JP2001277167A - 3D pose recognition method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a three-dimensional attitude recognizing method for taking components one by one using a robot hand from a group of many piled plate- like components. SOLUTION: This attitude recognizing method is used for taking components one by one using the robot hand 3 from the group 4 of many plate-like components piled in the three-dimensional coordinates. A label figure of the components to be recognized in the two-dimensional coordinates is formed using a label having height information based on luminance of reflected light from the group 4, and an attitude of the components to be recognized is recognized based on this label figure.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for recognizing a three-dimensional posture of a recognition target component used when a robot hand takes out a large number of flat plate components one by one from a group of stacked components.

[0002]

2. Description of the Related Art From the viewpoints of reduction of production cost, improvement of safety, and the like, in a processing or assembly line (production line), large automation using a robot has been attempted. At the beginning of the automation, the robot was used in a manner in which parts stored in a predetermined position and in a predetermined posture were supplied, and was premised on a large number of small varieties, so-called mass production. However,
In recent years, due to the influence of the diversity of consumers, it has become necessary to supply small quantities and various types of parts on the same production line. On the contrary, the problem that it takes time to arrange the parts has come to be caused. Therefore, a method of recognizing the three-dimensional posture of a target object (part, work) by a computer for controlling a robot has been developed, and a device for supplying parts using this method has been developed. I have.

[0003] In order to supply a small number of parts of various types, it is only necessary to (1) individually recognize the stacked parts and (2) determine the attitude of each part. For example, in Japanese Patent Application Laid-Open No. 05-134731, "high-speed picking device for piled parts", an image input camera processes a video signal of a work in a piled state by an object recognition device, and a line segment image obtained from a contour is compared with a matching model. Recognize the posture of the work by comparing it with (a model corresponding to a specific part consisting of a visually recognizable simple shape of the work shape) and calculate the picking part from the quantitative relative positional relationship with the specific part A device has been proposed.

Japanese Patent Application Laid-Open No. 10-089943 discloses a method for positioning an object by detecting a feature amount of the object and associating it with a reference model. No. 118975 "Handling Position Recognition Method and Recognition Apparatus" processes distance images obtained by projecting slit-shaped pattern light on the surface of the work, and projects illumination light onto the work surface with the slit fully open. The method of recognizing the handling position of the work using both the processing of the gray image to be performed and the method described in JP-A-11-190620 “Method and apparatus for recognizing the posture of a work, and a recording medium for a program for recognizing the posture of a work” are described in Japanese Patent Application Laid-Open No. 11-190620. The first using a sensor
The edge of one long side of the work is detected by the sensor, the sensor member is rotated counterclockwise of the first sensor in the detection state, and the edge of the long side is detected by the second sensor, and the posture of the work is determined. Each of the identifying techniques is disclosed.

[0005]

As is apparent from the above-mentioned conventional techniques, in each case, the recognition target component is grasped using a non-contact sensor or an image device (camera), and the sensor information or the image information is obtained. Is obtained by obtaining the posture information of the component by processing the robot and operating the robot based on the posture information. These basic techniques remain the same,
Each method has a different posture recognition method. For example, Japanese Patent Application Laid-Open No. 05-134731, Japanese Patent Application Laid-Open No. 10-08
In No. 9943, there is a method of collating with data stored in advance so as to correspond to a reference model. However, if these are changed in a recognition target part, it is necessary to change the data to be stored. There is.

In Japanese Patent Application Laid-Open No. Hei 10-118975, a recognition target component is distinguished by using both a distance image by slit light and a grayscale image by illumination light, but the components are piled up as a component group as the object of the present invention. When parts are separated, there is a possibility that a grayscale image cannot be obtained correctly. In JP-A-11-190620, there is a problem that it takes too much time to recognize one component, for example, rotation of a sensor is required. Therefore, a technique for easily and quickly recognizing the three-dimensional posture of a recognition target component when taking out a large number of flat components from a piled component group one by one with a robot hand was studied.

[0007]

What has been developed as a result of the study is a posture recognition method used when a robot takes out a large number of plate-like components one by one from a group of components stacked in three-dimensional coordinates. A label figure of a part to be recognized in two-dimensional coordinates using a label having height information based on the luminance of the reflected light from the part group, and a three-dimensional posture for recognizing the posture of the part to be recognized from the label figure It is a recognition technique. Specifically, in order to accurately carry out the gripping by the robot hand, it is configured from the posture determination in the preceding stage and the determination of the gripping direction. First, in the posture determination, in the above-described method, the maximum luminance in the reflected light from the component group is set as the maximum point T, and a label having different height information is attached according to a change in luminance from the maximum point T. Among the labels having a luminance higher than a predetermined luminance threshold, a label graphic of the recognition target component in the two-dimensional coordinates is formed, and a plurality of radial directions around the highest point T of the label graphic are formed. An average luminance value is calculated from the sum of the luminances.
A bisector is determined. If the intersection distance between the bisector and the contour of the label graphic is larger than a predetermined posture discrimination value, the recognition target component is in a horizontal posture. Is determined to be a vertical posture.

In the above method, the height from the parts group is
The luminance measuring device is defined as a reference axis in a vertical direction toward the component group, and a laser is irradiated from a direction inclined from the reference axis by an angle α, and the horizontal distance S between the reflection position of the laser and the reference axis is used as the reference axis. The height Z of the reflection position is obtained as Z = S / tanα. The highest point T is the height Z obtained in each measurement by repeating the measurement with the laser a plurality of times for the entire part group.
Is determined by comparing. As the measuring device, an image camera or an optical sensor can be used. Basically, the measuring device and the laser irradiating device constitute a set of measuring mechanisms, displace the components on the component group at equal intervals in the XY directions, and sample the height of the entire component group. Preferably, the laser is a line laser L, and a linear reflection position obtained for each measurement is measured as a horizontal distance from a coordinate axis parallel to the line laser L including the reference axis, and the measuring mechanism is set in the X direction or Y direction. Displacement at equal intervals in the direction can reduce the time required for measurement.

In a component group in which a large number of flat components are piled up, the three-dimensional postures of the individual flat components are randomly overlapped, but the brightness of the reflected light varies depending on the height. In the present invention, by utilizing the relationship between the height and the intensity of the luminance, a recognition target component having the highest point T which can be easily taken out by the robot hand is selected from the component group, and the three-dimensional posture of the recognition target component is determined. You do it. It is used to select such recognition target parts and determine the three-dimensional posture.
A label having different height information in accordance with a change in luminance, which is a feature of the present invention.

[0010] The plate-shaped component that provides the reflected light with the maximum brightness is located at the highest position in the component group. Since the plate-shaped component located at the highest position naturally has a continuous surface, it should provide reflected light having a luminance that continuously decreases from the maximum luminance. Therefore, a label graphic composed of only labels having a luminance higher than a predetermined luminance threshold represents only a flat part located at the highest position, and this is set as a recognition target part. When a plurality of flat parts are overlapped and each of them produces reflected light of the same level of brightness, the label figure obtained by adjusting the brightness threshold value appropriately can be corrected, or each of the complicated flat parts can be corrected. It is predicted that the contour of the constituent label graphic changes discontinuously, and by detecting the discontinuity of the contour, a single flat component can be set as the recognition target component.

Although it is possible to detect the three-dimensional coordinates and the inclination of the label figure to some extent by using only the label figure, in order to ensure accuracy, in the present invention, the posture is roughly classified (horizontal posture or vertical posture). Then, the three-dimensional coordinates and inclination of the recognition target component including the label graphic, that is, the three-dimensional posture, are determined based on each suitable technique. Therefore, first, (1) the luminance average value is calculated from the sum of the luminance in a plurality of radial directions centered on the highest point T of the label figure, and (2) the radial direction having a sum greater than this luminance average value. Determines the bisector in the label figure from (3) (a)
If the intersection distance between the bisector and the contour of the label figure is larger than a predetermined posture determination value, the recognition target component is in the horizontal posture. It is determined as a vertical posture. The highest point T of the label graphic always exists on the outline of the label graphic, that is, on the outline of the recognition target component, unless the recognition target component is completely horizontal. Therefore, when the sum of the luminances in the plurality of radial directions is respectively calculated, the sum of the luminances is lower than the average luminance in the radial direction including the outside of the recognition target component, and conversely, the sum of the luminances is averaged in the radial direction including the inside of the recognition target component. Brightness will be exceeded. The direction of the bisector thus determined represents the direction and horizontal angle when the recognition target component is projected on a horizontal plane. The length of the intersection point between the bisector and the contour of the label figure is proportional to the inclination of the recognition target component. Thus, using the label figure,
The posture of the recognition target component can be grasped.

More specifically, (a) for the recognition target component determined to be in the horizontal orientation, a regression plane R of the recognition target component is calculated from a point on the contour of the label figure, and the three-dimensional coordinates of the regression plane R are calculated. The gripping direction of this recognition target component is determined from the horizontal angle and the vertical angle with respect to (b) For the recognition target component determined to be in the vertical orientation, the brightness of the label aligned in the diagonal direction of the rectangle (filet diameter) surrounding the label figure The sums are compared, the surface direction of the recognition target component is determined in a diagonal direction in which the sum of the luminances increases, and the horizontal angle formed by the diagonal direction with respect to the two-dimensional coordinates is determined as the gripping direction of the recognition target component. The reason for roughly classifying the three-dimensional posture of the recognition target component is that a robot hand that holds the recognition target component from a substantially vertical direction in a horizontal posture, and a robot that holds the recognition target component from a substantially vertical direction in a vertical posture. This is because a hand is used. In this way, by using different robot hands according to the three-dimensional posture of the recognition target component, only the recognition target component can be reliably and quickly taken out from the component group without error.

An actual flat component is not a simple flat plate, but has some irregularities and steps, and there is usually a distinction between front and back. Therefore, it is preferable that the recognition target component taken out by the robot hand undergoes a front / back determination once before being transferred to the processing or assembly line in the next process. The above-described measurement mechanism can be applied to the front / back determination of the recognition target component. For example,
When a convex part is formed on the surface of the flat part, the extracted recognition target part is made horizontal at the measurement position, and a laser is irradiated obliquely toward the part where the convex part is formed. If it is present, it is determined that the projection is facing upward because the convex is upward, and (2) if the reflected light is located backward, it is determined that the convex is downward and is facing backward. The recognition target component for which the front / back determination has been completed may be temporarily stored in the stocker separately for the front and back, and then sent to the next process as appropriate.

[0014]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a layout diagram of a component supply device configured as a verification model of the three-dimensional posture recognition method of the present invention, FIG. 2 is a partial side view of a measuring mechanism 1 for explaining an algorithm for detecting a height, and FIG. FIG. 4 is a flowchart showing a process leading to posture classification after detecting the highest point T shown in FIG. 4, FIG. 4 is a state diagram of the recognition target component 2 when it is determined to be a horizontal posture, and FIG. FIG. 6 is a diagram showing the relationship between the inclination θ and the horizontal angle に お け る in the recognition target component 2, FIG. 7 is a flowchart showing the determination of the gripping direction of the recognition target component 2 in the horizontal posture by the robot hand 3, and FIG. FIG. 9 is a flowchart showing the determination of the gripping direction of the posture recognition target component 2 by the robot hand 3, FIG. 9 is a side view showing an example of the robot hand 3 used in this verification model, and FIG. Is a plan view that shows the case where the identified is, FIG. 11 is a plan view of the recognition target part 2 represents the case where identified as a table in the front and back judgment.

This example is a verification model of the three-dimensional recognition method according to the present invention. An annular flat plate (see FIG. 12 or FIG. 13) having a distinction between front and back surfaces due to unevenness is piled up in a case 5 to form a part group 4. And the robot hand 3 takes out each one.
As shown in FIG. 1, the verification model includes a posture discriminating device 6 in which the measuring mechanism 1 is displaced in the Y-axis direction on the piled component group 4, and a recognition target component 2 whose holding direction is determined by the posture discriminating device 6. A robot 7 that moves a robot hand 3 that takes out the object, and a recognition target component 2 that is taken out are placed in a horizontal position on a judgment table 8, and a front / back discrimination device 9 that judges front / back is configured, and these are controlled by a computer 10. The control of each device 6, 9 or the robot hand 3 can use a conventionally known technique.

The posture discriminating apparatus 6 has a stay 14 having both arms 13, 13 standing on a base 12 displaced in the Y-axis direction on a cross table 11, and the stay 14 constitutes the measuring mechanism 1. The measuring mechanism 1 has a laser device 16 that directs a CCD camera (measuring device) 15 from the center of the stay 14 in the vertical direction and irradiates the line laser L obliquely from both arms 13. In the present invention, the orientation can be determined by providing one laser device 16 that irradiates the line laser L obliquely with respect to the direction of the CCD camera 15 = the vertical direction. Since accurate measurement cannot be performed with only the line laser L from one direction, two laser devices 16 and 16 are used in this verification model.

In the present invention, first, the highest point T in the component group 4 piled up in the case 5 must be detected, and the component 2 to be recognized must be selected. For this reason, the posture determination device 6
While displacing the measurement mechanism 1 above the parts group 4 at a constant interval, the luminance of the reflected light by the line laser L is measured by the CCD camera 15 to obtain the position of the maximum luminance. In particular,
As shown in FIG. 2, the reflection position of the line laser L irradiated from the oblique direction of the angle α is compared with the distance S from the reference line of the CCD camera 15, and the distance S is compared with the height difference from the CCD camera 15 ( As Z = S / tan α), the height of the component that causes each reflected light is obtained indirectly. The position of the reflected light measured by the CCD camera 15 is only relative coordinates with respect to the reference line defined by the CCD camera 15, but since the CCD camera 15 increases or decreases only the Y coordinate according to the base 12 displaced in the Y-axis direction, By reflecting the increase or decrease of the Y coordinate on the relative coordinates, the three-dimensional coordinates (x, y, z) of the position of each reflected light can be obtained.

At the positions of the respective reflected lights thus obtained, the coordinates having the maximum luminance are regarded as the highest point T of the component group 4 and the recognition target component 2 and the recognition target component is determined based on the flowchart shown in FIG. 2 is performed.
In the measurement of the entire part group 4, the detection of the highest point T may be roughly performed in order to reduce the time. For this reason, after selecting the recognition target component 2, the highest point T is measured again. If the highest point T of the entire component group 4 is accurately and precisely measured from the beginning, it is not necessary to measure the highest point T of the recognition target component 2 again.

The posture determination starts with labeling of the recognition target component 2 according to the luminance information. This verification model uses color labeling to monitor the results of labeling, etc.
(For example, a monitor of the control computer 10). First, in order to sort the recognition target component 2 from other components, an appropriate brightness threshold value Si that does not pick up the brightness of other components is determined so that labeling can be performed only with the reflected light from the recognition target component 2. I do. More specifically, the edge of the label Eg, Eg (the end where the vector of the outline of the image formed by the label changes suddenly, while the maximum luminance is set to the maximum value of the luminance threshold Si and the luminance threshold Si is gradually reduced, see FIG. 1) of the brightness threshold value Si at which the distance between
Immediately before this, a brightness threshold value Si at which the brightness of other components is not picked up is set. When labeling with parts overlapping,
The distance between the edges Eg and Eg in the label becomes shorter due to the overlapping of the parts and the outline is interrupted. From this, it is possible to compare the distance between the edges Eg and Eg, and determine the one immediately before the luminance threshold Si whose length has suddenly become short as the luminance threshold Si for separating the other component from the recognition target component 2. . The label graphic 17 is configured as a label image of light reflected by only the recognition target component 2 based on the luminance threshold Si.

The procedure is as shown in the flow chart. First, the luminance threshold value Si is set to 255 (in the flow chart, the initial value is 256 for calculation and 1 is subtracted at the beginning of the loop). The reflected light of only points having brightness is acquired, and the image acquired by the CCD camera 15 is
Convert to value and label the same color (red in this verification model). The distance between the edges Eg, Eg is determined from the outline of the label. Next, the luminance threshold value Si is reduced to 254, labeling is performed in the same manner, the distance between the edges Eg, Eg is obtained, and compared with the distance between the previous edges Eg, Eg. After that, the brightness threshold is set to 1
Edges Eg, Eg corresponding to each luminance threshold
While obtaining the distance between the edges Eg, a luminance threshold value Si at which the distance between the edges Eg is maximized is obtained. Since this luminance threshold value Si obtains a label including light reflected from other components,
The label figure of only the recognition target component 2 has the luminance / luminance threshold Si
+1 is used for binarizing and labeling again. The previously calculated label graphic 17 may be stored, and the stored label graphic 17 may be used.

Next, the posture of the recognition target component 2 is determined. First, as shown in FIG.
The sum of a plurality of luminances in the radial direction is obtained with the highest point T at 17 as the center. In this verification model, the sum of luminances in a fixed range in 24 equally spaced directions is obtained. The sum of the luminances is further summed and divided by 24 to obtain an average luminance value. The bisector 18 of the label graphic 17 is determined from these directions, leaving only directions above the average luminance value. The highest point T must exist on the outline of the label graphic 17, and the sum of the luminance in the radial direction outside the recognition target component 2 is lower than the average luminance value, and conversely, the sum of the luminance in the radial direction inside the recognition target component 2. It is obvious that is higher than the luminance average value, and thus the bisector 18 can be obtained.

Then, an intersection between the bisector 18 and the contour of the label graphic 17 and a horizontal distance D between the intersections are obtained. The length of the horizontal distance D changes depending on the inclination θ of the recognition target component 2, and is compared with a preset posture determination value K.
The posture of the recognition target component 2 can be roughly classified into a horizontal posture or a vertical posture. In this verification system, taking into account the gripping ability of the robot hand 3 described later, the posture determination value K is calculated by setting θ = 70 degrees, and if the inclination θ of the recognition target component 2 is smaller than 70 degrees, the lateral posture (FIG. 4) If the inclination θ is 70 degrees or more, the posture is vertical.
(Fig. 5).

In order to hold the recognition target component 2 with the robot hand 3, the part to be recognized of the recognition target component 2 is determined and its three-dimensional coordinates are obtained. Is important. The grip direction is determined by the inclination θ and the horizontal angle の of the recognition target component 2 as shown in FIG. The posture has already been roughly divided, but this is just robot hand 3
Therefore, the gripping mechanism is selected, and then the horizontal gripping and the vertical gripping are separated, and the correct gripping direction is determined by different procedures (FIGS. 7 and 8).

In the case of the horizontal posture, the flowchart shown in FIG. 7 is followed. Based on the label figure already determined (the reflected light may be measured again with the luminance threshold value Si + 1, the label figure 17 may be formed through binarization and labeling), the label figure 17 is first arranged on the contour of the label figure 17 The three-dimensional coordinates of each of the plurality of points are obtained. It is certain that the three-dimensional coordinates exist on the recognition target component 2, and the regression plane R of the label graphic 17 is obtained based on the three-dimensional coordinates, and the inclination θ is specified from the normal vector H of the regression plane R. I do. The calculation of the regression plane R uses a conventionally known method. The horizontal angle 法 that the normal vector H has with respect to the XY plane is the horizontal angle の of the recognition target component 2.
Becomes Thus, (1) the highest point T (three-dimensional coordinates of the maximum brightness) of the recognition target component 2 necessary for gripping by the robot hand 3, (2) the inclination θ at the highest point T, and (3) the horizontal at the highest point T Since all the angles ψ have been obtained, the robot hand 3 may be moved in accordance with the information, and the recognition target component 2 may be pinched and taken out.

In the present verification system, all vertical postures at an inclination θ of 70 degrees or more can be grasped by approaching the robot hand 3 from above. Only the angle ψ needs to be specified. The procedure is shown in the flowchart of FIG.
The label figure 17 already determined (again, the luminance threshold value Si + 1
The reflected light may be measured by binarizing and labeling may be configured through labeling) .First, the Feret diameter of the label graphic 17 is determined, and the sum of the luminances on the diagonal lines T1 and T2 at this ferret diameter is obtained. Find R1 and R2. The luminance sums R1 and R2 are compared, and if R1> R2, the recognition target component 2 is determined to be parallel to T1, and the horizontal angle of T1 is determined as the horizontal angle の of the recognition target component 2. Conversely, if R1 <R2, the horizontal angle of T2 is determined as the horizontal angle の of the recognition target component 2. Thus, robot hand 3
(1) the highest point T of the part 2 to be recognized,
Since the inclination θ at the highest point T (= 90 degrees) and (3) the horizontal angle に お け る at the highest point T are all obtained, the robot hand 3 is moved according to the above information, and the recognition target component 2 is held. I just need to take it out.

In the present verification model, the components to be recognized 2 are taken out in the horizontal position and the vertical position, because the target object of the present invention is a flat plate, and the gripping mechanism at the end position in the horizontal position and the vertical position. This is because if the robot hand 3 has only the robot hand 3, the operation amount of the robot hand 3 or the robot 7 becomes large, control becomes difficult, and interference with other devices may occur. So robot hand 3
FIG. 9 shows an example in which the robot hand 3 is configured to be able to appropriately switch the holding portion while suppressing its own operation range. The robot hand 3 has two vertical and horizontal air cylinders 19 and 20. The horizontal air cylinder 20 is actuated by moving the horizontal chuck 21 toward the recognition target component 2 in the horizontal posture from the gripping direction, and The vertical chuck 22 is moved closer to the recognition target component 2 from the gripping direction, and the vertical air cylinder 19 is moved.
Activate In addition, it is assumed that it is difficult to make the chuck claw go under the recognition target component in the vertical position, and in such a case, a magnet or suction cup is attached so that it can be picked up by suction. It may be attached to a robot hand.

In the present invention, the purpose is to take out one by one from the component group of the flat plate components piled up by the robot hand, and to recognize the three-dimensional posture, each component is regarded as a flat plate. The shape of the part is undulating even if it is small, and there is usually a distinction between front and back. Therefore, in this verification model, the recognition target component 2 taken out by the robot hand 3 is placed on a horizontal judgment table 8 of a front / back discrimination device separately.
The line laser L is applied obliquely to the line laser L.
The front and back are determined based on which of the axes is shifted inside or outside.

More specifically, as shown in FIG. 10 and FIG. 11, if the recognition target component 2 is an annular component having a star-shaped protrusion 23 protruding to the front side, the recognition target component 2 crosses the protrusion 23. The line laser L is irradiated as described above, and the front and back sides are determined based on the position of the reflected light of the line laser L. Here, if the recognition target component 2 placed on the judgment table 8 faces the back side, the reflected light by the line laser L appears partially inside (the side close to the line laser L), and conversely, the recognition target component 2 , The light reflected by the line laser L appears partially outside (on the side far from the line laser L). The front / back discrimination device 9 can also perform quality judgment by recognizing the shape of the component based on the relationship of placing the recognition target component 2 at a predetermined position on the judgment table 8 at the time of front / back determination.
In addition, without using the front / back discrimination device 9, the robot hand 3 takes a specific posture (the recognition target component 2 is made horizontal at a specified position) while holding the recognition target component 2, and reflects the line laser L irradiated from a specific direction. The front and back may be determined from light.

[0029]

According to the three-dimensional recognition method of the present invention, without using a database storing the data of the shape of the component to be extracted, only the processing of the label figure derived from the luminance of the reflected light is performed from the component group. It has the feature that it can select the recognition target parts and specify the gripping direction of the robot hand. Since a part that is a three-dimensional object is converted into a two-dimensional figure (label figure) having height information and processed, it is difficult to target a three-dimensional part with a complicated external shape. If so, the use of the three-dimensional posture recognition method of the present invention enables quick and accurate control of the robot hand.

As is evident from the above example of the verification model, the hardware configuration includes existing devices, sensors, personal computers, etc., other than the robot hand having different holding parts by discriminating the recognition target component into the horizontal posture or the vertical posture. Can be used to configure the system, and there is also an advantage that the present invention can be applied to various kinds of plate-shaped components by modifying software. In the illustrated example, the device configuration for the flat annular component is shown, but it is needless to say that a general component supply device for other flat components can also be configured. As described above, the basic method of the present invention is widely applicable, and realizes the provision of a general-purpose component supply device in a future small-lot multi-product line.

[Brief description of the drawings]

FIG. 1 is a layout diagram of a component supply device configured as a verification model of a three-dimensional posture recognition method of the present invention.

FIG. 2 is a partial side view of a measuring mechanism for explaining an algorithm for detecting a height.

FIG. 3 is a flowchart showing a process leading to posture classification after detecting a maximum point T indicating a maximum luminance.

FIG. 4 is a state diagram of a component when it is determined that the component is in a horizontal orientation.

FIG. 5 is a state diagram of a component determined to be in a vertical posture.

FIG. 6 is a diagram illustrating a relationship between a tilt θ and a horizontal angle に お け る in a recognition target component.

FIG. 7 is a flowchart showing a process of determining a gripping direction of a recognition target component in a lateral posture by a robot hand.

FIG. 8 is a flowchart showing a process of determining a gripping direction of a recognition target component having a vertical posture by a robot hand.

FIG. 9 is a side view illustrating an example of the robot hand used in the present verification model.

FIG. 10 is a plan view illustrating a case where the recognition target component is identified as the back in the front / back determination.

FIG. 11 is a plan view illustrating a case where the recognition target component is identified as a front in the front / back determination.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Measuring mechanism 2 Recognized part 3 Robot hand 4 Parts group 6 Attitude discriminating device 9 Front / back discriminating device 10 Computer 15 CCD camera (measuring device) 16 Laser device 17 Label graphic 18 Bisection line T Highest point α Laser irradiation angle S Horizontal distance Z Height of reflection position R Regression plane K Posture discrimination value θ Tilt of recognition target component 水平 Horizontal angle of recognition target component Si Luminance threshold H Normal vector of regression plane

 ──────────────────────────────────────────────────続 き Continued on the front page F term (reference) 2F065 AA04 AA12 AA21 AA24 AA35 AA37 AA51 BB01 BB27 DD06 FF01 FF02 FF09 FF42 GG04 HH05 HH12 HH14 JJ03 JJ09 JJ26 NN17 NN20 PP02 QQ00 QQ08 QQQAQQQAQQAQQAQQAQQAQAQAQ DB07 DB10 DC07 DD11 FA03 FA05 FB01 FB12 FB22 FC04 FC13 FC14

Claims (5)

[Claims] 1. A posture recognition method used when a robot hand picks up a large number of plate-shaped parts in a three-dimensional coordinate system from a group of parts one by one, based on the luminance of light reflected from the parts group. Using a label with height information 2
A three-dimensional posture recognition method in which a label graphic of a recognition target component is configured in dimensional coordinates, and the posture of the recognition target component is recognized from the label graphic. 2. An attitude recognition method used when a robot hand takes out a large number of plate-shaped components from a pile of components in three-dimensional coordinates one by one by a robot hand. As the highest point T, a label having different height information is attached in accordance with a change in luminance from the highest point T, and only the label having a luminance higher than a predetermined luminance threshold among the labels is in two-dimensional coordinates. A label figure of the part to be recognized in step (a), a luminance average value is calculated from the sum of luminance in a plurality of radial directions centering on the highest point T of the label figure, and a sum larger than the luminance average value is calculated. Determine a bisector in the label figure from the radial direction
A three-dimensional posture recognition method in which the recognition target component is determined to be a horizontal posture when the intersection distance between the equal line and the contour of the label graphic is larger than a predetermined posture determination value, and the recognition target component is determined to be a vertical posture when the intersection distance is small. . 3. The height from the part group is determined by irradiating a laser from a direction inclined at an angle α from the reference axis with the vertical direction facing the luminance measuring device as the reference axis. 3. The three-dimensional posture recognition method according to claim 2, wherein a height Z of the reflection position is obtained as Z = S / tanα from a horizontal distance S between the reflection position and the reference axis. 4. For a recognition target component determined to be in a horizontal posture, a regression plane R of the recognition target component is calculated from a point on the contour of the label figure, and a horizontal angle and a vertical angle of the regression plane R with respect to three-dimensional coordinates are calculated. The three-dimensional posture recognition method according to claim 2, wherein a gripping direction of the recognition target component is determined from the following. 5. For a recognition target component determined to be in a vertical orientation, the sum of the luminances of labels arranged in a diagonal direction of a rectangle (filler diameter) surrounding the label figure is compared, and the recognition is performed in a diagonal direction in which the sum of the luminances increases. 3. The three-dimensional posture recognition method according to claim 2, wherein a surface direction of the target component is determined, and a horizontal angle formed by the diagonal direction with respect to the two-dimensional coordinates is determined as a gripping direction of the target component.