FIG. 6 shows a mechanical position correction mechanism. To briefly explain this operation, the component 32 attracted by the suction nozzle 31 is moved by the regulating claws 33 and 34 that approach at the same rate at the same time. The amount of positional deviation in the X direction and the y direction is corrected. As a result, the suction nozzle 31 and the part 32
The positional relationship between the suction nozzle 31 and the component 32 is always kept constant, and the positional deviation between the suction nozzle 31 and the component 32 that occurs mainly when the component supply section picks up the component is corrected.

In addition, as a position correction method using image processing, a silhouette image of the part is captured by an imaging means such as a television camera, and the digital image obtained by quantizing the video signal from the imaging means is processed to determine the position and inclination of the part. A common method is to detect

Although there are several specific methods for detecting the position and inclination of a component, two typical methods will be described here.

The first method is position detection using template matching. To briefly explain using FIGS. 7a, b, and c, subpatterns 42 and 43 containing the characteristics of the pattern are cut out from the image pattern 41 of the recognition target part held in the frame memory as a digital image. Figure 7a
). The cut out sub-pattern is held in a memory separate from the image pattern 41, and is used as the template A44,
Template C45 is used (FIG. 7b). This is the teaching process. Through this process, the position of the image pattern 41 can be expressed by the coordinates of the upper left corner (PA shown in FIG. 7a), which is the representative point of the template, and the inclination of the image pattern 41 can be expressed by , can be expressed by the slope of a straight line connecting the representative points of the two templates (PA and PC shown in FIG. 7a). During recognition processing, matching is performed while moving the template on the digital image stored in the frame memory, and the position where the degree of matching between the digital image and the template is highest is detected. FIG. 7C shows that as a result of matching the image pattern 46 taken into the frame memory during recognition processing with template A44 and template 045, the matching degree is maximized in PA and PC, respectively. and PC'' are the positions detected by template matching.The slope of the image pattern 46 is represented by the slope of a straight line connecting two points PA'' and PB''.

In the second method, corners of the pattern are detected by processing the contour shape of the image pattern, and the pattern is recognized based on the number, type, and distance of the detected corners. Figure 8a.

To briefly explain using b, a, and d, the first contour point Po of the pattern is detected from the point P3 inside the image pattern 61 of the recognition target part held in the frame memory as a digital image. And this contour point P. The contour points for one rotation of the pattern are tracked clockwise and detected one after another. Next, the declination angle θ of a straight line 52 connecting contour points separated by a certain number of points is calculated for the contour points for one round of the pattern, and the difference in the declination angle Δθ is calculated (FIG. 8b). Four corners A of the image pattern 61 shown in FIG. 8a,
B, C, and D are Δθ53,5 of one group in Figure 8.
4,55.56, and the value obtained by adding Δθ in each group becomes the amount of angle change of that corner. 4 corners shown in Figure 8, A.

Since B, C, and D are all corners on the outline that change the direction of travel by 90 degrees to the right, these will be referred to as plus 900 corners (minus corners change the direction of travel to the left). The representative position of each corner is the point where the sum of Δθ's in each group is half the value of the sum of all Δθ's. During the teaching process, the number of corners detected through the above process, the type of each corner (how many corners there are), the arrangement of the corners, and the distance between adjacent corners are held. Image pattern 5 captured during recognition
Similar processing is performed for 7, and matching is performed on the n-axis in order to check which of the corners held at the time of teaching each corner obtained as a result corresponds to. This matching is performed by comparing the number, type, arrangement, and distance of the corners held at the time of teaching, and determining the correspondence between the other corners. In Fig. 8d, the first detected corner Ds
Corner A63 by comparing s with the teaching data in Figure 8.
However, the distance conditions with the adjacent corners do not match. Then, when matching is attempted by regarding corner A59 as corner A53, all conditions match. In this way, by correlating the corners detected during recognition with the corners during teaching, the position and inclination of the pattern can be recognized.

Problems to be Solved by the Invention There are two problems that arise when a mechanical position correction mechanism is used. Correction errors occur due to wear of one regulating claw and parts fall. Naturally, since the regulating claw comes into contact with parts, wear cannot be avoided, and sooner or later it will wear out. M
FIG. 9a shows from the side how the component 62 sucked by the suction nozzle 61 is corrected for positional deviation by the regulating claw 63.

The regulating claws 63 hold the part 62 when the part 62 is sandwiched from both sides.
It is tapered so that a force is applied to press the suction nozzle 61 against the suction nozzle 61. Therefore, the worn regulating claw 64 has a shape as shown in FIG. 9.

Moreover, in reality, parts with various sizes and hardnesses are calibrated using the same standard pawl, so the manner in which the standard pawl wears out is quite complicated, and abnormal external forces are applied when the parts are sandwiched. This may cause correction errors or parts to fall. Another problem is that at least two types of regulating portions must be provided depending on the size of the component. In recent years, with the development of high-density packaging technology, the tendency for electronic components for circuit boards to be made into chips has increased more than ever. Small is 1.6 X O, BMM
Parts from 101 uI square in size are commercially available.
Moreover, there is an increasing demand for a mounting machine that can mount all of these various chip components in one machine.

In order to mount chip components of different sizes in one mounting machine, at least two parts are required: a regulating part for small components as shown in FIG. 10a, and a regulating part for large components as shown in FIG. 10. A different type of alignment is required, leading to maintenance and cost problems combined with the alignment wear problem discussed above.

Therefore, a non-contact electric position correction method was adopted instead of a contact-type mechanical position correction method, but the following methods also apply to conventional techniques such as template matching and corner matching using contour shape processing. There are problems.

First, the biggest problem with the template matching method is that it is vulnerable to pattern rotation. This is because, as shown in FIG. 11, the template 71 cut out from the image pattern during teaching always moves within the screen in the horizontal-vertical direction, so that a reliable and high match can be obtained for a greatly tilted image pattern 72. The position indicating the degree cannot be detected. The tilt angle that can be detected by template matching is at most ±10
Considering that the tilt angle is about .degree., it is hard to say that template matching is suitable for detecting the position of a chip component in a chip component mounting machine, where a tilt larger than that is expected when the component supply section picks up the chip component.

Furthermore, corner matching based on contour shape processing cannot detect the accurate position of the chip component. Some chip components have sharp corners, but some chip components 81 have blunt corners as shown in FIG. 12a.

In the case of such parts, the way the corners are blunted is not uniform, so if you calculate the representative position of the corner, it will be the position PA as shown in Figure 12, or if the corner has a large degree of dullness, it will be the position P''. The difference between the positions of PA and PA" becomes an error. Furthermore, in the case of a chip component 82 having a radiused portion as shown in FIG. If not, Figure 12d
Chip component 83 with dust attached to the corner as shown in
In this case, a problem arises in that the position of the corner D is shifted or the type (angle amount) of the corner changes.

It is an object of the present invention to solve the above-mentioned conventional problems and to provide a non-contact component recognition method that is resistant to rotation of a target component and has high accuracy.

Means for Solving the Problems In order to achieve the above object, the present invention includes means for approximating the outline of a part by a plurality of straight lines, means for calculating the declination angle of the straight line, and means for approximating the outline of the part by a plurality of straight lines. approximating the contour of the part by using a means for calculating the angle formed by two adjacent straight lines among the plurality of straight lines and the coordinates of the intersection, and using the angle formed by the two adjacent straight lines and the coordinates of the intersection as a set of corner data; find the number of sets of corner data equal to the number of the plurality of straight lines,
The object of the present invention is to provide a component recognition method that detects the position and inclination of a component using some of the corner data.

Note that in detecting the plurality of straight lines that approximate the outline of the part, first, contour points for one round of the pattern are detected from the image pattern obtained by digitizing the image obtained by imaging the part, and then a certain number of points are detected. Calculate the declination angle of a straight line connecting two contour points that are separated by the same amount as the contour points for one round of the pattern, and define the section in which the difference between adjacent declination angles is continuously zero or a value small enough to be considered zero. It is preferable to use a method of detecting as a straight line.

In addition, when a straight line that approximates the contour of a part is divided, a method is used to restore the divided straight line to a single straight line by calculating and comparing the declination angles of each divided straight line. is preferred.

Function: Since the present invention has the above configuration, the position and inclination of components can be detected without contact, and maintenance and adjustment are easy. Furthermore, regardless of the size of the component, it can be recognized as long as it can be imaged, and it can also be expected to have the advantage of saving space when designing a mounting machine. Furthermore, since parts are recognized by detecting straight lines that make up most of the part's outline, it is resistant to part rotation and can accurately detect the position and tilt of chip parts with blunt corners or rounded corners. can.

Furthermore, when a straight line is divided, restoring it to a single straight line helps to improve recognition efficiency.

In addition, as a method for detecting the contour points for one rotation of the pattern from the image pattern, a known method is adopted, such as determining the connection direction of the contour points based on the contents of the 8-neighborhood data and detecting the contour points while tracking them. be able to.

EXAMPLE Hereinafter, an example of the present invention will be described with reference to FIGS. 1 and 2.

Figure 1 shows that a chip component with a somewhat complex outline is captured in a frame memory as a silhouette image, and then
The flow of processing until the position and inclination are detected is shown in order from a to e.

Reference numeral 1 is an image pattern of a chip component, and first, a contour point P□ of the pattern is detected from a point Ps inside this image pattern. Then, the contour points for one rotation of the image pattern are detected clockwise from the contour point P□.To detect the contour points, the connection direction of the contour points is determined based on the contents of the 8 neighborhood data 2, and the contour points are tracked. A method of detection is used. Next, the declination angle of a straight line connecting two contour points separated by a certain number of points is calculated for the contour points for one round of the pattern, and the difference between adjacent declination angles is continuously zero (or can be considered zero). The section where the value is as small as 3 to 1o are straight lines detected in this way. Then, the angle of argument is determined for each of the straight lines 3 to 10, and it is determined whether straight lines with the same angle of argument are adjacent to each other. Since it is found that straight lines 6 to 8 have the same angle of deviation and exist adjacent to each other, one straight line 11 is restored from these straight lines 6 to 8. Further, at this stage, the remaining straight lines are compared with a predetermined minimum length, and straight lines 12 and 13 shorter than the minimum length are considered to be noise and removed. Straight lines 14 and 15 left after the above processing
, 16.17 are straight lines that approximate the contour of the image pattern of a chip component having a somewhat complicated shape. Therefore, each corner data of a polygon that approximates the image pattern 1 of the chip component is calculated by determining all the angles formed by two adjacent straight lines and the coordinates of the intersections of each straight line. Once the corner data is determined, the subsequent processing is exactly the same as the corner matching by contour shape processing described as one of the conventional examples. That is, the two types of number, arrangement, and distance of advantageous corners at the time of recognition are compared with each data held at the time of teaching, and the other corner is matched. This correspondence makes it possible to detect the position and inclination of a chip component with a somewhat complex contour. 18-21 are straight lines 14-17
It represents the intersection of , and serves as the reference position when calculating the position and inclination.

FIGS. 2a to 2d are diagrams showing how to obtain a straight line that approximates the outline of a pattern from the outline points of the pattern, and the image pattern 22 includes noise 23 and recesses 24. The contour points of this image pattern are obtained by detecting contour points P□ from points Ps inside the pattern, and then tracing the contour points one after another clockwise from Po. Then, the declination of a straight line connecting two points separated by a certain number of points (m in Figure 2) (in Figure 2, the straight line connecting Pn and Pn+.
1 and Pn+1+. Declination angle θyu and θn+1 of the straight line connecting
) is calculated for the contour points of one round of the pattern. Further adjacent declination angle θ. , θn+1 is calculated. Figure 2C
is a graph in which the vertical axis is the difference in argument Δθ and the horizontal axis is n, and represents the four corners pA and pB of the image pattern 22.

PC and PD are represented by upwardly convex curves 25. Further, the noise 23 is represented by an upwardly convex curve 26, and the recess 24 is represented by a downwardly convex curve 27. Here, for convenience of explanation, the term "upward convex curve" or "downward convex curve" is used, but in reality, n on the horizontal axis takes discrete values, so each curve 25.26.
27 represents Δθ of one group, and the value obtained by adding Δθ in each group represents the angle of each group (that is, 26 is a corner, 26 is noise, and 27 is a concave portion). Therefore, 26 represents the angle of the noise 23, but since the individual Δθ, which is a component of 26, is small enough to be considered zero, there is one straight line 28 between the corners P A and P B. expressed. On the contrary, the concave portion 24 between corners PB and PC becomes a large negative coat as shown by 27, so that it is divided into two straight lines 29. Also,
There is no noise or recess between the corners PD and PA, but due to processing reasons, the line is divided by P() into two straight lines 3o. For the straight line 29.30 divided in this way, if we calculate the declination of all the detected straight lines and integrate the straight lines with the same declination next to each other, we can obtain 1.
It can be restored as a book straight line 31.32.

Through the processing described above, a straight line that approximates the contour of the pattern is obtained from the contour points of the pattern.

In the above embodiment, the corner angle is calculated by adding the differences Δθ in argument angles. However, by taking the arithmetic average of several adjacent Δθ values before addition, all the differences Δθ in argument angles can be calculated. The corner angle may be determined by substituting the average Δθ of the differences in argument angles and adding the average Δθ of the differences in argument angles.

Furthermore, in this case, the value of Δθ for the noise 23 shown in FIG. 2 becomes even smaller due to smoothing, and compared to the case where the difference Δθ in argument is added as is,
This makes it difficult to be affected by noise.

Furthermore, in the explanation of Figure 1, it was assumed that straight lines 12.13 shorter than the minimum length were removed as noise when compared with a predetermined minimum length, but straight lines 12.13 are longer than the minimum length. If the image pattern 1 of the chip component has a shape, the outline of the image pattern 1 of the chip component is approximated by four or more straight lines as shown in Fig. You may choose from among them.

Furthermore, components that can be recognized by the component recognition method according to the present invention are not only chip components but also ICs with short leads as shown in FIG.
It can also be applied to the component 110. In the case of such parts, when calculating the declination angle of a straight line connecting two contour points separated by a certain number of points, the distance between the contour points is widened, and the declination angle to be added when calculating the corner angle is This is because when calculating the average difference Δθ, by increasing the number of data to be averaged, irregularities on the contour can be smoothed and approximated by a straight line, and virtual corners 111 to 114 can be detected.

Figures 6a and 6b show a conventional component recognition method using corner detection (Figure 6a) and a component recognition method according to the present invention (Figure 6b).
) are shown in comparison. According to the method of the present invention, it is more resistant to noise 120 than when directly detecting the corner itself and using it for position correction as in the conventional example, and it can be recognized with high accuracy even when there are dull corners or rounded portions 121. I understand.

Effects of the Invention According to the component recognition method of the present invention, the contours of chip components having a wide variety of sizes and shapes, and IC components with short leads having uneven contours are approximated by straight lines, and the approximated straight lines are By determining corner data from two adjacent straight lines, the position and inclination of the component can be determined without contact. In addition, it is resistant to noise, can be recognized accurately even when there are blunt corners or rounded corners, and is highly resistant to rotation.

[Brief explanation of the drawing]

1 and 2 show an embodiment of the present invention, and FIG.
-e is an explanatory diagram showing the rough processing flow according to the component recognition method of the present invention, Fig. 2 a - d is an explanatory diagram showing how to obtain a straight line that approximates the contour from contour points, and Fig. 3 is an explanatory diagram showing how to approximate the contour. FIG. 4 is an explanatory diagram showing another example of how to find a straight line; FIG. 4 is an explanatory diagram showing a position detection state of an IC component with a short lead;
Figures 6a and b are comparison diagrams with the conventional example, WJe figures to Figure 12 are explanatory diagrams of the conventional example, Figure 6 is an explanatory diagram of the mechanical adjustment method, and Figures 7a-C are template matching. Figures 8a-d are explanatory diagrams showing position detection by corner matching using contour shape processing, Figures 9a, b, and 10a, b are mechanical An explanatory diagram showing the disadvantages of the regulation method, FIG. 11 is an explanatory diagram showing the disadvantages of the position detection method using template matching, and FIG. 12a
= d is an explanatory diagram showing the disadvantages of a method of detecting a position by corner matching based on contour shape processing. 1...Image pattern of chip parts, 16-16
....straight line approximating the outline of the image pattern, 18
~21...Detection position. Name of agent: Patent attorney Toshio Nakao and one other person
2 Figure 3 Broom Figure 4? J5 Figure 8 Figure 9 Figure 10 Figure 11 Figure 12 Figure 8

Claims (2)

[Claims] (1) means for approximating the outline of a part by a plurality of straight lines; means for calculating the angle (hereinafter referred to as an argument) that the straight line makes with a reference axis; and means for approximating the outline of the part by a plurality of straight lines; The plurality of lines approximate the contour of the part by using means for calculating the angle formed by two adjacent straight lines and the coordinates of the intersection point, and using the angle formed by the two adjacent straight lines and the coordinates of the intersection point as a set of corner data. A component recognition method comprising: obtaining a number of sets of corner data equal to the number of straight lines, and detecting the position and inclination of the component using some of the corner data. (2) In detecting the plurality of straight lines that approximate the outline of the part, first, the outline points for one round of the pattern are detected from the image pattern obtained by digitizing the video obtained by imaging the part, and then a certain number of points are detected. Calculate the declination angle of a straight line connecting two contour points that are separated by the same amount as the contour points for one round of the pattern, and define the section in which the difference between adjacent declination angles is continuously zero or a value small enough to be considered zero. The component recognition method according to claim 1, which detects the line as a straight line. (2) When the straight line that approximates the contour of the part is divided,
2. The component recognition method according to claim 1, wherein a divided straight line is restored to a single straight line by calculating and comparing the deviation angles of each divided straight line.