JPS62159292A - Recognizing method for position of machine screw hole - Google Patents

Abstract

PURPOSE:To estimate the center of a machine screw hole by using a part of the periphery, even when a difference of brightness of the machine screw hole and a background is not so large, by providing a machine screw hole position calculating process for calculating a position of the center of the machine screw hole from a line drawing data. CONSTITUTION:The titled method consists of a line thinning process for generating an image data by extracting a part where brightness is varied, from a two-dimensional variable density image data which has been inputted by an image input device, a machine screw hole existence deciding process for calculating a curvature of a curve on a line drawing data and deciding whether a machine screw hole of a given diameter exists in the original image or not, and a machine screw hole position calculating process for calculating a position of the center of the machine screw hole from the line drawing data. According to this constitution, a round pattern of the machine screw hole can be discriminated from other graphic of the same size by calculating a curve, and also in case when brightness of the background of the machine screw hole is not uniform as shown in the figure by using an image operator, and even when a difference of the brightness of the hole and the background is not so large, a contour of the machine screw hole can be detected.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of Industrial Application The present invention relates to a screw hole position recognition method for calculating screw hole positions from image data.

BACKGROUND OF THE INVENTION In recent years, as a screw hole position recognition method, a method has been used in which the screw hole position is calculated using data obtained by binarizing image data using a fixed threshold value and projecting it in the x and y directions.

The conventional screw hole position recognition method described above will be described below with reference to the drawings.

FIG. 1 is an explanatory diagram of screw holes. Figure 2 shows the screw holes in Figure 1 inputted from a camera and binarized. Figure 3 shows I from the binarized data in Figure 2.

This is data converted to projection data in the y direction. FIG. 4 shows image data of a screw hole when the background brightness of the screw hole is not uniform. FIG. 6 shows the binarized image data of FIG. 4. Figure 6 is from the binarized data in Figure 6.
+ It is converted into projection data in 7 directions.

Next, the conventional screw hole position recognition method described above will be specifically explained with reference to the drawings.

First, the image data shown in FIG. 1 is input using an image input device such as a camera and stored in a storage circuit. The data stored in the memory circuit is multi-gradation image data, which is binarized at an appropriate binarization level to obtain the image shown in FIG. Regarding binarized data, the sum of the data with the same X coordinate on the image is calculated for each X coordinate, and that is the X coordinate projection data, and the same calculation for Y is the y coordinate projection data. It is. FIG. 3 shows projection data of the X and Y coordinates of the binarized data of FIG. 2. If a screw hole exists in the image, a figure whose width and height are both equal to the diameter of the screw hole should appear on the projection data graph. Therefore, in the projection data,
Find a shape whose width and height are both equal to the diameter of the screw hole,
By calculating its center coordinate, the X coordinate of the screw hole,
The y coordinate is obtained.

Problems to be Solved by the Invention However, with the above configuration, it is not possible to distinguish between the round pattern of screw holes and other shapes of similar size, as described above, and the problem shown in FIG. If the background brightness of the screw hole is not uniform, or if there is not much difference in brightness between the hole and the background, the round pattern of the screw hole will not appear even after binarization, as shown in Figure 6. Unfortunately, as shown in FIG. 6, no figure corresponding to the screw hole appears in the projection data, so there is a problem that the position of the screw hole cannot be determined.

In view of the above problems, the present invention has a round pattern of screw holes,
It makes it possible to distinguish shapes from other shapes of similar size, and even when the background brightness of the screw hole is not uniform or there is not much difference in brightness between the screw hole and the background, it is possible to distinguish the shape from one circumference. The present invention provides a screw hole position recognition method that allows the center of the screw hole to be estimated using the area.

Means for Solving the Problems In order to solve the above-mentioned problems, the screw hole recognition method of the present invention extracts parts where the brightness changes from two-dimensional grayscale image data input using an image input device and generates line drawings. a thinning step of generating data; a screw hole existence determination step of determining whether a screw hole of a given diameter exists in the original image by calculating the curvature of the curve on the line drawing data; This process consists of a screw hole position calculation step of calculating the center position of the screw hole from the line drawing data.

Operation The present invention has the above-mentioned configuration, and uses a previous image operator to distinguish between a round pattern of screw holes and other shapes of similar size by calculating the curvature. The outline of a screw hole can now be detected even when the background brightness of the screw hole is not uniform, as shown in the figure, or when there is not much difference in brightness between the hole and the background.

EXAMPLE Hereinafter, an example of the present invention will be described with reference to the drawings.

FIG. 7 shows the extended Laplacian operator.

FIG. 8 shows a binarized Laplacian image. FIG. 9 shows a mask area for thinning lines from which noise has been removed from the image shown in FIG. The tenth is an explanatory diagram of ridge point extraction. Figure 11 shows the 8-neighbourhood of the vector. FIG. 12 is a condition score correspondence table for thinning. FIG. 13 shows line drawing data. FIG. 14 is a diagram showing the correspondence between direction vectors and bits.

FIG. 15 shows line drawing data directly expressed as a branch vector. In FIG. 16, line drawing data is expressed as a branch vector by setting a constraint so that a loop does not occur. FIG. 17 is an explanatory diagram of branch points. FIG. 18 is an explanatory diagram of branch cutting. FIG. 19 is a diagram showing the relationship between the radius of the circumference and the amount of change in angle. FIG. 20 is an explanatory diagram of curvature calculation of a line drawing. FIG. 21 is an explanatory diagram of arc center calculation. FIG. 22 is a diagram showing arcs extracted from line drawing data and the estimated position of the center of the arc.

Next, the screw hole position recognition method of this embodiment will be specifically explained.

[,1+lIl Lineization Step] First, the operation of the thinning step of generating line drawing data by extracting a portion where the brightness changes from two-dimensional grayscale image data input using an image input device will be described.

(-) Filtering As a premise for thinning the image data, the operator shown in FIG. 7 is used to perform filtering processing that emphasizes areas where brightness changes. This operator is equal to the difference between the normalized smoothing operator and the original image. Since the smoothing operator extracts low frequency components, this operator that takes the difference between the smoothing operator and the original image is a high frequency component extraction operator. This operator is 3x3
The Laplacian operator is expanded to a size of 11x11, and like the Laplacian operator, it has the characteristics of an edge enhancement filter, so it is called the extended Laplacian operator, and images converted using this operator. is called a Laplacian image.

(b) The method for extracting line elements from a thinned Laplacian image is described in the 10th
This is achieved by finding the ridge of the image when brightness is considered as height, as shown in the figure.

In this method, a binary image obtained by removing the nozzle component from a binarized Laplacian image using a fixed threshold value is used as a mask, and only points within the mask that are determined to be ridge points are set as line drawing points. . FIG. 8 shows the above-mentioned Laplacian image binarized using a fixed threshold value, but due to the characteristics of the operator, the image contains a very large number of noise components. Therefore, ridge points are extracted using a mask region obtained by removing connected regions whose area is smaller than a certain value from a binary image. FIG. 9 becomes the mask area. Extraction of ridge points is achieved by the following two steps.

(Step 1) When extracting ridge points from the Laplacian image, calculate the score of each point. Here, in the horizontal, vertical, and two diagonal directions, if the brightness of the central point is greater than the other two points, that line is called a central maximum. Since the distance of this central maximum line is different in the horizontal/vertical direction and in the diagonal direction, it is necessary to weight it. Therefore, the scores shown in FIG. 12 are added for each direction that satisfies the conditions, and the score is determined as the center point. The value of each score is selected to be approximately 1.4:1 in the horizontal or vertical direction and in the diagonal direction.

(Step 2) Pixels whose scores obtained in step 1 exceed 3 are defined as line drawing points, and pixels with scores less than 3 are defined as non-line drawing points. If the score is equal to 3, only the pixels that satisfy the following two conditions are set as line drawing points, and the other pixels are set as non-line drawing points.

(However, Xa=Xo) (2) T(XO)=1or T(X2)=1or
T(X4)=1or T(X6)=1 However, L T(Xi)=1:P(Xi))3=o:Other p(Xi) is the score corresponding to each point in the 8 neighborhood shown in Figure 11 .

(C) Vectorization The line drawing data obtained in the above manner (FIG. 13) contains a lot of noise. Therefore, we used the concept of branch vectors to remove unnecessary noise.

The branch vector will be explained below.

As shown in FIG. 14, in the branch vector, each point on the image has an 8-bit value, and each bit represents the direction component of the pixel connected to that pixel. However, the first
If the bits corresponding to all directions shown in Figure 4 are set, many loops will occur, so the following constraints were set when converting line drawing data into branch vector representation.

When connected components exist in the vertical (04, 40) and horizontal (01, 10) directions, line drawing points in diagonal directions on both sides thereof are not considered as connected components.

If this constraint does not exist, the line drawing data is converted as shown in FIG. 15, and if the constraint is added, it is converted as shown in FIG. 16. When line drawing data is converted into a branch vector using the above method, the value becomes 0 because the isolated point has no neighbors. That is, isolated points can be removed by converting to a branch vector.

[Screw hole presence determination process]

The screw hole presence determination step of determining whether a screw hole with a given diameter exists in the original image by calculating the radius of curvature of the curve on the line drawing data will be described below.

(-) All points in the branch/cut branch vector image are 1'-4 branch points, as shown in FIG. However, in order to determine circularity, the line must consist of only end points and midpoints. Therefore, in the "branch cutting" process, all three and four branch points are converted into one and two branch points. This method is realized by converting 3 branch points into 1 branch point and 2 branch points, and converting 4 branch points into 2 branch points and 2 1 branch points, that is, by performing cutting. When making this cut, we followed the standard: ``Only the line with a radius of curvature close to the target screw hole is left, and the others are cut.''

The branch cutting process will be explained using FIG. 18. FIG. 18(a) is an example of three branch points, including point p75.
and P Ao A1. P Bo B1. PC
.

It has branches in three directions of C1. Points Ao, Bo, C.

are the points where an arm of length L is extended from point P, and point A1.
B1. C1 is the point Ao, Bo, C respectively.

This is the point where an arm of length L is extended from. For these points, take the elevation angle θik as shown in the figure. Using these elevation angles, the angle change amount of the curve dθ1=θ, 1−θi when the line is traced by the length of L for all combinations of lines in two directions that can be branched from the point P.

Calculate dθ2=θko−π−θ10 dθ3″″θko−θ as 1. Now, when the radius of a given screw hole is r, as shown in FIG. 19, the amount of angular change when tracing the circumference by L is L/r.

Therefore, dθ1. dθ2. Sum of differences between dθ3 and L/r 1dθ1−L/r1+1dθ2 L/ r l +
Leave only the point where l dθ3L/r l becomes minimum and cut the rest.

0)) Arc extraction Since the branched and cut line drawing data is a collection of curves without branches, all you have to do is search for a line that has the same radius of curvature as the diameter of the screw hole. However, when converting grayscale image data to line data, even if the arc is beautiful in the original image, it may become a perfect arc when converted to line drawing data due to uneven brightness on the screen. It is normal not to. Therefore, in order to be able to calculate the curvature even if it is not a complete arc, the curve is divided into parts, the curvature is calculated for each part of the curve, and the average of these (hereinafter referred to as the average curvature) is approximated. curvature of the curve.

Let me explain the calculation method in more detail.

Let us represent a curve as a set of points on the curve as follows.

(p x (i), py (i)) + 1≦l≦L However, L is the length of the curve, and the subscripts of the points are the same order as on the curve. (p + q corresponds to the length of the line segment when dividing the curve and the distance between the points at which the angle change is measured, respectively.

It would be better to define it empirically to make it easier to recognize, but
As a guideline, both p and q are set so that they are large enough not to be affected by unevenness on the screen, and the number of sample points for calculating the partial curvature is not too small. The angle of the line connecting the i-th point of the curve and the (i + p )-th point is a
rg(i) (see Figure 20).

Next, from the difference in angle between the i-th point and the (i + q)-th point arg (i + q) - arg (i), the curvature curv (i) at this position is calculated as curv (i) =
(arg (i + q) - arg (i month/d (i, q) where d (i 1 (1) is calculated as the distance between p (i) and p (i + q).

The average curvature is the average of curv (i) over points on the curve (excluding the ends of the line).

Furthermore, when calculating the average, the variance is also calculated (hereinafter referred to as curvature variance).

The smaller the curvature variance, the more constant the curve of the line, so
Almost a perfect arc. If you choose one whose average curvature is close to the given screw hole diameter and whose curvature variance is small,
You can find the screw holes.

[Screw hole position calculation process]

The screw hole position calculation section that calculates the screw hole center position from line drawing data will be explained below.

Body) Arc center estimation As mentioned above, the extracted circular arc is not necessarily a perfect circular arc, so the center of the circular arc is estimated by the method described below.

The arc is divided, and the center of the partial arc is determined for each part of the divided arc using the direction of the line and the average curvature that has already been calculated, and the average is set as the center of the arc. The specific calculation formula is as follows: The estimated center of the arc (
estx(i), esty(i)) are as shown in Figure 21, estx(i)=px(i)+cos(
arg(i)+yr/2)/mc=px(i)-s
in(arg(i))/mCe5ty(i)=py(i
)+sin(arg(i)+yr/2)/mC=p
It is calculated as y(i)+cos(arg'(i))/mc.

The estimated center of the arc is the average over the points on the curve of estx(i) and esty(i). With the above method, even if the arc extracted from the line drawing is distorted,
Since averaging is performed at three points: curvature calculation and center position calculation, a center position close to reality can be obtained.

FIG. 22 shows the circular arcs extracted by the above method and the estimated center position (plus mark) for each circular arc.

Effects of the Invention As described above, the present invention includes a thinning step of generating line drawing data by extracting a portion where brightness changes from two-dimensional grayscale image data input using an image input device such as a television camera; A screw hole existence determination step in which the radius of curvature of the curve on the line drawing data is calculated to determine whether a screw hole with a given diameter exists in the original image; It consists of a screw hole position calculation section that calculates the position, and is the binary version of conventional image data.

Compared to the method of calculating the center of gravity from data projected in the y-direction, it is more reliable in identifying screw holes, and can be used when the background brightness of the screw hole is not uniform due to poor lighting, or when the brightness between the hole and the background is different. Even when the difference is small, the screw hole position can be estimated using a part of the circumference, which has a great practical effect.

[Brief explanation of drawings]

Figure 1 is the image data of the screw hole, Figure 2 is the binarized image data of Figure 1, and Figure 3 is the conversion of the binary data of Figure 2 to projection data in the X and 7 directions. Fig. 4 shows the image data of the screw hole when the background brightness of the screw hole is not uniform, Fig. 6 shows the binarized image data of Fig. 4, and Fig. 6 shows the image data of the screw hole when the background brightness of the screw hole is not uniform. From the binarized data in Figure 5, “ + 1
Figure 7 shows the extended Laplacian operator, Figure 8 shows the binarized Laplacian image, and Figure 9 shows the image in Figure 8 after being thinned by removing noise. Figure 10 is an explanatory diagram of ridge point extraction, Figure 11 is a diagram showing 8 neighborhoods of a vector, Figure 12 is a diagram showing condition score correspondence for thinning, and Figure 13 is a diagram showing the condition score correspondence for line thinning. Figure 14 is a diagram showing line drawing data. Figure 14 is a diagram showing the correspondence between direction vectors and bits. Figure 15 is a diagram showing how line drawing data is expressed as a branch vector. Figure 16 is a diagram that sets constraint conditions to prevent loops from occurring for line drawing data. Figure 17 is an explanatory diagram of branching points, Figure 18 is an explanatory diagram of branch cutting, Figure 19 is a diagram of the relationship between the radius of the circumference and the amount of angular change, and Figure 20 is a line drawing. FIG. 21 is an explanatory diagram of calculation of the arc center, and FIG. 22 is a diagram showing the arc extracted from line drawing data and the estimated position of the arc center. Name of agent: Patent attorney Toshio Nakao and 1 other person No. 1
Figure 2 Figure 4 Figure 6 Figure 7 Figure 8 Figure 9 VA 1011m Figure 1 Figure 12 Figure 13 Figure 14 Figure 15 Figure 16 Figure 17 Figure 18 ((L) b) Figure 20 "-211 Figure 21

Claims (1)

[Claims] A thinning process in which line drawing data is generated by extracting parts where brightness changes from two-dimensional grayscale image data input using an image input device, and the original image is created by calculating the curvature of the curve on the line drawing data. A screw hole position recognition method comprising: a screw hole existence determination step of determining whether a screw hole of a given diameter exists in the screw hole; and a screw hole position calculation step of calculating the position of the center of the screw hole from the line drawing data.