JPH02176409A - Body recognizing apparatus - Google Patents

Abstract

PURPOSE:To make it possible to compute the position of a body accurately by recognizing one of the bodies based on the width and the height of a part where a slit image is shielded with the body, and obtaining the coordinates of the body based on the inclinations and the distances of two cameras when the reference image is cut. CONSTITUTION:Bodies to be recognized 12a-12d are mounted on a horizontal stage 10. An irregular reflecting surface 14 is arranged at the rear side, and a means 16 for projecting multi-slit light beams is arranged. A slit image 18a on a reference surface 30 on the irregular reflecting surface is shielded with the body. The width (d) and the height (h) of the shielded part are obtained with a camera 20. The image is processed in an image processor 22. The result is compared with the shape data of the bodies 12a-12d which are stored in the memory of a computer 24. Thus one of the bodies is recognized. Then, the coordinates of the center of the body are obtained by a specified computing expression based on the inclination of the recognized body with respect to the optical axis of the camera and the distance between both cameras by using the coordinates of two cameras where the image 18a of the slit image on the reference surface 30 is cut.

Description

[Detailed Description of the Invention] [Summary] Regarding an object recognition device used as a three-dimensional visual recognition unit for industrial use or intelligent robots, this is essential for the +A architectural lighting method using multi-slit light or multi-spot light. The objective is to be able to easily recognize objects without being bothered by the problem of corresponding points or to be subject to many restrictions on the surface material of the object to be recognized, and to accurately determine the three-dimensional position of the recognized object. , From the coordinates of the end points of the two cameras where the slit image or spot image of the reference plane on the image plane is interrupted by the object identified from among the multiple objects, the inclination of the identified object with respect to the reference plane and the distance between the two cameras are calculated. The configuration is such that the center coordinates of the object are finally determined based on the distance.

[Industrial application field]

The present invention relates to an object recognition device used as a three-dimensional visual recognition unit for industrial use or intelligent robots.

When an industrial robot or the like performs assembly work, it is necessary to accurately recognize the three-dimensional shape of the assembled parts, and then use a hand to grasp a predetermined part of the object based on the recognition results, and then perform the assembly work. be. Therefore, there is a need for an object recognition device that can recognize the three-dimensional shape of an object using a simple method and can be used for three-dimensional visual recognition by a robot.

[Conventional technology]

Well-known methods for measuring the three-dimensional shape (depth) of an object include the binocular stereo method and the structured illumination method. Many of these methods use triangulation, and the former applies geometry to a triangle formed by one point on the object and two points on both eyes (cameras). Here, the latter structured illumination method replaces one of the eyes with a light source. This structured illumination method is also called an active distance measurement method because it actively irradiates light onto the object. In contrast, the binocular stereo method is a passive measurement method. In addition to spot light and slit light, grind and regular patterns have been used as patterns of irradiated light. A photometric stereo method using multiple light sources has also been proposed.

However, to be precise, among these methods, spot light,
Many methods other than those using slit light are not based on the triangular &lI method.

Methods based on the principle of triangulation have been used for a long time because the geometric calculations are simple and the mechanism is relatively easy to implement.
This has been achieved to some extent. Historically, the binocular stereo method is the oldest because it has been used by living things, and it is superior both from the constraints of the applicable environment and from the ease of manufacturing the device. Nevertheless, there are the following circumstances behind the appearance of the spot method and the slit method.

First, structured illumination methods such as the spot method and the slit method are suitable for visual purposes in robotics.

Here, since the working range is narrow, irradiation light (projection light) with a high signal-to-noise ratio can be obtained. The second is a fundamental one. ap
As a matter of course, points belonging to the same point on the object must be selected as the three points for performing the geometric calculation of triangulation.

However, under common circumstances, it is difficult to find a point in the left eye image that corresponds to a point in the right eye image. This problem is usually called the corresponding point problem. In contrast,
In the method of irradiating one spot or one slit, the geometric conditions regarding the light source are known, so the corresponding point problem does not occur.

One problem with the structured illumination method that illuminates one spot or one slit as described above is that
Data processing requires time. ? J! To obtain 3D information about any object in the field, it is necessary to scan the entire field of view using spot light and slit light.0 Image 1
If only 1 slit and 1 slit are projected in the frame, it will take many frames to cover the entire field of view, and 1/3
It will take time just to input the number of frames of 0 seconds.

This naturally led to the idea of increasing the speed by emitting a large number of spot lights or slit lights at a time (but if you simply do this, the problem of corresponding points comes up again.Image The problem is in which direction the base point or base line above is caused by the beam irradiated. However, if the object is a smooth and diffusely reflecting surface, the spot light, slit light, etc. Since the slit lights can be counted sequentially, the correspondence in this case is good. There are many objects that are missing.

As described above, the multi-spot method and multi-slit method have the above-mentioned problem of corresponding points and the problem of limitations on the material of the object surface.

Therefore, in order to solve these problems, the present inventor filed a patent application No. 63-13133 for an object recognition method using multi-spot light or multi-slit light for objects whose two-dimensional shape as seen from a camera remains unchanged. Disclosed in issue.

That is, in this method, a diffusely reflecting surface such as a vertical plane wall is installed behind the object to be recognized, and when this diffusely reflecting surface is irradiated with multi-slit light or multi-spot light, the field of view of the camera set at an appropriate position is When there is no object, --like straight line groups or spont groups are reflected. When an object is placed in the field of view in this situation, the slit image or spot image is blocked by the object, and the object is recognized based on the position where the slit image or spot image is interrupted.

[Problem to be solved by the invention]

However, since it was necessary that the two-dimensional shape of the object as seen from the camera be constant, - for objects with a boat-like structure, the object must always be placed on a horizontal platform in the same attitude relative to the camera. Without it, this condition could not be met, making it extremely difficult to arrange objects.

Therefore, the present invention provides a multi-slit light or a multi-slit light 7)
Objects can be easily recognized without being bothered by the corresponding point problem that was essential in the structural lock method using light, and without being too limited by the material of the surface of the object to be recognized. The object is to provide an object recognition device that can accurately determine the position.

[Means to solve the problem]

The present invention focuses on the fact that more degrees of freedom are allowed when the object to be recognized is axially symmetrical.

As long as an axially symmetrical object is always placed upright on a horizontal table, its two-dimensional projected shape will be the same no matter where and in which direction it is placed, and its horizontal cross-sectional shape will be circular. Furthermore, the radius of the object cut by the reference horizontal plane is measured in advance (
Since this is known, the three-dimensional position of the object can be determined accurately.

From this point of view, in the present invention, as shown in FIG.
Recognition target object 12a placed upright on a horizontal table 10
12d, a diffused reflection surface 14 provided upright behind the objects 12a to 12d, and a slit image 1 on the diffused reflection surface 14.
8 or multi-slit light 1 for forming a spot image
8 or multi-spot light irradiation means and the diffused reflection surface 14
The reference plane 30 is a horizontal plane passing through the lens center 26C so that the intersection line with the object 12a overlaps with one of the slit images 18 or spot images formed on the diffuse reflection surface, and
Two cameras 20 and 21 installed in front of 12d,
The shape information of the objects 12a to 12d is stored, and the reference plane 3 is displayed on the image plane obtained by at least one of the cameras.
The image 18a of the slit image or spot image of 0 is the object 12.
The height h of the image of the object from the intersection line of the radiation d blocked by a to 12d and the reference surface and the diffuse reflection surface 14 is determined, and the ratio of the height h and the radiation d is compared with the shape information. By doing so, each of the objects 12a to 12d in uniform number is identified, and furthermore, the identified object causes the slit image of the reference plane 30 or the image 18a of the spot 7) image to be interrupted on the image plane. From the coordinates of the end point (ycyl), Dtt, 3'l) by the camera, calculate the inclination of the identified object with respect to the optical axis of the camera, and calculate the distance between the two cameras and the inclination. and calculation means for calculating the central coordinates (xo, yo) of the object in a R-like manner.

[For production]

According to the invention, the camera 20.21 allows the diffuse reflection surface 14 to be
The object 12a together with the upper slit image 18 or spot image
~Photograph 12d. Then, an image 1 of the slit image or spot image of the reference plane 30 is displayed on the image plane of the camera 20.21.
Find the width d where 8a is blocked by objects 12a-12d and the height h of the image of the object from the image 18a of the reference plane,
From this ratio of height h and width d, a plurality of objects 12a to 12d
Identify each one.

Next, the coordinates (yLl, y
The center coordinates (xo, yo) of the identified object are calculated from ()'t*, yez).

In this case, in the present invention, the inclination of the center line from both cameras to the center coordinates (xo, yo) of the identified object with respect to the reference plane is calculated as +, kx, and these inclinations are +, kg and the distance between both cameras. Finally, the center position of the object [1
(We are looking for X(+, yo).

In this way, with two cameras and the slope of the center line obtained by these cameras, the path between the end points j! 1yL
+-y□ or)'tg Ymx is removed from the calculation in the calculation means, and )'tt Y□ or yL
Measurement errors when ly□ is small can be removed.

〔Example〕

Hereinafter, the present invention will be described in detail based on embodiments shown in the drawings.

FIG. 1 is a block diagram of an embodiment of an object recognition device according to the present invention. In this embodiment, the calculation means is an image processor 22.
, a calculation 1924 , and a monitor 25 . An image taken by a camera 20 (camera 21 is not shown) is image-processed by an image processor 22 , and one image processor 22 displayed on the monitor 25 is a computer. 24
and is controlled by a computer 24.

FIG. 2 is a side view with a part of FIG. 1 omitted, in which a plurality of axially symmetric objects 12a, 12b, 12c
, 12d are placed upright at arbitrary positions. Further, it is assumed that all the objects 12a to 12d are placed so that they do not overlap when viewed from the camera 20. 14 is a vertical diffused reflection surface such as a wall surface, and this diffused reflection surface! A slit image 18 is formed on the slit image 18 by irradiating multi-slit light from a slit light generating device 16 provided at an angle of 15 to 20 degrees downward with respect to the horizontal plane.

Considering a horizontal plane passing through the lens center 26c of the lens 26 of the camera 20, this is taken as a reference plane 30, and the intersection line between this reference plane 30 and the diffused reflection surface 14 is an arbitrary one of the slit images 18 formed on the diffused reflection surface. The position of the camera 20 is set in front of the objects 12a to 12d to be recognized so as to overlap with the objects 12a to 12d. In addition, the height from the horizontal base IO to the reference plane 30 is assumed to be ■. 28 is an imaging plane placed parallel to the lens 26. Normally, the camera 20 is installed horizontally so that the optical axis of the lens 26 coincides with the reference plane 30;
It may be installed with a slight inclination upward or downward with respect to zero. However, even in this case, it is necessary to install the camera 20 so as to pass through the lens center 26c of the l&quasi-field surface 30 lens 26, and it goes without saying that the imaging plane 28 is not perpendicular to the reference plane 30.

An example of displaying an image taken by the camera 20 and displayed on the monitor 25 is shown in FIG. Although the camera 21 is not shown, unless otherwise specified below, it can be considered in exactly the same way as the camera 20, so its explanation will be omitted.

On the other hand, after measuring the shapes of the objects 12a to 12d in advance,
This shape information is stored in the memory (not shown) of the computer 24.No matter where the objects 12a to 12d are placed on the horizontal table 10, the reference plane 30 cuts the objects at the same location. Therefore, the portion where the slit image 18 on the reference plane 30 is blocked by the object corresponds to a horizontal section at the height H from the bottom of the object.

Measure the width of this part on the image plane as shown in Figure 3.
shall be. Since the slit image is not interrupted from the upper end of the object, the height h of the object image from the reference plane 30 can be found by searching for the interrupted portion from the top. Here, the reason why the height from the reference plane 30 is used instead of the height from the bottom is that in the lower part of the object, objects 12a to 12d
This is because it is difficult to distinguish between the horizontal table 10 and the horizontal table 10, and the slit image cannot be observed well.

The ratio of the above-mentioned height h to the radius d of the cut plane of the object by the reference plane is the same whether it is measured on the object or on the image plane, and is a unique value for each of the objects 12a to 12d. Therefore, by comparing the ratio h/d of the height h to the width d with the shape information of the objects 12a to 12d stored in the memory of the computer 24, one of the plurality of axially symmetrical objects 12a to 12d is
”:) One can be identified.

If this ratio h/d happens to be in the same direction for multiple objects, by changing the height H of the reference plane, the ratio h/dJI
: Can be changed.

Figure 4 is based on Figure 2! This is a schematic plan view cut along a plane, and with reference to this plan view, a formula for calculating the center position coordinates of the object 12a is obtained based on the object 12a reflected on the continuous image plane 28 of the camera 20 and the slit image. For the 0-position l measurement, the coordinates (yt) of the end point where the reference slit image 18a is interrupted by the object 12a.

y, ) (see FIG. 3), it is assumed that the object 12a has already been identified from among a plurality of objects as described above,
Since the object 12a is axially symmetrical, the object 1 based on the reference plane 30
The radius of the circle of the cut 2a is known by measuring it in advance. Therefore, the problem of calculating the coordinates of the center position of the object 12a can be solved by arbitrarily giving the coordinates (yL, ya) of the end points and solving the problem as shown in FIG. The center coordinates of a circle with radius r that is tangent to two straight lines as in (
This is equivalent to calculating -Xs, )'e).

The calculation results are as shown in the following equations (1) and (2).

(υ ・・−・・−−−−−・−(2) In this way, the position +M (yt
, ya), the center coordinates of the identified object can be easily determined.

Next, let's analyze the error in measuring the position of the object. The cause of the error is incomplete coordinate configuration.

Possible causes include lens distortion, digitization noise, and image blurring, but here we will consider digitization noise. Since image blur is subjected to binarization processing, it can be considered by incorporating it into digitization noise.

Differentiating equation (1), we get ΔX.

From Δ(ya −yL) O νm −LL, similarly from equation (2), −・・−・・−・−・・・・・・ (5) When evaluated using actual camera data, Only the first term of equations (4) and +5) is dominant. Therefore, the center position of the object can be determined with the same degree of error as the measurement error of the end points of the slit.

The method for determining the center coordinates of an object as described above has already been disclosed by the present inventor in Japanese Patent Application No. 13134/1983.

However, the center coordinates (io', yo) of the object obtained using equations (1) and (2) above are y
When m-'Yt is small, that is, when the recognized object is small or located far away, a measurement error will occur.

Therefore, in the present invention, if the ratio between xo and yo is calculated, then as shown in FIG.
Drop a perpendicular line to the optical axis of 1, let its legs be H+, Ht, and 2
Let the center points of the two lenses be O+ and Os.

If the ratios obtained by applying equation (6) to the two cameras 20 and 21 are respectively + and *, then these ratios are +(y
L, +y□) −・・・・・・・・−・−・−
・−(7)÷(31L + y true) −・・
・・・・・・−・−・−・・(6), and the above (1)
, (2), the term Ym-VL is removed, so the specificity of measurement errors is eliminated.

However, with only equation (6), the direction of the object from the center of the lens,
That is, only the inclination of the center line from the center of the lens to the center of the object with respect to the reference plane 30 is known.

Therefore, a camera 21 is provided in addition to the camera 20, and X e ', , :'7 o is calculated by triangulation.

+(yLm+y0) −・−・・−・−・・
(8), where FL++ 7□ is the coordinate of the imaging screen 28 of the camera 20, and yti+3'+1'' is the coordinate of the imaging screen 28 of the camera 21.

Then, the coordinates of the center point A of the object 12a as seen from one camera, for example the camera 20, are (-x.

y), and assuming that the two cameras 20 and 21 are parallel and the distance between them is D, HIO1=Xo +H1O@
-) 10, AHg -yo D, AHt -yo, so kl -Fil / Xo,, k□-(
y.

D)/xo, and solving this gives the following equation. By substituting the above equations (7) and (8), the coordinates (-Xo, -yo) of the center point A can be found.

〔Effect of the invention〕

As described above, in the present invention, the slit image or spot image of the reference plane on the image plane is interrupted by the object identified from among a plurality of objects, and the identified object is determined from the coordinates of the end point obtained by the two cameras. Since the center coordinates of the object were finally constructed as shown in this paper based on the inclination with respect to the reference plane and the distance between both cameras, measurement errors caused by the short length between the end points of each camera were eliminated. can do.

4. Brief explanation of the drawings □.

Fig. 1 is a configuration diagram of an embodiment of an object recognition device according to the invention, Fig. 2 is a partially omitted side view of Fig. 1, Fig. 3 is a schematic diagram showing one image plane on a monitor, Figure 4 is a schematic plan view taken along the reference plane of Figure 2, and Figure 5 is an explanatory diagram of binocular stereoscopic vision according to the present invention.

10--Horizontal platform, 12a to 12d-object, 14-vertical diffuse reflection surface, 16--multislit light generator, 18--slit image, 20--camera, 22--image processor, 24--computer , 25 .--Monitor, 28 .--Imaging surface, 30 .--Reference surface.

In the figures, the same reference numerals indicate the same or corresponding parts.

Claims (1)

[Claims] An object to be recognized (
12a to 12d), a diffused reflection surface (14) provided upright behind the object (12a to 12d), and a multi-slit for forming a slit image (18) or a spot image on the diffused reflection surface (14). The lens is arranged so that the line of intersection between the light (18) or multi-spot light irradiation means (16) and the diffused reflection surface (14) overlaps with one of the slit images (18) or spot images formed on the diffused reflection surface. The horizontal plane passing through the center (26c) is the reference plane (30
) and installed in front of the object (12a to 12d);
a to 12d) is stored, and the reference plane (30) is stored on the image plane obtained by at least one of the cameras.
The image (18a) of the slit image or spot image of the object (
12a to 12d) and the height of the image of the object (
h), and by comparing the ratio of the height (h) and width (d) with the shape information, a plurality of objects (12a to 12d
), and further determine the coordinates (y_L_1, y_R_1), (y_L_2, y_R_
2), find the tilt (k_1, k_2) of the identified object with respect to the optical axis of the camera, and calculate the tilt (k_1, k_2).
and a calculation means for finally determining the center coordinates (x_o, y_o) of the object based on the distance (D) between both cameras.