JP3048718B2 - Vertex detector - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a vertex detecting device which can be used for a component inspection device and a three-dimensional position device in the field of, for example, factory automation technology (FA). The present invention relates to a vertex detection device that can be used.

[0002]

2. Description of the Related Art FIG. 6 is a block diagram of a conventional vertex detecting device. In this conventional example (hereinafter referred to as Conventional Example 1), a grayscale image input unit 101, a neighborhood variance value calculating unit 102, and a vertex detecting unit 103 are provided. And in a region near a pixel that is a vertex of a figure in a grayscale image, the vertex of the figure is detected by utilizing the fact that the variance of the brightness (density value) between the pixels is large. I have.

That is, the gray-scale image input unit 101
For example, it is configured by a camera, a frame memory, or the like, and is configured to output a density value signal representing the density value of each pixel of the read grayscale image to the neighborhood variance value calculation unit 102.

A neighborhood variance value calculation unit 102 calculates a variance value of brightness between pixels in a region near a target pixel based on a tone value signal of each pixel input from the grayscale image input unit 101. It is configured as follows.

[0005] That is, the neighborhood variance value calculation unit 102 uses, for example, an operator called "Hannah's interest operator" to obtain directional variance values in four directions, up, down, left, and right for each pixel in the image, and further calculates It is configured to obtain (minimum value of ratio of directional dispersion) × (variance value).

The value obtained by the neighborhood variance value calculation unit 102 is output to a vertex detection unit 103, and the vertex detection unit 103 determines a pixel having a local maximum value in a neighborhood area from the value input from the neighborhood variance value calculation unit 102. It is configured to detect a vertex as a vertex pixel.

FIG. 7 is a block diagram of another conventional vertex detecting apparatus. This conventional example (hereinafter referred to as a second incoming example) includes a grayscale image input section 201, an edge detecting section 202, an edge extending section 203, and a vertex. The detection unit 204 is configured to detect the vertex of the figure in the grayscale image as an edge line of the target object, track the edge line, and detect a point where the edge line is bent as the vertex.

That is, the gray-scale image input unit 10 of the conventional example 1
A density value signal corresponding to the density value of each pixel of the grayscale image read by the grayscale image input unit 201 having the same configuration as that of 1 is output from the grayscale image input unit 201 to the edge detection unit 202,
The edge detection unit 202 is configured to detect an edge line of the target graphic from the grayscale image.

[0009] The edge detecting unit 202 is provided with, for example, “So
The value of the edge strength is calculated using a first-order differential operator called a “bel operator”, binarization processing is performed, and edge line thinning processing is performed to determine the axis of the edge line.

[0010] The edge extension unit 203 is
The extension processing is performed on the divided edge line in the binarization processing performed in step 2 at the divided end point, and the connected edge line is reproduced.

The vertex detecting section 204 is configured to track the edge line obtained by the edge extending section 203, search for a bending point, and detect the bending point as a vertex of a figure.

[0012]

The prior art 1 shown in FIG.
In the data after the variance value calculation, although the data of the vertex pixel has a local maximum value in the neighboring area, in order to detect another vertex adjacent to one vertex without error,
It is necessary to make the size of the local maximum value detection area for searching for one local maximum value such that adjacent vertices cannot exist in that area.

For this reason, even though no vertex exists in one maximum value detection area, the vertex detection unit 103 detects, for example, a pixel corresponding to a point on the edge line between the vertices in the maximum value detection area. Is determined to be a pixel having the maximum value of, and a point on the edge line may be erroneously detected as a vertex.

When a vertex existing in a dark area and a point on an edge line present in a bright area on an image are simultaneously present in one local maximum detection area, the data value of the point on the edge line is larger. This may make it difficult to distinguish between a vertex and a point on an edge line in the binarization process.

On the other hand, in the conventional example 2 shown in FIG. 7, information on a plurality of edge lines is mixed near a vertex where a plurality of edge lines are gathered, which makes it difficult to correctly detect and extend the edge lines. .

In view of the above points, the present invention provides a vertex detecting apparatus which does not erroneously detect a point on an edge line as a vertex and can detect a vertex in which a plurality of edge lines are gathered by a simple process. The purpose is to do.

[0017]

According to the present invention, a gray-scale image input section for reading a gray-scale image and outputting a density value signal for each pixel, a density value signal from the gray-scale image input section, and inputting the density value signal to the density image signal. A neighborhood variance value calculation unit that calculates a variance value of the density between pixels in a neighborhood area of each pixel of the grayscale image based on the grayscale image, a vertex candidate detection unit that detects a vertex candidate point based on the variance value, and a grayscale image input unit Input the density signal from
An edge detection unit that detects the intensity and direction of the edge line based on the density value signal; and a thinned edge direction that generates a thinned edge direction image from the edge line intensity and direction data obtained by the edge detection unit. When the cosine value of the difference between any two edge direction values in the neighborhood of the pixel on the thinned edge direction image with respect to the vertex candidate point is smaller than the cosine value of a preset value, A vertex detection unit that detects a vertex candidate point as a vertex.

[0018]

The grayscale image input device outputs a density value signal corresponding to each pixel of the read grayscale image to the neighborhood variance value calculator and the edge detector.

A vertex candidate point is detected from the density value signal output to the pixel neighborhood variance value calculation unit by, for example, the same procedure as that of the vertex detection of the first conventional example. That is, the neighborhood variance value calculation unit calculates a variance value of brightness between pixels in a region near the target pixel based on the gray value signal of each pixel input from the gray image input unit, and calculates the variance value. The vertex candidate detection unit detects a vertex candidate point using a pixel having a local maximum value in the neighboring area as a vertex candidate pixel. Therefore, the vertex candidate points include the vertices and the vertices and the points on the edge line that are erroneously detected.

The intensity and direction of the edge line are detected from the density signal output to the edge detecting section, and further, the edge direction thinned by the thinning edge direction image generating section is detected from the detected data. An image is generated.

In the thinned edge direction image,
Only pixels having an edge intensity value equal to or greater than a certain threshold have a corresponding edge direction value (-180 ° to 179 °), and the other pixels have meaningless values (for example, 1000). Therefore, it is guaranteed that the pixel having the edge direction value is always a pixel on the edge line, and in the thinned edge direction image, the pixel corresponding to the actual vertex is defined by the value θt set in advance in the vicinity thereof. It has two pixels with edge direction values where the cosine value of the difference is smaller than the cosine value.

The vertex detecting section determines whether each vertex candidate point has two pixels having an edge direction value in which the cosine value of the difference is smaller than the cosine value of the preset value θt in the neighboring area. It is determined whether or not the vertex candidate point is a vertex, and the vertex is detected. That is, data in an area of, for example, M × N pixels centered on a pixel on the thinned edge direction image corresponding to a certain vertex candidate point is examined, and data other than a value that is insignificant when present in the area is examined. That is, when there is only one edge direction value, or the cosine value of the difference between any two edge direction values θ and θ ′ among the edge direction values in the M × N pixel area is set in advance. If the value is equal to or greater than the cosine value of the value θt, it is determined that the candidate vertex point is not a vertex. In other cases, that is,
When the cosine value of the difference between any two edge direction values θ and θ ′ among the edge direction values in the area of M × N pixels centered on the vertex candidate point is smaller than the cosine value of the preset value θt. Then, the vertex candidate point is detected as a vertex.

[0023]

1 to 5 show an embodiment of the present invention.
A specific description will be given based on FIG.

FIG. 1 is a block diagram of a vertex detecting apparatus according to an embodiment of the present invention. The apparatus includes a grayscale image input section 1, an edge detecting section 2, a thinned edge direction image generating section 3,
The apparatus includes a neighborhood variance value calculation unit 4, a vertex candidate point detection unit 5, and a vertex detection unit 6.

The gray-scale image input unit 1 captures a target object having a gray scale as shown in FIG. 2A or 3A, for example, and converts the gray-scale image, that is, the gray-scale image into a large number of pixels. What is necessary is just to be configured to output the density value signals corresponding to the shading of each pixel in a predetermined order, for example, a video camera, a frame memory and the like.

The density value signal output from the grayscale image input unit 1 is simultaneously input to the edge detection unit 2 and the neighborhood variance value calculation unit 4.

The edge detecting section 2 for inputting the density value signal is configured to detect the edge line of the target figure based on the density signal value. For example, a primary differential operator called "Sobel operator" is used. When the density of a pixel at coordinates (x, y) is f (x, y), this operator

[0028]

(Equation 1)

And the edge direction value (−180
° to 179 °)

[0030]

(Equation 2)

It is configured to be determined according to the following.

From the edge intensities obtained by the edge detection unit 2, the detected edge line has a certain width, for example, as shown in FIG. You can get the image you have. A similar image can be obtained for the edge direction value. )

The data of the edge strength and the edge direction value obtained by the edge detection unit 2 are used for thinning edge direction image generation unit 3
, And the thinned edge direction image generation unit 3
A pixel having an edge strength value equal to or higher than a certain threshold value is given a 1 and other pixels having a value equal to or less than the threshold value, such as a background noise region, are given a 0, thereby performing a binarization process.

Thereafter, the thinning edge direction image generation unit 3
Performs thinning processing of an edge line detected with an unnecessary width for a portion having a value of 1, and extracts a core line of the edge line.

Thereafter, the thinned edge direction image generating section 3 gives the pixel having a value of 1 an edge direction value corresponding to the pixel obtained by the edge detecting section 2 and assigns a value of 0 to the pixel. An insignificant value (for example, 1000) is given to the possessed pixel, and a thinned edge direction image as shown in FIG. 3C is generated, for example.

A series of processes performed by the edge detecting section 2 and the thinned edge direction image generating section 3 provide an image in which pixels having edge direction values are guaranteed to be pixels on the edge line.

In the thinned edge direction image, at the stage of the binarization processing, pixels having a value of 0 are generated although they are edge lines, and as shown in FIG. In general, pixels having meaningful edge direction value data are interrupted.

The neighborhood variance value calculation unit 4 which inputs the density value signal of each pixel from the grayscale image input unit 1 outputs the target pixel to the grayscale image input from the grayscale image input unit 1 in the same manner as in the prior art. Is calculated so as to calculate a variance value of brightness between pixels in a neighborhood area of “Hannah”.
An operator called “interest operator” is used.

That is, the neighborhood variance value calculation unit 4 calculates directional variance values in four directions of up, down, left, and right for each pixel in the image.
Further, it is configured to obtain (minimum value of ratio of directional dispersion in four directions) × (dispersion value). As a result of this processing, a large value is obtained at a vertex, a point on an edge, or the like where a change in the density value between pixels in the vicinity area is large.
An image as shown in FIG.

The value obtained by the neighborhood variance value calculation unit 4 is output to the vertex candidate detection unit 5, and the vertex detection unit 5
A pixel having a local maximum value in a neighboring area from a value input from the above is set as a vertex candidate pixel, and a vertex candidate point is detected.

That is, the vertex detection unit 5 performs a binarization process on the data value obtained by the neighborhood variance value calculation unit 4 so that the data value of the noise portion in the background becomes 0, and further sets the target pixel to A process of determining whether or not the value is a local maximum value in a region of M × N pixels at the center and leaving the value if the local maximum value, and replacing the value with 0 if the local value is not the local maximum value (non-local maximum value suppression process)
And vertex candidate points are detected.

In this case, since the area of M × N pixels detects adjacent vertices without omission, the size is such that no other vertex exists in the M × N pixel area regardless of the center of any vertex. Need to be set. Also, where
It can be easily understood that the points detected as the vertex candidate points include the vertex and the vertex in the first conventional example and the point on the edge line that is erroneously detected, as shown in FIG.

The vertex detector 6 is configured to detect only vertices from the thinned edge direction image obtained by the thinned edge direction image generator 3 and the vertex candidate points obtained by the vertex candidate detector 5. You.

Here, each vertex candidate point is determined by whether or not it has two pixels having an edge direction value in which the cosine value of the difference is smaller than the cosine value of the preset value θt in the vicinity area for each vertex candidate point. Is a vertex, and the vertex is detected. If this judgment formula is expressed by a mathematical formula,

[0045]

Cos (θ−θ ′) ≧ cos (θt)

Is as follows.

That is, the data in the area of M × N pixels, for example, in which the other vertex candidate points are located around the pixel on the thinned edge direction image corresponding to a certain vertex candidate point, are examined. When there is only one edge direction value, or any two edge direction values θ and θ ′ among the edge direction values in the area of M × N pixels. If the cosine value of the difference is equal to or greater than the cosine value of the preset value θt, it is determined that the candidate vertex point is not a vertex. In other cases, that is,
When the cosine value of the difference between any two edge direction values θ and θ ′ among the edge direction values in the area of M × N pixels centered on the vertex candidate point is smaller than the cosine value of the preset value θt. Then, the vertex candidate point is detected as a vertex.

For example, the case where the vertex candidate point at the center of the area A shown in FIG. 2C is detected as the vertex will be described with reference to FIG.

As shown in FIG. 4A, an M × N surrounded by a bold line around a vertex candidate point (indicated by hatching)
= 7 × 7 pixels are set in the area A, and the thinned edge direction image in the area A includes a pixel to which the edge direction value of the pixel is given as shown in FIG. There are pixels indicated by blanks given no value (for example, 1000).

The pixels in the area A are examined, and a value other than a meaningless value such as 1000 is stored in an array prepared in advance. The values of the stored data are sequentially examined, and after confirming that there are two or more values other than meaningless values, a preset value θt (for example, 20 in this case) and an arbitrary value among them are set. Two values θ, θ ′ (eg, 114 and 8)
9) Compare and evaluate the magnitude relationship between the difference and the cosine value.

In this example, cos (114-89) = 0.9063 <cos (2
0) = 0.9397, so that the vertex candidate point in the area A is determined as a vertex and detected.

A case where it is determined that the candidate vertex at the center of the area B shown in FIG. 2C is not a vertex will be described with reference to FIG. 5 as follows.

As shown in FIG. 5 (a), M × N = encircled by a thick line around a vertex candidate point (indicated by hatching)
A region B of 7 × 7 pixels is set, and the thinned edge direction image in the region B has no meaning as the pixel given the edge direction value of the pixel as shown in FIG. There is a pixel indicated by a blank given a value (for example, 1000).

The pixels in the 7 × 7 area B surrounded by the bold line are examined, and a value other than a meaningless value such as 1000 is stored in an array prepared in advance. The stored data values are sequentially examined to confirm that there are two or more values other than meaningless values, and then a preset value θt (for example,
In this case, 20) and any two values θ,
The magnitude relationship between the difference between θ ′ (for example, 71 and 64) and the cosine value is compared and evaluated.

In this example, cos (71-64) = 0.9932> cos (20)
= 0.9397, and in this case, a similar relationship holds for all combinations of the stored data, so that it is determined that the vertex candidate point of the area A is not a vertex.

If the number of stored data is one, it is assumed that the candidate vertex point of the area is not a vertex.

In this way, the vertex detecting section 6 detects only the vertices from the thinned edge direction image obtained by the thinned edge direction image generating section 3 and the vertex candidate points obtained by the vertex candidate detecting section 5. By doing so, it is possible to prevent a point on the edge line from being erroneously detected as a vertex, and to detect a vertex in which a plurality of edge lines are gathered in a short time by a simple process.

[0058]

As described above, according to the present invention, a point on an edge line is not erroneously detected as a vertex, and a vertex where a plurality of edge lines are collected can be detected in a short time by a simple process. it can. Furthermore, in the detection of the vertices of a figure of a grayscale image in which erroneous detection is more likely to occur than a graphic without grayscale,
The vertices can be detected satisfactorily without the above problem.

Further, the present invention is not limited to a planar figure,
For example, the present invention can also be applied to a case where a three-dimensional shape of a target object composed of straight lines is recognized using a stereo method. The amount of processing data can be reduced, and the processing speed can be increased.

[Brief description of the drawings]

FIG. 1 is a block diagram of the present invention.

FIG. 2 is an explanatory diagram sequentially showing processing results up to detection of a vertex candidate point according to the present invention;

FIG. 3 is an explanatory diagram sequentially showing processing results up to the generation of a thinned edge direction image according to the present invention.

FIG. 4 is a diagram illustrating processing of a vertex detection unit according to the present invention.

FIG. 5 is a diagram illustrating processing of a vertex detection unit according to the present invention.

FIG. 6 is a diagram showing an example of a prior art of a method of detecting a figure vertex of a grayscale image.

FIG. 7 is a diagram illustrating an example of a conventional technique of a figure vertex detection method of a grayscale image.

[Explanation of symbols]

 Reference Signs List 1 Shading image input unit 2 Edge detection unit 3 Thinned edge direction image generation unit 4 Neighborhood variance value calculation unit 5 Vertex candidate point detection unit 6 Vertex detection unit

────────────────────────────────────────────────── ─── Continuation of the front page (56) References JP-A-2-18681 (JP, A) JP-A-3-40183 (JP, A) (58) Fields investigated (Int. Cl. ⁷ , DB name) G06T 7/60

Claims (1)

(57) [Claims] 1. A vertex detecting device for reading a grayscale image and detecting vertices of a figure composed of straight lines of the grayscale image, comprising: a grayscale image input unit for reading a grayscale image and outputting a density value signal of each pixel; A neighborhood variance value calculation unit that calculates a variance value of density between pixels in a neighborhood area of each pixel of the grayscale image based on a density value signal input from the grayscale image input unit,
A vertex candidate detection unit that detects a vertex candidate point based on the variance value, an edge detection unit that inputs a density value signal from a grayscale image input unit, and detects the intensity and direction of an edge line based on the density value signal, A thinned edge direction image generation unit that generates a thinned edge direction image from the edge line intensity and direction data obtained by the edge detection unit; and a pixel on the thinned edge direction image for the vertex candidate point. A vertex detector that detects the candidate vertex point as a vertex when a cosine value of a difference between any two edge direction values in the vicinity area is smaller than a cosine value of a preset value. Detection device.