JPH09167234A - Three-dimensional recognizing method and its device using card information - Google Patents

Abstract

PROBLEM TO BE SOLVED: To more correctly control a robot at higher speed while restricting its cost at the time of proceeding the automization of a production line by comparing CAD graphic characteristic quantity with camera graphic characteristic quantity and discriminating the kind and figure, etc., of a specified object caught by a camera. SOLUTION: In the three-dimensional recognizing device, a camera 3 is attached to the robot handle 2 of a robot 1, one of objects 6-8 is discriminated by a computer 5 while catching a specified object 4 by the camera from above and, at the same time, the robot handle 2 is operated in accordance with data of the clamping position of the object which is discriminated by the computer 5. At this time, CAD graphic characteristic quantity introduced by projecting the respective objects on the orthogonal surface in a prescribed direction from respective kinds of CAD information of the objects 6-8 is calculated and camera graphic characteristic quantity in a two-dimensional picture obtained from the specified object which is caught by the camera in a prescribed direction is calculated. Then, both kinds of characteristic quantity are compared and the kind and the figure, etc., of the specified object caught by the camera 3 is discriminated.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and apparatus for three-dimensionally recognizing a target object by a computer that controls a robot for assembling an automated production line.

[0002]

2. Description of the Related Art In order to reduce production cost and enhance safety, a manufacturing or assembling production line is being greatly automated by using a robot. So-called CAM is CAD information created in advance, for example, CA.
It is widely seen in machining processes such as grinding and cutting a predetermined material while controlling the robot based on D data. In the assembly process for assembling such a processed product, the computer controlling the robot is used to set the object to be processed. Automation has not yet been fully achieved because it has to be dimensionally recognized. However, as a result of the rapid development of computer technology, an object can be captured by a camera, and a computer can recognize the object three-dimensionally from the image and control the robot.

In the conventional method of three-dimensionally recognizing an object, the direction in which the object is captured is generally set in advance, and the image data of the object viewed from that direction is stored in the computer. The procedure of determining the type of the target object by matching it with the graphic data of the comparison object in the above-mentioned direction is used. I needed a camera to shoot it.

[0004]

In the conventional three-dimensional recognition method, a plurality of cameras for obtaining an image are prepared or one camera is directed toward an object in order to capture the object in multiple directions. This required a device to change the production line, which was a factor in raising the cost of automation of the production line. In addition, in a production line aiming at high-mix low-volume production, there are many objects that have similar appearances, and such conventional methods may cause misrecognition.

Therefore, it has been decided to develop a new three-dimensional recognition method with the goal of controlling the robot more accurately and at high speed while controlling the cost of the automation of the production line.

[0006]

What has been developed as a result of the study is to derive each two-dimensional figure obtained by projecting each of these objects on a plane orthogonal to a predetermined direction from the CAD information of each of the plurality of objects. The CAD graphic feature amount such as the center of gravity of the figure, the contour line, the direction of the normal to the contour line at each point arranged on the contour line, the distance from the center of gravity to each point arranged on the contour line, and the like are calculated by the camera in a predetermined direction. A camera figure characteristic amount similar to the CAD figure characteristic amount in the two-dimensional image obtained from a specific target object (hereinafter referred to as a specific object) captured from the CAD figure characteristic amount and the camera figure characteristic amount are calculated. By comparing, C for determining the type, posture, etc. of the specific object captured by the camera C
This is a three-dimensional recognition method using AD information.

The CAD information is preferably three-dimensional CAD data, but in some cases, the three-dimensional CAD data may be first generated from the two-dimensional CAD data forming the three-view and the three-dimensional CAD data may be used. The file format of the CAD data is not specified, but the DXF format is preferable in consideration of versatility.

First, a three-dimensional figure of an object made of CAD data is converted into a two-dimensional figure viewed from an arbitrary viewpoint. Generally, since a three-dimensional object has a stable posture under gravity (hereinafter referred to as a stable posture), image data of a specific object captured by a fixed camera is limited to the specific object in this stable posture. The angle of horizontal rotation may be treated as an offset during calculation as a phase difference. Obtained by converting 3D figures 2
The three-dimensional figure may be created only for the type of the image data captured by the camera. The 3D figure viewed from an arbitrary viewpoint is the vertex (x, y, z) that is translated so that the center of the 3D figure overlaps the CAD data of the object corresponding to the specific object so as to overlap the origin of the 3D space. ) Is Rx rotation around the X axis (Equation 1), Ry rotation around the Y axis (Equation 2), Z
It can be obtained by moving to coordinates (X, Y, Z) by Rz rotation (Equation 3) about the axis.

[0009]

[Equation 1]

[0010]

(Equation 2)

[0011]

(Equation 3)

In order to convert the thus rotated three-dimensional figure into a two-dimensional figure, the distance between the virtual screen on which the three-dimensional figure is projected and the viewpoint E is D, and the three-dimensional coordinate (X, Y, Z).
Is converted into two-dimensional coordinates (MX, MY) (Equations 4 and 5).

[0013]

(Equation 4)

[0014]

(Equation 5)

According to the present invention, the two-dimensional figure thus obtained is filled and binarized to extract the feature quantity. First, from the binarized two-dimensional figure, the center of gravity (Gx, Gy) (Equations 6 and 7)
And a point (DX) that represents a contour line by contour line tracking processing (“Image processing”: 1983, P187 to P188, Hiroshi Ozaki, Keiji Taniguchi, Hideo Ogawa)
j, DYj) (j = 1, 2, ..., Nc) is obtained. In addition to the center of gravity and the contour line, the feature amount includes the distance from the center of gravity of the object to the contour line (CLj, equation 8) and the normal direction of the contour line (Vj, equation 9).

[0016]

(Equation 6)

[0017]

(Equation 7)

[0018]

(Equation 8)

[0019]

(Equation 9)

The specific object taken in from the camera is binarized,
Distance ILn (n = 1, 2, ..., Dr) from the center of gravity of the image data of the binarized specific object, and the distance from the center of gravity of the object derived from the CAD data to the contour line. C
Compare with Lj to determine identity. In this case, since ILn in the image data and CLj from the CAD data do not always match the scale and the number of data, the scale constant is multiplied by CLj and the number of data is calculated. ILn and IL in ILn
A linear interpolation is performed between n + 1 and n + 1 to convert it into a continuous broken line, and this continuous broken line is subdivided by Nc (the number of CLj) to achieve matching as MLj.

[0021]

[Equation 10]

Image data taken from the camera and CA
A two-dimensional figure in which a three-dimensional figure of D data is viewed from an arbitrary viewpoint is the maximum value of CLj {CLj1, CLj2, ..., CLjm} and M
The degree of matching Pmax (Equation 11) obtained from the difference between the distance CLj between each point and the center of gravity when matching the maximum value max {MLj} (1 ≦ j ≦ Nc) of Lj and MLj, and each of CLj Minimum value (CLk1, CL
k2, ..., CLkt} and the same degree of coincidence Pmin (Equation 11) when the minimum value min {MLj} (1 ≦ j ≦ Nc) of MLj are matched are the most identical. Can be determined to be high, and in this case, it is possible to identify which CAD data object the particular object captured by the camera is, and from which direction the object is viewed.

[0023]

[Equation 11]

According to the method of recognizing a specific object according to the present invention, a virtual space is imagined by CAD information in a computer, and image data captured by a camera is a two-dimensional view of which object is viewed from which direction in this virtual space. It is determined whether or not it matches the image, and a considerable amount of data is required for one piece of CAD information. Therefore, in order to identify a plurality of objects, the amount of data is increased by the number of objects, and it is currently difficult to execute the calculation for recognition in real time. Therefore, if the feature amount from the CAD information is calculated in advance and the database is constructed, the processing speed can be increased. The scale of the image data of the camera is changed so that the number of data is matched with the assumption that such a database is used and the amount of data that needs to be calculated is reduced. For example, if a faster computer appears in the future, two-dimensional figures from CAD data can be calculated and compared each time.

With respect to the specific object whose type and direction are thus recognized, (1) the distance CLj from the center of gravity of the two-dimensional figure to each point arranged on the contour line among the CAD figure feature values
Of the maximum value CLj1 (DXj1, DYj1) and the point (DXj2, DYj2) of the next maximum value CLj2 excluding this maximum value are connected in the normal direction H (Equation 12), (2) A point (DXk1, DYk1) indicated by the minimum value CLk1 of the distance CLj from the center of gravity of the two-dimensional figure to each point arranged on the contour line, and a point (DXk2, DYk2) indicated by the next minimum value CLk2 excluding this minimum value. The direction H (Equation 13) connecting (3), (3) The method of the longest of the contour line parts in which the points having almost no change in the normal direction Vj to the contour line at each point arranged on the contour line of the two-dimensional figure are arranged Line direction V
a, or (4) the maximum value of the frequency distribution in the normal direction Vj with respect to the contour line at each point arranged on the contour line of the two-dimensional figure {Vb1, Vb
2, ..., Vbu}, and one of the sets of the normal direction, which is deviated by about π from the maximum value, is set as Vbw, which is the gripping direction of the specific object recognized by each method. Further, when a plurality of gripping directions are parallel to a direction in which a plurality of the gripping directions coincide with each other and pass through the center of gravity, the gripping position is determined.
A direction parallel to the gripping direction of (4) and passing through the center of gravity is determined as a gripping position, and the gripping direction and position of the object by the robot hand are recognized.

[0026]

(Equation 12)

[0027]

(Equation 13)

The recognition of a specific object is generally a premise for moving some device or machine (in some cases, recognition alone may be sufficient).
For example, the gripping direction and the gripping position of the recognized specific object by the robot hand are determined using the graphic feature amount. The gripping direction is used as data for recognizing the direction of a specific object, and can also be used for controlling a turntable or the like that horizontally rotates the specific object in a certain direction. Since these gripping directions and positions can be derived from the CAD figure feature values,
It is preferable to calculate in advance and build a database for the same reason as above. As for the gripping direction, a more stable grip by a robot hand or the like can be realized by comparing a plurality of calculated results and taking a majority decision.
If the results do not match, the normal direction Vj of the contour line calculated as the CAD figure feature amount is determined as the grip position as the most appropriate one for gripping by the robot hand or the like.

Most of the robot hands that are currently in use are those that cause a pair of fingers to approach and separate face-to-face, and in order to grab a specific object with such a robot hand,
Among the contours of this specific object, it is preferable that two opposite sides are parallel and long. However, the recognition of the gripping position by calculation is
Since it is not always possible to calculate what is suitable for gripping a robot hand, first, as gripping directions, a plurality of directions that meet the above conditions (two opposing sides are parallel and long) are calculated from the CAD figure feature amount, and which gripping direction is calculated. The specific gripping position is determined by determining whether the direction is the most preferable.

The method (1) is the point (DXj1, DYj1) indicated by the maximum value CLj1 of the distance CLj from the center of gravity of the two-dimensional figure to each point arranged on the contour line and the next maximum value excluding this maximum value. It is judged that the point (DXj2, DYj2) indicated by CLj2 is a point that is largely convexly curved on the contour line, and the point (D
Xj1, DYj1) and (DXj2, DYj2) are connected, the normal direction is defined as the gripping direction of the robot hand, assuming that there is a side having the least change in the distance CLj to the center of gravity, that is, a smooth side. decide. For example, in the case of trapezoid, 4
As a result of selecting two adjacent points among the two vertices, the normal direction of the side having these two points at both ends is determined as the gripping direction H. The method (2) uses the point (DXk
1, DYk1) and (DXk2, DYk2) are connected, and it is assumed that there is a side part where the distance CLj to the center of gravity changes the least, and the direction connecting the points is determined as the gripping direction H of the robot hand. To do.

In the method (3), the contour line portion in which points having almost no change in the normal direction Vj to the contour line at each point arranged on the contour line of the two-dimensional figure are arranged in a substantially straight line or a smooth curve. Since the contour line is the side of the object orthogonal to the gripping direction, the normal direction Va of the longest of the sides is determined as the gripping direction. In the method (4), since the normal direction Vj at the point on the side will have the largest number, the frequency distribution in the normal direction Vj with respect to the contour line at each point arranged on the contour line of the two-dimensional figure is calculated. The maximum value is examined, and it is determined that one set of the normal direction Vbw deviated from the maximum value by about π exists on two sides facing each other, and one of the normal directions Vbw is determined as the gripping direction.

If the object has a simple shape, the gripping directions calculated by the above methods should match. If at least a plurality of gripping directions match, the direction parallel to the matched gripping direction and passing through the center of gravity is The grip position can be recognized as the one that can be gripped most stably by the robot hand or the like. If all do not match, the gripping direction obtained by connecting at least two points that are facing each other by the method of (4) will be more stable and can be gripped by a robot hand, etc. The direction passing through is determined to be the gripping position.

The above three-dimensional recognition method can be configured as a three-dimensional recognition device including a camera and a computer. For example, CAD information handling means for the computer to create and store each CAD information of a plurality of objects, C
From the AD information, each two-dimensional figure obtained by projecting each of the objects on a plane orthogonal to a predetermined direction is derived, the center of gravity of the two-dimensional figure, the contour line, and the direction normal to the contour line at each point arranged on the contour line, CAD figure feature amount extraction means for calculating a CAD figure feature amount such as a distance from the center of gravity to each point arranged on the contour line, image data acquisition means for obtaining a two-dimensional image from a specific object captured by the camera from a predetermined direction, A camera figure feature amount extraction means for calculating a camera figure feature amount similar to the CAD figure feature amount in a two-dimensional image, a CAD figure feature amount and a camera figure feature amount are compared, and a specific object captured by the camera And a discriminating means for discriminating the type, the posture, and the like.

Further, in the computer, the C of the target object matched with the type, posture, etc. of the specific target object captured by the camera is displayed.
If a gripping determination means for calculating the gripping direction and gripping position of the object from the CAD figure feature amount of the two-dimensional figure derived from the AD information is added, for example, three-dimensional until the robot grasps the specific object captured by the camera. A recognition device can be configured.

Although it depends on the shape, type, and number of objects to be discriminated, a personal computer may be used in addition to an office computer, a minicomputer, etc. because the performance of the CPU has been remarkably improved these days. Further, one computer may be assigned to each means for distributed processing, or one computer may perform centralized processing because there is some time-series flow in executing the means. When using a plurality of computers, it is preferable to connect each computer via a network.

[0036]

BEST MODE FOR CARRYING OUT THE INVENTION A configuration of a three-dimensional recognition apparatus using the three-dimensional recognition method of the present invention will be described below with reference to the drawings. In FIG. 1, a camera 3 is attached to a robot hand 2 of a robot 1, and a specific object 4 is captured by the camera 3 from above, and a target object 6 is detected by a computer 5.
9 is a block diagram showing a device configuration of a three-dimensional recognition device that determines whether the robot 7 is 8 or 7, and at the same time operates the robot hand 2 according to the data of the gripping position of the object 7 determined by the computer 5. The computer 5 is the CA of the objects 6, 7 and 8.
It is also a means for creating and storing D data, and the computer 5 constructs the database 9 by extracting CAD figure feature amounts from CAD data of three-dimensional figures.

When the grip positions of the objects 6, 7 and 8 are calculated and attached to the database 9, the robot hand 2 can be operated immediately after the specific object 4 is discriminated. Further, if the database 9 can be constructed in advance by taking a sufficient time, the CAD figure feature amount of each of the objects 6, 7, 8 can be easily distinguished from the easily distinguishable object 8 and the object 6 or 7.
(The fall state and the standing state can be considered) are divided into groups, and in the actual discrimination, when it is recognized as an object 6 or 7 of a group that is difficult to discriminate, the robot hand 2 is moved,
The accuracy of recognition may be increased by allowing the camera 3 to catch the specific object 4 from diagonally above. Regarding the horizontal rotation of the specific object 4, in the matching of the CAD data feature amount and the camera image feature amount, the phase difference in the direction with respect to the center of gravity of the minimum value and the minimum value or the maximum value and the maximum value can be treated as an offset value. Good.

The CAD figure feature amount obtained from the CAD data is the center of gravity, the contour line, and the distance from the center of gravity to the contour line. From the feature amount of the distance from the center of gravity to the contour line, it is possible to obtain the maximum value and the minimum value, and the frequency characteristic obtained by FFT of the data of this distance. This FFT is extremely excellent as a feature amount in that the features of the object can be captured without being confused by the appearance such as the size and shape of the object. Since the means for calculating the CAD figure feature amount and the means for calculating the camera figure feature amount from the image data are the same processing, both means should be realized by the same device, in this example, the computer 5. You can

If the CAD figure feature amount and the camera figure feature amount are compared in the order of the flowchart shown in FIG. 2, the specific object can be efficiently recognized. In step 1, CA
Looking at the degree of coincidence between the FFT results of the D graphic feature amount and the camera graphic feature amount, if they are clearly different, the target of the CAD graphic feature amount extracted as a comparison in the computer 5 and the specific object 4 captured by the camera 3 Items 6 and 8 are excluded because it is not necessary to compare them. FFT agreement is CA
It is determined whether the frequency F1 indicated by the maximum value of the FFT value FCLq calculated from CLj of the D data and the frequency F2 indicated by the maximum value of the FFT value FILq of the ILn of the image data are substantially equal.

In step 2, the ratio between the CAD figure feature amount and the camera figure feature amount is made the same, and in step 3, the number of data of the CAD figure feature amount is subdivided to match the number of data of the camera figure feature amount, Put them in a state where they can be compared. When the distance from the center of gravity to the contour line is selected for each feature quantity, the ratio can be obtained by comparing and calculating the largest of the feature quantities, and the matching of the number of data is obtained by linearly interpolating the CAD figure feature quantity. Then, it can be realized by converting into continuous polygonal lines and subdividing by the number of data of the camera figure feature amount.

In step 4, the CAD figure feature amount and the camera figure feature amount are compared. In this comparison, when the distance from the center of gravity to the contour line is selected as each feature amount, first, the minimum value of the camera figure feature amount is detected and the minimum value of the CAD figure feature amount is made to match this minimum value one by one. Agreement Pmin
(Equation 11) is calculated, then the maximum value of the camera figure feature amount is detected, and the matching degree Pmax (Equation 11) is calculated while matching the maximum value of the CAD figure feature amount with this maximum value one by one. The lowest value of the matching degrees Pmin and Pmax is set as the matching degree P between the specific object 4 captured by the camera 3 and the target object 7 being compared in the computer 5.

The above steps 1 to 4 are carried out for each of the two-dimensional figures possessed by each of the objects 6, 7, 8 and belong to a group in which the CAD figure feature quantity of the object 7 to be compared is difficult to discriminate. For example, 2
The accuracy of recognition of the specific object 4 is improved by comparing with the CAD figure feature amount of the three-dimensional figure.

If it is possible to recognize which object 6, 7, 8 from which direction the specific object 4 captured by the camera 3 is looking, then grasping the object 7 associated with the CAD figure feature amount of the recognized object 7. Using the position data, the robot hand 2 is operated to grip the specific object 4. Since the data of the gripping position is calculated assuming that the specific object 4 follows the reference coordinates, the horizontal rotation offset value obtained above is added or subtracted to horizontally rotate the robot hand 4 in accordance with the gripping direction. I shall. Since the specific object recognized in the standing state has already been grasped three-dimensionally, the robot hand is first lightly hit to cause the robot to fall, and the gripping position is determined and the robot hand is grasped in the falling state. Is preferred.

[0044]

EXAMPLE An example based on the above embodiment will be described. The computer 5 is a Pentium (registered trademark) 60 Hz personal computer, the robot 1 is a general industrial robot of 6-axis articulated type, and the computer 5 and the robot 1 are connected by RS-232C. The camera 3 is attached to the robot hand 2 of the robot 1 so that the image data can be taken into the computer 5 by a dedicated image board (see FIG. 1).
reference). The CAD data of the object 4 shown in FIG. 3 is DX.
The file was created in the computer 5 in the F file format, and the coordinate file of each vertex (Table 1) and the surface definition file (Table 2) were created from this CAD data and saved.

[0045]

[Table 1]

[0046]

[Table 2]

From the CAD data, three-dimensional figures (perspective views 1, 2, 3) obtained by rotating the objects 6, 7, 8 in the three-dimensional space as shown in FIG. 4, FIG. 5 or FIG. Then, by painting the inside with the outermost contour line as a boundary, for example, in FIG.
Obtain a dimensional surface model (Fig. 7). The CAD figure feature amount is calculated for this two-dimensional surface model. As the CAD figure feature amount of this embodiment, (1) centroid,
(2) contour line, (3) distance from the center of gravity to the contour line, (4) maximum and minimum values in the distance from the center of gravity to the contour line, and (5) FFT data of the distance from the center of gravity to the contour line are selected. . FIG. 8 is a distribution diagram of the distance (CLj) from the center of gravity to the contour line (3), and FIG. 9 is a frequency characteristic diagram of FFT data of the distance from the center of gravity to the contour line (5).

The CAD figure feature amount is collected in the database 9 for the CAD data of each object. Table 3 shows
Figure 4 (corresponding to the leftmost number 1), Figure 5 (corresponding to the leftmost number 2)
7 is a table listing a database 9 (see FIG. 1) of CAD figure feature amounts of a two-dimensional surface model obtained from FIG. 6 (corresponding to the leftmost number 3). The name in the second column from the right is the group name, and none is the specific object 4 just seen from above.
Group that can be matched with the image data of
And type_b are groups that must be matched with the image data of the specific object 4 viewed from above and obliquely above, respectively (a and b are groupings according to the appearance of the figure, and are further increased to c, d, ...). These data are classified into groups and stored.

[0049]

[Table 3]

After the specific object 4 is placed below the robot hand 2, the camera 3 captures the specific object, a two-dimensional surface model is obtained by binarizing this image data, and the camera image characteristics are obtained based on the above procedure. Calculate the amount. And
The camera image feature amount is matched with the CAD figure feature amount database 9. In this embodiment, F
It will be explained from the stage after the rough agreement by the FT data is obtained. First, the CAD data feature amount (CLj) is multiplied by the proportional constant K to match the width of the distribution of the camera image feature amount (ILn). Next, as you can see in the contour map of Figure 10,
The number of camera image features is matched with the number Nc of CAD data features by primary interpolation. In this state, the specific object 4
If the target object 7 and the target object 7 are the same, the superposition of the distribution charts is substantially the same (see FIG. 11), and if they are different, they are displaced (FIG. 12).
reference). The above comparison is carried out in the computer 5 as the calculation of the degree of coincidence P.

The object 7 having the CAD data feature amount with the lowest coincidence P is the specific object 4 captured by the camera 3, and CA
The stable posture of the specific object 4 is determined from the type of the D data, and the minimum and minimum values of the CAD data feature amount and the camera image feature amount or the maximum value and the maximum value of the maximum value are obtained. Is recognized as a phase difference between the target object 7 and the specific object 4, the robot hand 2 horizontally rotates the gripping direction by this phase difference, and based on the grip position calculated in advance, the specific object 4 is detected. Grab. For the operation of the robot hand 2, the existing operation means using the computer 5 was used.

To determine which object the specific object captured by the camera is, the calculation and comparison may be performed according to the above procedure, but the actual orientation of the specific object is not always easy to recognize. Not necessarily. In some cases, in order to recognize the parts sent by the belt conveyor, you can move them,
It is convenient to defeat certain objects that are upright. In the present embodiment, the direction of this specific object is also identified when the computer recognizes the specific object by the above procedure, and for example,
The upright specific object is added with a routine that makes it easier for the robot hand to hold, such as when it is tilted. This is also an advantage because the specific object can be recognized three-dimensionally by using the three-dimensional CAD data, so that the specific object can be recognized in the height direction.

[0053]

As described above, this method uses CAD information in the assembly process as well as the conventional machining process using CAD information, and realizes consistent data management from design to processing and assembly. The feature is that you can do it. Moreover,
Since the CAD information used for recognizing the specific object is three-dimensional data, there is an advantage that the specific object can be recognized three-dimensionally.
This enables more flexible operation of the robot hand and the like. This means increasing the intelligence of the conventional industrial robot.

Since a robot used in a factory production line or the like is required to have a considerably high speed, at present, the CAD figure feature amount and the gripping position data are collectively stored in a database from the CAD information and specified. At the stage of object recognition, it is convenient to compare this database with the camera figure feature amount of the specific object. However, if the processing capacity of the computer is further improved in the future, the CAD information can be transferred to the computer in real time, and the specific object can be immediately recognized, which enables wider use.

At the present stage, the present invention greatly contributes to the unmanned operation of the factory by teleoperation. In recent years, a computer network has been widely taken up, and CAD information is often exchanged on a personal computer network. With the method and apparatus of the present invention, if the CAD information of an object is sent from a remote location to a computer that operates the robot and the robot can recognize and grip a specific object three-dimensionally based on the data, the Unmanned factories can be realized through operations.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a three-dimensional recognition device that operates a robot hand.

FIG. 2 is a flowchart showing an order of comparing respective feature amounts.

FIG. 3 is a perspective view of an object used in an example.

4 is a perspective view 1 of a three-dimensional figure obtained by rotating the CAD figure of the object of FIG. 3;

5 is a perspective view 2 of a three-dimensional figure obtained by rotating the CAD figure of the object shown in FIG. 3;

6 is a perspective view 3 of a three-dimensional figure obtained by rotating the CAD figure of the object shown in FIG. 3;

FIG. 7 is a surface model obtained from the three-dimensional figure of FIG.

FIG. 8 is a distribution diagram of a distance from a center of gravity, which is a CAD graphic feature amount, to a contour line.

FIG. 9 is a frequency characteristic diagram of FFT data of a distance from a center of gravity, which is a CAD graphic feature amount, to a contour line.

FIG. 10 is a distribution diagram of contour lines in a state where the CAD graphic feature amount and the camera image feature amount are matched in size and number.

FIG. 11 is a distribution diagram of contour lines in a state where the contour lines substantially match each other in the state of FIG.

FIG. 12 is a distribution diagram of contour lines in which each contour line is different and shifted in the state of FIG.

[Explanation of symbols]

 1 Robot 2 Robot Hand 3 Camera 4 Specific Object 5 Computer 6 Object 7 Object 8 Object 9 Database

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Yoshiomi Sozawa 1-161, Nise, Okayama-shi, Okayama (72) Inventor Takao Uchida 4-4, 419, Miyamae, Kurashiki-shi, Okayama

Claims (8)

[Claims] 1. A center of gravity, a contour line, and an array on the contour line of the two-dimensional figure are derived by deriving two-dimensional figures obtained by projecting the respective objects on orthogonal planes in a predetermined direction from the CAD information of the plurality of objects. Two-dimensional obtained from a specific object captured from a predetermined direction by the camera by calculating the CAD figure feature amount such as the normal direction to the contour line at each point and the distance from the center of gravity to each point arranged on the contour line The same camera figure feature amount as the CAD figure feature amount in the image is calculated, and the CAD figure feature amount and the camera figure feature amount are compared to determine the type, posture, etc. of the specific object captured by the camera. A three-dimensional recognition method using CAD information. 2. The CAD figure feature quantity according to claim 1, wherein the maximum value of the distance from the center of gravity of the two-dimensional figure to each point arranged on the contour line and the next maximum value excluding the maximum value are shown.
A three-dimensional recognition method using CAD information that discriminates the normal direction of the direction connecting the dots as the gripping direction of the object by the robot hand. 3. The CAD figure feature quantity according to claim 1, which shows a minimum value of a distance from a center of gravity of a two-dimensional figure to each point arranged on a contour line and a next minimum value excluding the minimum value.
A three-dimensional recognition method using CAD information for discriminating the direction connecting the dots as the gripping direction of the object by the robot hand. 4. The CAD figure feature quantity according to claim 1, which is the most among the contour line portions in which points having almost no change in the normal direction to the contour line at each point arranged on the contour line of the two-dimensional figure are arranged. A three-dimensional recognition method using CAD information that discriminates the normal direction of a long object as the gripping direction of an object by a robot hand. 5. The CAD figure feature quantity according to claim 1, wherein the maximum value of the frequency distribution in the normal direction to the contour line at each point arranged on the contour line of the two-dimensional figure is examined, and the maximum value is calculated. A three-dimensional recognition method using CAD information that discriminates one of the pairs of normal directions deviated by about π from the grip direction of the object by the robot hand. 6. When the specific object according to claim 1 has a plurality of gripping directions, a gripping position is defined as a direction parallel to a gripping direction in which two or more of the gripping directions match and passing through the center of gravity. If all are different, the maximum value of the frequency distribution in the normal direction to the contour line at each point arranged on the contour line of the two-dimensional figure in the CAD figure feature amount according to claim 1 is examined, and the maximum value A three-dimensional recognition method using CAD information in which one of the pairs of normal directions deviated by about π from is set as the gripping direction of the object by the robot hand, and the direction parallel to the gripping direction and passing through the center of gravity is determined as the gripping position. . 7. A CAD information handling means comprising a camera and a computer, the computer creating and storing CAD information of a plurality of objects, and projecting each of the objects from the CAD information onto an orthogonal plane in a predetermined direction. CAD for guiding each two-dimensional figure, such as the center of gravity of the two-dimensional figure, the contour line, the normal direction to the contour line at each point arranged on the contour line, and the distance from the center of gravity to each point arranged on the contour line CAD figure feature quantity extraction means for calculating figure feature quantity, image data acquisition means for obtaining a two-dimensional image from a specific object captured by a camera from a predetermined direction, camera figure similar to the CAD figure feature quantity in the two-dimensional image A camera figure feature quantity extracting means for calculating a feature quantity, and a distinguishing means for comparing the CAD figure feature quantity and the camera figure feature quantity to determine the type, posture, etc. of a specific object captured by the camera. A three-dimensional recognition device using CAD information, including: 8. The computer according to claim 7, wherein the object is selected from the CAD figure feature amount of a two-dimensional figure derived from the CAD information of the object matched with the type, posture, etc. of the specific object captured by the camera. A three-dimensional recognition device using CAD information, which is added with a grip determination unit that calculates the grip direction and grip position of the.