JPS63174890A - Method of recognizing body - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to an object recognition method, particularly an object recognition method in which objects stored in bulk in a container are photographed using a TV right camera, and the recognition and extraction work is performed. Regarding improvements.

[Prior Art] An object recognition method using a TV right camera is widely used today to recognize parts stored in bulk in a container and take them out using a robot hand.

Such recognition is performed using a TV right camera and a device that performs image processing on the signal, and the image processing device performs binarization processing and gray processing on the video signal output from the TV right camera. Alternatively, it is formed to perform distance image processing or the like to recognize the various images described above.

By the way, there is a strong desire for automation in manufacturing processes in factories today, and line automation is progressing for this purpose, but object recognition technology for automatically supplying parts to the line is lagging behind, and its effectiveness is limited. countermeasures were desired.

(In other words, parts, especially cast parts, etc., that are brought into the line entrance are randomly stacked on pallets. Therefore, each part must be accurately recognized and taken out from among them and placed on the line.) There is a problem in that the work of arranging and supplying objects is technically extremely difficult to automate, and the work of recognizing and extracting such objects must rely on manual labor.

Among other things, such recognition extraction work requires 1. Since this is a monotonous and repetitive work, and the parts to be removed are often heavy, there is a strong desire to automate this process.

For this reason, the development of a robot system for picking up bulk parts using image processing technology and robot technology has been progressing. Perform recognition work to measure the position and orientation of the part, give the position and orientation data of the part recognized in this way to the robot, and use the robot to grasp the part and feed it to the production line in the determined orientation. Efforts are underway to do so.

FIG. 3 shows an example of such a robot system, and FIG. 4 shows an example of an object recognition method used in this robot system.

This robot system uses a TV left camera 4 to project parts 12 stored in bulk on a pallet 10.
The video signal is outputted to the image processing device 16.

At this time, if the hand 24 or arm 28 of the robot 20 is within the field of view of the TV left camera 4, it is
After moving out of the field of view of the left camera 4, object recognition work is started.

Based on the video signal input from the TV left camera 4, the image processing device 16 recognizes the coordinates, shape, and orientation of the desired part 12 that can be taken out from the parts stacked in bulk on the pallet 10, and uses the data. The signal is output to the robot control device 18.

That is, the image processing device 16 performs image processing on a human-powered video signal, recognizes the position and orientation of the removable parts 12 in the pallet 10, and controls the robot controller 1.
8 outputs a recognition completion signal, converts the recognized position and orientation of the part 12 into values in the robot coordinate system, and prepares for transmission to the robot control device 18.

After confirming that the recognition completion signal has been output, the robot control device 18 receives recognition data from the image processing device 16 and controls the articulated robot 20 based on the data. The desired part 12 is picked up from the pallet 10 using a hand 24 and placed on a conveyor 26.

That is, upon receiving data, particularly component position data, from the image processing device 16, the robot control device 18 moves the robot hand 24 above the component 12 based on the data, and then moves the robot hand 24 toward the target component 12. Start approaching.

At this time, the robot control device 18 controls the robot hand 1.
The hand 24 continues to approach while monitoring the signal from the workpiece confirmation sensor installed at part 8, and when the workpiece confirmation signal turns on (work present), the hand 24 stops approaching and moves the hand 24 toward the part 12.
grasp.

Then, the component 12 gripped by the hand 24 is moved toward the component supply conveyor 26 and placed on the conveyor 26 .

At this time, when the hand 24 and arm 28 of the robot 20 are removed from the field of view of the TV left camera 4, the robot system resumes the recognition operation for the next part 12 and moves the part from the pallet 10 to the conveyor 26. Repeat the process of extracting the item.

[Problems to be Solved by the Invention] In such a conventional system, a judgment as to whether or not parts 12 stored in bulk on a pallet 10 can be taken out is obtained by image processing a video signal. The part dimensions and
The actual dimensions of the component are compared with preset dimensions, and the determination is performed based on whether or not the two match within a predetermined recognition tolerance.

However, the dimensions obtained by image processing the video signal are often different from the dimensions of the parts actually stored in the pallet 10, and (a) some parts are stored in the pallet 10 overlapping with other parts. , unremovable parts 12
was mistakenly determined to be removable, and the robot 20
(b) The parts that overlap each other in a complicated manner little by little on the pallet 10 may be judged as impossible to take out and are left behind on the pallet 10. Therefore, there is a problem in that the parts 12 stored in the pallet 10 cannot be accurately recognized and efficiently taken out, which is a major cause of a decrease in the part 12 removal and supply capability.

The inventors analyzed how such recognition errors occur and found the following causes. Let us briefly explain the cause of this occurrence.

First, as shown in FIG. 6, when the shaft part 12 (shown in FIG. 5) is placed on the bottom plate of the barre 1-10 (position A in FIG. 6), the TV left camera 4 and the distance between the article 12 and the article 12.

Assuming that, the bottom plate of the pallet 1o.

Let the height be the reference height (position B in FIG. 6).

And the standard height within /rut10. When the shaft part 12 is placed horizontally in the position, the TV left camera 4
Assuming that the length seen from the TV left camera 4 is 1, the reasons why the length of the shaft component 12 looks different when seen from the TV left camera 4 can be broadly classified into the following three reasons.

■The first cause is when the shaft component 12 is tilted at an angle θ with respect to the horizontal plane, as shown in FIG.
For example, the shaft component 12 has a reference height. , when it is tilted by θM□ with respect to the horizontal plane, its length becomes fl cosθ
It is understood that it looks short as MAX.

■The second cause is when the distances of the shaft parts 12 stored in the pallet 10 with respect to the TV camera 14 are different; for example, the shaft parts 12 are higher than the standard height. At a high position (position C in Fig. 6), the length appears longer as J (hc-ho)/(h.-2ho), and is also longer than the reference height position. At a low position (': position A in the j56 figure), J! (hc-ho)/h.

It looks short.

That is, when the shaft component 12 is located at a higher position than the reference height CB, the component 12 as seen from the TV left camera 4
It will be understood that the length of B) appears to be longer than the actual length of the part, and also appears shorter than the actual length of the part if it is located lower than the reference height B).

■The third cause is when a plurality of shaft parts 12 overlap each other.

For example, when shaft parts overlap each other as shown in FIG. 8, the length of the part as seen from the TV camera 14 is 1 even though the actual length of the part 12 is 1.
It will be mistakenly recognized as ``.

In order to solve the problems caused by these causes, for example, the video signal is image-processed, and the shaft part 1
When determining dimension 1 of item 2, refer to (■) above.

In anticipation of the error caused by the cause of (2), it is conceivable to take measures such that even if the recognized dimension of the shaft component 12 is somewhat different from the actual dimension 1 of the component, it is recognized as a removable component.

However, if the error tolerance is set to ±20 to 30%,
As a result, as shown in the cause of (■) above, a part that cannot be taken out due to overlapping other parts is mistakenly determined to be able to be taken out, and the robot hand 24 frequently makes mistakes in taking out the work. A problem arises in that this occurs.

On the other hand, if the error tolerance range is set strictly, as shown in FIG. 9, if some shaft parts overlap each other little by little or stand up,
Even though there are shaft parts 12 that can be taken out within the pallet 1δ, it is determined that all the parts 12 are short outside the tolerance range and cannot be taken out, and the pallet 1
A problem arises in that removable parts within 0 are left behind.

For this reason, there has traditionally been a demand for the development of an object recognition method that simultaneously satisfies these two conflicting demands, that can accurately recognize parts stacked in bulk on the pallet 10, and that can perform the task of removing the parts. was.

[Related technology] As a related technology to the present invention, Japanese Patent Application Laid-open No. 59-198307
This application relates to a tracking determination method for apparatus objects that can make accurate judgments even when rod-shaped objects are in contact with each other, overlap each other, and are scattered. This method may be similar to the present invention in that a video signal taken by a TV camera is used.

However, this application is characterized in that a bar-shaped object is tracked and searched diagonally from a transverse position of the bar-shaped object, and the recognition tolerance range for the reference dimension of the object to be recognized is set in multiple stages. This is clearly different from the present invention, in which a robot hand is used to take out objects stored in bulk in a container based on an appropriate and optimal tolerance range.

[Object of the Invention] The present invention was made in view of the above-mentioned conventional problems, and its purpose is to accurately recognize objects stored in bulk in a container and to efficiently recognize objects using a robot hand. By providing a method for recognizing objects that allows them to be easily retrieved.

[Means for solving the problem] In order to achieve the above-mentioned object, the present invention uses a TV camera to photograph objects stored in bulk in a container, and based on the video signals, images are taken from inside the container according to the following steps. It is characterized by repeatedly performing a series of operations in which a desired removable object is recognized and the recognized object is taken out to a predetermined position outside the container using a robot hand.

(b) Image processing of video signals is performed by setting recognition ranges in at least three levels: the strictest tolerance range, the relatively loose tolerance range, and the loosest tolerance range, with respect to the standard dimensions of the object to be recognized. A first step in which the recognition work is performed while sequentially lowering the recognition tolerance range until it is recognized that there is an object that can be taken out from the container by the robot hand, based on the object dimensions obtained by the robot hand.

(b) In the first step, if it is determined that there is an object that can be taken out based on the strictest recognition tolerance, the object is taken out to a predetermined position outside the container using a robot hand, and then the Returning to the first step, the same operation is repeated, and if it is determined that there is an object that can be taken out based on the relatively loose recognition acceptance criteria, the object is moved to a predetermined location outside the container using a robot hand. At the same time, it is determined which of the above-mentioned tolerance ranges the object dimensions obtained during the recognition process of the taken-out object correspond to, and then the process returns to the first step and the next step is performed from the determined tolerance range. The second step is to restart the recognition task for the object.

(c) In the first step, if it is recognized that there is an object that can be taken out based on the loosest recognition tolerance, a robot hand is used to take out and stir the object, and the removal and stirring It is determined which of the above-mentioned tolerance ranges the object dimensions obtained during object recognition work fall under, and then the process returns to the first step and restarts the recognition work for the next object from the determined tolerance range. and a third step of terminating the object recognition task if it is recognized in the first step that there is no retrievable object even based on the loosest recognition tolerance range.

[Operation] As described above, according to the present invention, by fluidly setting the permissible object recognition range according to the situation, it is possible to sequentially take out objects starting from the one that is easiest to take out under each condition. can be carried out reliably and efficiently.

In addition, even when conventionally it would be determined that there are no more removable objects in the container, by stirring the objects using a robot hand, the objects can be untangled and stored in the container. This makes it possible to make a removable object removable, thereby preventing work errors such as leaving a removable object behind even though the removable object is stored in the container.

[Embodiment] Next, the object recognition method according to the present invention will be explained by taking as an example the case where it is applied to the bulk parts retrieval robot system shown in FIG. 3. Components corresponding to those shown in FIGS. 3 to 9 are designated by the same reference numerals, and their explanations will be omitted.

The system of this embodiment photographs the shaft parts 12 stored in bulk on a pallet 10 using the TV left camera 4, and processes this video signal to determine whether the shaft parts 12 recognized can be taken out by a robot. The robot hand 24 is configured to take out the removable parts 12 from the pallet 10 onto a conveyor 26 by using a robot hand 24.

In the present invention, the determination as to whether or not the shaft component 12 in the pallet 10 can be taken out is based on the entire length of the shaft component 12 recognized through image processing using a video signal, that is, as shown in FIG. , based on whether the length of the cylindrical portion of the shaft component 12 falls within a preset object recognition tolerance range.

However, as mentioned above, even if the entire part 12 is visible when viewed from the TV left camera 4, as already mentioned, (1) the distance from the TV left camera 4 to the part 12, (2)
The length of the cylindrical portion of the component 12 appears to be different depending on the inclination of the component 12.

For this reason, in the system of this embodiment, as shown in FIG. The reference height is set to the position B in FIG. 6), and the length J of the shaft component 12 installed horizontally at this reference height position is registered in advance in the image processing device 16 as the reference length.

Further, the present inventor has proposed that when the parts 12 stored in bulk in the pallet 10 can be taken out by the robot hand 24 even if the parts 12 are tilted up to the maximum σMAX with respect to the horizontal plane, the TV
An analysis was performed on how the dimensions of the removable shaft component 12 as seen from the right camera 4 change.

The results of this analysis are shown in FIG. 2, in which the horizontal axis represents the storage position (height) of the shaft component 12 within the pallet 10, and the vertical axis represents the shaft component as viewed from the TV right camera 4. Twelve dimensions are shown.

In the figure, the shaft component 12 that can be taken out by the robot hand 24 can be easily determined from geometrical relationships to occupy the range shown by diagonal lines in the figure. Moreover, when the storage position of the shaft component 12 in the pallet is lowered, one end of the shaft component 12 hits the bottom of the pallet 10.
The same figure also shows that the possible inclination angle becomes smaller.

The first characteristic feature of the object recognition method according to the present invention is that a plurality of recognition tolerance ranges are provided for the reference dimension 1 of the shaft component 12, and the recognition tolerance range is set in multiple stages until the target shaft component 12 is recognized. The purpose is to recognize the shaft parts 12 while lowering them sequentially.

In this embodiment, in order to determine whether the part 12 recognized by the image processing device 16 can be taken out, three allowable length ranges are provided as follows.

(a) Tolerance range 1; (J!-0) or j! +
Range a) (range indicated by A in Figure 2) (b) Tolerance range 2; (J!-b) or J! +
a) (Range indicated by B in Figure 2) (c) Tolerance range 3: Width W of the robot hand or more and (
Jl+a) The recognition work for the part 12 is continued while switching between these three levels of allowable ranges depending on the situation.

FIG. 1 shows a preferred example of the object recognition method according to the present invention. A recognition operation is performed to determine whether or not there is a part 12 (step 2.3).

When it is recognized that there is a part 12 that can be taken out, the robot hand 24 is used to take out the part 12 (step 4), and a series of such article recognition and removal work (step 2. 3.4), tolerance range 1
This process is repeated until it is determined that there are no more parts 12 that can be taken out.

Here, the part 12 that is determined to be removable based on the tolerance range 1 is a part having a length shown in the range A in FIG. 2, that is, the reference height within the pallet 10.

Therefore, the parts 12 that are longer than the length 1 of the parts placed on the pallet 10 and for which the recognition and extraction work is performed based on the above-mentioned tolerance range 1 are located at a position higher than the reference height ho in the pallet 10 and have a slope. It is understood that only small and easily removable items are covered.

In this way, according to the present invention, from the parts 12 stacked in bulk on the pallet 10, parts stacked at the top of the pallet 10, with a small inclination, and which can be taken out reliably are taken out in order, which is extremely rational. This makes it possible to greatly reduce parts removal errors and improve work efficiency.

Then, in the recognition work based on tolerance range 1, if it is determined that the removable parts 12 are no longer found (step 3), the tolerance range is lowered by one rank, and the removable parts 12 are no longer found based on tolerance range 2. It is determined whether the part exists or not (step 5.6).

If it is recognized that there is a part 12 that can be taken out (step 7), the robot hand 24 is used to similarly take out the part (step 8).

Here, the part 12 that is determined to be removable based on the allowable range 2 is a part with a length shown in the range B in FIG. Therefore, this kind of parts are
It will be understood that they exist evenly at all height positions within the pallet 10, and that their inclination angle θ is within the maximum inclination angle OMAX determined by the constraints of the robot hand 24.

When the extraction work for the component 12 that has been determined to be removable based on the tolerance range 2 is completed, the robot hand 24 collides with other parts during the extraction operation, or the parts collide with each other, causing the parts to overlap each other. Alternatively, it is conceivable that the packaging state of the parts that are standing may change, and the parts 12 may appear in the pallet 10 in a position that is easier to take out.

The second characteristic feature of the present invention is that the object recognition and extraction work is performed with a view to this point.

In other words, if the part recognition and extraction work is performed based on a tolerance range other than the strictest tolerance range (tolerance range 2 and tolerance range 3 in the example), the dimensions of the part 12 obtained during the recognition work are (dimensions obtained by image processing) falls within the tolerance range 1.2 or 3 (step 9, step 15, step 16).
).

Thereafter, the recognition and extraction work for the next part 12 is started from the allowable range determined in this way.

In the embodiment, when performing the component recognition and extraction work (step 6.7) based on the tolerance range 2, the length of the component 12 to be extracted is temporarily stored.

When the recognition and extraction work for the part is completed (step 8), the length of the next extracted part 12 is within the tolerance range 1.
(Step 9).

As a result, if it is determined that there is no part 12 with a length that belongs to tolerance range 1, the recognition and extraction work based on tolerance range 2 (step 6.7.8) is performed in the same way, and this extraction work is completed. Thereafter, it is determined whether the length of the part 12 taken out falls within the tolerance range 1 (step 9).

According to this embodiment, such a series of operations is repeated to sequentially take out the parts 12 recognized based on the tolerance range 2 from the pallet 10.

By the way, as the work to take out the parts 12 is continued in this way, the packaging condition inside the pallet 10 gradually changes as described above, and parts 12 in positions that are easier to take out appear on the pallet 10. There are many.

According to the present invention, since it is also determined at the same time whether or not the removed part 12 belongs to the tolerance range 1 (step 9), the packing appearance in the pallet 10 changes and the part 12 is placed in a position that makes it easier to take out the part 12. appears, the next recognition task will be performed based on the tolerance range one rank higher, that is, tolerance range 1.

Therefore, in this case, we restart the extremely rational work of taking out the parts 12 stacked in bulk on the pallet 10 in order of parts that are stacked at the top of the pallet 10, have a small slope, and can be taken out reliably. It is possible,
It becomes possible to further improve the work efficiency.

Further, according to the recognition method of the embodiment, if it is determined that there is no removable part 12 even by recognition based on the tolerance range 2 (step 7), the recognition level is further lowered by one rank, and the recognition level is further lowered by one rank. Parts 1 that can be removed based on 3.
Perform recognition work to determine whether or not 2 exists (step 11)
.

That is, if it is determined that there are no parts 12 that can be taken out based on the tolerance range 2, the parts 12 stored in the pallet 10 may overlap or stand on top of each other as shown in FIG. Either this is the case, or all the parts from the pallet 10 have been used up.

Therefore, according to the present embodiment, in such a case, the recognition level of the article is lowered by one rank, the tolerance range 3 is selected, and it is recognized whether or not there is an article that can be taken out (steps 11 and 12). .13).

This recognition range 3 is set so that if even a small part of a part is visible in the pallet 10, it is determined that it can be taken out, so if it is determined that there is no part that can be taken out within this tolerance range 3, (Step 13) is a case where there is no component 12 already in the pallet 10, and in such a case, the article recognition and extraction work is promptly completed.

Further, if it is recognized that there is a part 12 that is determined to be removable based on the tolerance range 3 (step 13)
, the robot hand 24 goes to take out the part (
Step 14).

In this case, the parts that are judged to be removable actually overlap or stand on the pallet 10, so the retrieval operation by the robot hand 24 is not necessarily successful; It is possible to change the packing form into a shape that makes it easy to take out the parts 12.

In particular, when the robot hand 24 is used to take out the component 12 that has been determined to be removable based on the tolerance range 3, unlike the cases of the tolerance range 1 and tolerance range 2, the robot stirs the component 12 itself. It is preferable to control the hand 24, and by doing so, the packing form of the parts can be greatly changed and the parts stored in the pallet 10 can be easily taken out.

When the part recognition and extraction work based on the tolerance range 3 is completed, the dimensions of the part 12 determined by image processing during the recognition and extraction work belong to the tolerance range 1 or fall within the tolerance range 2. A judgment is made as to whether it belongs to the tolerance range 3 or not (steps 15 and 16).

If it is determined that the previously extracted component 12 falls within the tolerance range 3, the series of recognition and extraction operations in steps 12 to 15 are repeated as described above.

If it is determined that the item belongs to tolerance range 2, the tolerance range is raised by one rank, and the recognition and extraction work for the article based on tolerance range 2 is performed in the same manner as shown in steps 5 to 9.

Further, if it is determined that the object belongs to the tolerance range 1, the tolerance range is raised by two ranks, and a series of recognition extraction operations based on the tolerance range 1 are performed as shown in steps 1 to 4.

As described above, according to the recognition method of the present invention, at least three or more recognition tolerance ranges are set for the reference dimensions of the object to be recognized, and the recognition tolerance range is always lowered from the strictest recognition tolerance range in order. In order to recognize whether or not there are items that can be removed, the parts 12 stacked in bulk on the pallet 10 are taken out in order from the parts 12 that are stacked on top and have a small slope, that is, the parts that can be taken out reliably. It will be understood that this makes it possible to carry out the work in an extremely rational manner, and that errors in taking out parts are greatly reduced.

Further, according to the present invention, when a situation arises in which the parts 12 cannot be easily taken out from the pallet 10, that is, in the embodiment, when parts that can be taken out only based on the tolerance range 3 can be recognized. Since the robot hand 24 can be used to stir the pile of parts on the pallet 10 to create a package that is easy to take out, it is possible to recognize and take out the items present on the pallet 10 reliably and efficiently. In particular, it is possible to achieve an excellent effect in that the recognition and extraction work can be performed until the parts 12 are completely removed from the pallet 10.

Therefore, by applying the method of the present invention to a robot system such as that shown in FIG. 3, the time cycle for supplying parts to the conveyor 26 can be shortened, and the unloading errors can be reduced to make the unloading cycle constant. As a result, when supplying parts to the production line via the conveyor 26,
It becomes easy to set extra buffers for synchronization with post-processing and allowances for takeout capacity, making it possible to significantly contribute to cost reduction on the production line.

Furthermore, according to the present invention, there are almost no parts 12 left on the pallet 10 that cannot be taken out, so in a robot system as shown in FIG. 3, for example, the ability to supply parts to the conveyor 26 can be greatly increased. becomes possible.

In this embodiment, shaft parts stored in bulk on a pallet 10 are used as an example to be recognized. However, the present invention is not limited to this, and can be applied to recognition work for other objects having shapes other than this. can be done similarly.

Furthermore, in this embodiment, the case where the recognition tolerance range for an article is provided in three stages is explained as an example, but the present invention is not limited to this, and it is also possible to provide four or more recognition tolerance stages as necessary. be.

[Effects of the Invention] As explained above, according to the present invention, it is possible to accurately recognize articles stored in bulk in a container and efficiently take them out using a robot hand. It can be used for a wide range of purposes, and is particularly effective in automating manufacturing processes.

[Brief explanation of the drawing]

FIG. 1 is a flowchart showing a preferred embodiment of the article recognition method according to the present invention, FIG. 2 is an explanatory diagram of the recognition tolerance range used in this embodiment, and FIG. 3 is a flow chart showing a preferred embodiment of the article recognition method according to the present invention. An explanatory diagram of a robot system for taking out parts from a bulk stack; Figure 4 is a flowchart showing the general operation of the robot system shown in Figure 3; Figure 5 is an explanatory diagram showing an example of an article to be taken out; Figure 6 is an image. An explanatory diagram showing how the length of the object recognized through processing changes depending on the distance from the TV left camera. Figures 7 to 9 show parts stored in bulk on the pallet shown in Figure 3. It is an explanatory diagram showing a state. 10 ... Pallet 12 ... Parts 14 ... TV left camera 6 ... Image processing device 18 ... Robot control device 20 ... Robot 24 ... Robot hand 26 ... Conveyor 28 ...・Arm Fig. 3 Fig. 4

Claims (1)

[Claims] (1) Objects stored in bulk in a container are photographed using a TV camera, and based on the video signal, a desired object that can be taken out from the container is recognized according to the following steps, and the recognized object is transferred to a robot hand. A method for recognizing an object, the method comprising repeatedly performing a series of operations in which an object is used and taken out to a predetermined position outside a container. (b) Image processing of video signals is performed by setting recognition ranges in at least three levels: the strictest tolerance range, the relatively loose tolerance range, and the loosest tolerance range, with respect to the standard dimensions of the object to be recognized. A first step in which the recognition work is performed while sequentially lowering the recognition tolerance range until it is recognized that there is an object that can be taken out from the container by the robot hand, based on the object dimensions obtained by the robot hand. (b) In the first step, if it is determined that there is an object that can be taken out based on the strictest recognition tolerance, the object is taken out to a predetermined position outside the container using a robot hand, and then the Returning to the first step, the same operation is repeated, and if it is determined that there is an object that can be taken out based on the relatively loose recognition acceptance criteria, the object is moved to a predetermined location outside the container using a robot hand. At the same time, it is determined which of the above-mentioned tolerance ranges the object dimensions obtained during the recognition process of the taken-out object correspond to, and then the process returns to the first step and the next step is performed from the determined tolerance range. The second step is to restart the recognition task for the object. (c) In the first step, if it is recognized that there is an object that can be taken out based on the loosest recognition tolerance, a robot hand is used to take out and stir the object, and the removal and stirring It is determined which of the above-mentioned tolerance ranges the object dimensions obtained during object recognition work fall under, and then the process returns to the first step and restarts the recognition work for the next object from the determined tolerance range. and a third step of terminating the object recognition task if it is recognized in the first step that there is no retrievable object even based on the loosest recognition tolerance range.