JPH0624424A - Facing/p0sitioning method for container - Google Patents

Abstract

PURPOSE:To make it possible to perform the facing/aligning of various types of containers easily by a method wherein information an a sensed image of a container which is carried facing an optional direction is collated with image information which is previously registered, to judge the facing direction of the container, and the container is turned to the direction for a purpose. CONSTITUTION:An electronic camera 11 is arranged at an angle of view by which the picture of a major part of a container 10 can be taken, and an illumination part 12 is arranged in the vicinity of the electronic camera 11. When a container sensor 13 detects that the carried in container 10 is mounted on a rotary table 19, an image signal from the electronic camera 11 is registered in an input image memory 15 by the output of an operation control unit 14. Then, the degree of conformity of the input image is calculated at an image recognition operation unit 17 based on the registered image of the total periphery of the container 10 which is previously registered in a register image memory 16, and a position at which the degree of conformity becomes largest is retlieved. Then, the retlieved position number is transmitted to a rotation driving control unit 18, and the rotary table 19 is rotated by a pulse motor 20 to make the front surface of the container 10 face the front surface of the electronic camera 11, and the alignment is completed.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a container orientation alignment method, and more particularly to a container orientation alignment method for aligning a container such as a can supplied in an arbitrary orientation by rotating the container in a predetermined direction. .

[0002]

2. Description of the Related Art Conventionally, in the manufacture of food and household goods,
It has a manufacturing process in which a lid is attached to the container or an outlet is provided. In this manufacturing process, the container is rotated in a predetermined direction to align the ejection port with the printed pattern or logo mark on the outer peripheral surface of the container according to the function and aesthetic purpose of the container, and then the alignment is performed. There is. When such a container is rotated in a desired predetermined direction for alignment, a register mark which is easily sensed by a photoelectric sensor is provided at a specific location on the outer peripheral surface of the container. Then, the rotation of this container is scanned to detect the position of the register mark, the rotation is stopped, and the position is adjusted to a predetermined position. As the register mark, besides a mark provided as a sensing mark for a photoelectric sensor, a characteristic portion in a printed pattern or a characteristic protrusion or depression may be used.

[0003]

However, in the above-described conventional container orientation alignment method, the position, color, or contrast of the register mark is changed for each product type and model of the container. Therefore, it is necessary to reset the position and detection sensitivity of the sensor each time this change is made. In addition, due to the product design of the container, it may not be possible to attach the register mark to the container. In this case, since it is difficult to automate the orientation position of the container, the alignment is performed manually. In this way, when aligning the conventional container in the intended predetermined direction, in either case the productivity decreases,
Particularly in the recent small-lot production of a wide variety of products, there is a problem that an increase in the line stop time is a major factor of a decrease in productivity, and solving these problems is an issue for improving productivity.

The present invention has been made in view of the above-mentioned problems, and a container such as a can that has been supplied in an arbitrary direction is accurately and automatically rotated in a predetermined predetermined direction to perform alignment. The purpose is to provide a possible container-oriented positioning method. In addition, there is no need to modify the sensor position by changing the type and model of the container, which is conventionally done, and it is possible to shorten the mold changing time and significantly improve productivity. To aim.

[0005]

In order to achieve the above object, the container orientation alignment method of the present invention pre-registers image information of the entire circumference of the container whose orientation is to be aligned, and subsequently, an arbitrary The container is supplied in the orientation to obtain the captured image information of the container, and the registered image information is compared with the captured image information to determine the orientation of the supplied container. Based on this, the container is rotated in a predetermined predetermined direction for alignment. Then, the registered image information and the photographed image information are collated by the normalized two-dimensional correlation method that determines the area on the registered image having the maximum correlation value as the same portion as the input image, or the registered image. The information and the photographed image information are collated, and the orientation of the supplied container is determined by a one-dimensional cross-correlation number.

Further, the image information of the entire circumference of the container to be registered is obtained from one side of the image information created by connecting a number of distortion-corrected screens, a plurality of subdivided screens, or a predetermined height. The data is one-dimensional brightness data obtained by photographing the container to be registered. Further, the image information of the container supplied is separated from the orientation position of this container, and when the amount of alignment rotation is more than half a rotation, the container is rotated by reverse rotation to achieve the target position. The position is aligned in the direction.

[0007]

According to the container orientation alignment method of the present invention, a pre-registered image of the entire circumference of the container is compared with a partially photographed image of the container whose orientation position is inspected for alignment. , The direction of the container is judged, and based on this judgment data, the container supplied in any direction is accurately and
The position is automatically adjusted by rotating in a predetermined direction.

Generally, in the case where the container is manually positioned by visual judgment, the pattern of the entire circumference is preliminarily stored even if the entire circumference of the container is not visually observed by some training, and the container is based on this memory. It is possible to discriminate an arbitrary direction of, and correct it to a predetermined direction, that is, a predetermined direction. Here, the image recognition processing technique based on such a concept is adopted, and the container is accurately and automatically rotated in a predetermined predetermined direction for alignment. That is, the input image, which is a partial image photographed from an arbitrary direction, is searched from among the registered images that are images of the entire circumference of the container registered in advance,
Recognizing the degree of coincidence with that portion, the position number on the registered image assigned to the coincident portion is detected as the angle at which the container should rotate, and the container is rotated by this angle to move in the desired direction. Make a match.

FIG. 1 shows a schematic structure relating to the orientation and alignment of the container. In FIG. 1, the orientation image is inspected, the container 1 to be aligned is photographed by the electronic camera 3, the input image 4 is superimposed on the registered image 2, and the input image 4 is moved pixel by pixel to determine the degree of coincidence. Ask for. And
The position number 6 in the registered image 2 of the pattern matching recognition area 5 in the registered image 2 having the highest degree of coincidence is determined to be the direction in which the container 1 is currently facing. In FIG. 1, it is determined that the number is “8” in the position number 6, and thereafter, the position is automatically rotated in the target predetermined direction for alignment.

[0010]

EXAMPLES Next, examples will be described in detail. First, the first embodiment will be described with reference to FIG. In FIG. 2, the container 10 is placed on the turntable 19 in any orientation. The electronic camera 11 is arranged at an angle of view capable of photographing the main part of the container 10. That is, the comparison axis (X) is arranged orthogonal to the rotation axis (Y) of the turntable 19. Further, the container 10 is illuminated by an illumination unit 12 arranged near the electronic camera 11. When the container sensor 13 detects that the container 10 is placed on the turntable 19, the arithmetic control unit 14 to which the signal is input registers the image signal from the electronic camera 11 in the input image memory 15. Then, the image calculation process described below is performed under the control of the calculation control unit 14.

In this image calculation processing, first, a registered image of the entire circumference of the container 10 which is registered in the registered image memory 16 in advance is called, the image recognition calculation section 17 calculates the degree of coincidence with the input image, and this Search for the position with the highest degree of coincidence.
The registered image registered in the registered image memory 16 includes the container 1
A position number corresponding to the direction of 0 (position number 6 in FIG. 1) is assigned starting from the front side, and the searched position number is sent to the rotation drive control unit 18. The rotation drive control unit 18 reversely rotates the pulse motor 20 by an angle corresponding to this number and returns the turntable 19 so that the front surface of the container 10 faces the front surface of the electronic camera 11. The container 10 thus aligned in the target predetermined direction is moved from the front of the electronic camera 11 to the next stroke by the container transfer control unit 21 which is operated by the operation end signal from the rotation drive control unit 18. A lid or the like oriented by another device (not shown) is attached. The electronic camera 11 in the first embodiment uses a conventional CCD monochrome type. The number of pixels is 512 × 512. The illumination unit 12 uses a halogen bulb of 50 W to illuminate the photographing position in the center of the container 10 with substantially uniform brightness. A conventional eddy current type proximity switch is also used as the container sensor 13 that obtains a detection signal of the start of image capturing.

The configuration of the apparatus for implementing the above method is not limited to the above. For example, if there is only a large pattern, the electronic camera 1 using a CCD with a smaller number of pixels
Even if the value is 1, it may be determined in consideration of the image content of the container 10 and its property. A color electronic camera can be used instead of the CCD monochrome type electronic camera 11, but in the first embodiment, an appropriate light and dark pattern is sufficient, and it is not particularly necessary. The eddy current sensor of the container sensor 13 is not suitable for plastic or paper containers. Therefore, a photoelectric sensor or the like is appropriately selected and used depending on the material of the container.

Next, an image input process and its registration process in the first embodiment will be described. In FIG. 2, the image signal of the container 10 obtained by the electronic camera 11 is digitized and stored in the input image memory 15. This image-capturing range, that is, the angle of view was set to a range of about 90 degrees obtained by dividing the entire circumference of the container into about four. For a container with a diameter of 50 mm, such as general aerosol cosmetics, the size of the angle of view is about 40 mm on the outer circumference of the container, and you can enter about 3 to 5 characters such as a brand name or a logo mark indicating the product name. , Its characteristics can be fully expressed. Further, the lateral resolution of the image for expressing the characteristics of the image without deteriorating the image does not need to be so high. Here, the input image memory 15
Capacity of 128 × 128 pixels, 1 pixel brightness contrast is 8
It was a bit. In this case, the number of pixels is determined by the density of the pattern of the container 10 to be handled. When the pattern is large and the contrast is clear, a smaller number of pixels and gradation are sufficient. In such a case, it is possible to reduce the size of the image at the stage of image calculation to speed up the process. For example, in the case of a large pattern of black and white, 16 × 16 pixels have a 1-bit gradation, that is, it is possible to sufficiently search and recognize an image even with a binary image.

Further, the image recognition calculation process in the first embodiment will be described. First, it is recognized which part of the registered image (the registered image 2 in FIG. 1) is the input image (the input image 4 in FIG. 1). In this case, if there is no extreme difference in the relative brightness of both images, it is ignored and only the shape of the images is focused. In order to perform the recognition of the degree of coincidence as an image recognition calculation process, a search is performed at a place where the difference in brightness between corresponding pixels is the smallest or the degree of similarity is the highest.

In the first embodiment, the normalized two-dimensional correlation method is used in which misrecognition is small even when there is a difference in brightness or out-of-focus between images. That is, the area on the registered image with the maximum correlation value is determined to be the same portion as the input image,
The position number assigned in advance on the registered image is output as a numerical value to the rotation drive control unit 18. When the correlation value indicating the similarity of images is expressed as 100 when all the pixels are the same and 0 when they are completely different, it is the same image unless there is a large distortion, a difference in brightness or a focus blur. A value of 70 or more is shown. When the maximum value of the obtained correlation value is less than this, it is considered that different kinds of products are mixed, and the container 10 is excluded from the manufacturing line by an excluding mechanism (not shown).

Further, the image registration processing in the first embodiment will be described. Prior to work, the container 10
The developed image of the entire circumference of must be registered. There are two possible ways to do this. First, the first is a method of photographing and registering a product using the electronic camera 11 for capturing an input image, and the second is a method of using an original image used for plate making for printing the container 10. The second method is limited to those having image-distortion-free features and having a printed pattern all around. Therefore, it may not be applicable when the shape of the handlebar is characteristic. In addition, in order to register the image of the entire circumference of the container 10, it is necessary to specify the front part of the container 10 and also to shoot the images of the rear surface in succession. I shot it, corrected the distortion of the image in the peripheral area, and joined it together.

Next, the distortion correction processing will be described in detail. To simplify the description, a case where the body is photographed from the side when the container 10 has a cylindrical shape will be described. There is no distortion on the barrel surface on the optical axis passing through the center of the container 10, and on the surface off the optical axis, the distortion increases in inverse proportion to the cosine of the angle with the normal. To correct the image distortion, for example, as shown in FIG. 2, a black vertical line that is divided into 32 equal parts along the generatrix of the outer periphery of the white container 10 is photographed with an electronic camera for photographing the input image. A thinning process is performed, and a value necessary for changing each section imaged at unequal intervals at this time so as to be equal intervals is obtained as a correction coefficient, and the correction pixel is divided into eight equal parts. This is done by multiplying by the number and matching the number of corrected pixels in each section. As an example, when the diameter of the container is 53 mm and the distance from the center of the container to the camera lens is 200 mm, the number of pixels in each section when the diameter of the barrel is photographed in full screen, and when it is corrected to 16 pixels each The correction coefficient of is shown in FIG. In addition,
The image distortion may be corrected using a geometrically calculated correction coefficient. For example, an image of a cylindrical body taken from the side has no distortion on the can barrel surface that intersects with the line of sight passing through the center of the container, and 1 / COS of the angle with the normal line on the surface that intersects with the line of sight deviating from the center of the container. The distortion increases in proportion to. Therefore, the angle may be geometrically calculated from the diameter of the container, the distance of the camera, and the angle of view for correction.

Further, the image connecting process in the first embodiment will be described. The range of the image obtained by one shot is 1/4 of the entire circumference as described above, and to make this the image of the entire circumference, the image is shot several times and connected.
The concept is shown in FIG. In FIG. 4, first, the first photographing is performed, the distortion is corrected as described above, and then the image is registered in the registered image memory 16 in FIG. Next, when performing the second shooting, about 1/4 of the angle of view is overlapped with the first angle of view, and a vertically long image with a small pixel width of that part is subjected to the normalized two-dimensional correlation method. Perform a location search and connect to the first image. Similarly, the third and subsequent images are also connected to the previous image one after another. If you repeat this operation, the same part as the first image will come around. This also searches whether the beginning of each image is a part of the first image,
The registration of the image is completed by detecting that the container 10 has rotated once. The registered images are cyclically linked in the registered image memory 16. That is, the registration is performed as if an image of the body of the actual container 10 was imagined. This work is similar to the case of cutting and pasting finely cut photographs to create a panoramic photograph. If the overlap between the adjacent images is insufficient in this shooting, the matching part cannot be searched, and a message of insufficient overlap is issued to prompt reshooting. In an actual device, a registration mode in which a pulse motor used later for the alignment operation is automatically rotated by a predetermined angle is provided in order to perform the registration operation efficiently without excess or deficiency of image overlap.

Next, a second embodiment will be described. Although the image distortion correction performed in the first embodiment enables highly accurate recognition, the processing time becomes long due to the distortion correction. Both the registered image and the input image are images with a narrow range of view angle in which distortion can be ignored, and even if distortion correction is omitted, a sufficiently accurate image recognition search can be performed. A method that does not perform the image distortion correction will be described with reference to FIG. In FIG. 5, if the entire circumference of the body of the container 54 is divided into a large number of small images of ⅛ to 1/16, the image distortion will be within a few percent. Therefore, as the registered image in the registered image memory 50 without correction, the input image in the input image memory 51 has an angle of view that is slightly larger than twice the one image in the registered image, and the registered image is stored in the registered image. The image recognition unit 5 determines which one is included.
I searched in 2. The relationship between the image to be searched and the image to be searched is opposite to that in the first embodiment, but there is no difference in the basic configuration. That is, the electronic camera 53, the container 54, and the turntable 5
The operations of 5, the pulse motor 56, the rotation drive unit 57, and the container transport control unit 58 are the same as those in the first embodiment.

By registering the divided images, the container 5
A large registration area for registering the entire circumference of the body of No. 4 as one continuous image is not required, and a program for connecting images is not required, and the overall configuration and processing can be simplified. It is appropriate to divide the registered image into about 8 to 16. If it is less than that, distortion in the peripheral area increases and correction is required. Above that, depending on the print design, there is only one plain screen with no patterns or characters. Inconvenience occurs because a plurality of places occur. Generally, the solid portion of the entire circumference of the body of the container 54 has about 1/20 round, and if it is subdivided less than that, the risk of misrecognition increases. Although it seems that the accuracy of recognition is reduced due to the absence of distortion correction, on the contrary, the accuracy was higher without multi-division correction. In other words, in a wide range of view angles, the number of pixels that a small character occupies in an explanatory text is small, so it was difficult to recognize a single character, but in a narrow range of view angle, the number of pixels occupied by one character increases. , The correlation value changes greatly even with a small difference in one letter.

In the first embodiment and the second embodiment, the input image is photographed and the orientation of the container 54 is aligned on the same rotary table. However, these can be performed on different stages. . In other words, if the image processing is completed while the container 54 that has been imaged is conveyed to the next rotary stage, the processing time due to the division of labor can be shortened and the processing speed can be increased.
Further, when the amount to be rotated is equal to or larger than half the rotation of the container 54, the container 54 can be rotated in the reverse direction so that the container 54 can be operated at a higher speed.

Next, a third embodiment will be described. The first and second embodiments are so-called advanced methods using a two-dimensional image, but when the container is a simple cylindrical shape, similar alignment can be performed even with one-dimensional simple light-dark value data. I can. The concept is shown in FIG. In FIG. 6, the input data 63 detected while rotating the container 60 by a photoelectric sensor 62 having a predetermined height, which starts from an arbitrary position, and the front surface at the same height as the input data 63 detected in advance. The detected shift amount X65 with respect to the input data 61 of the brightness change along the circumference from the starting point is obtained, and the alignment is performed based on this shift amount X65. The shift amount X65 is the shift amount that maximizes the correlation value by calculating the cross-correlation function between the two by real-time correlation. The photoelectric sensor 62 is a normal photodiode type,
A focus having a diameter of about 1 cm was used. The size of the focal point is not particularly important, but if the focal point is too sharp, the risk of misjudgment increases due to the positional fluctuation of the photoelectric sensor 62, and if it is too large, the alignment resolution tends to decrease due to data blur. Correlation is registration data 61 and input data 63
Although it is a simple 8-bit 128 words, it requires only a few milliseconds since it is a one-dimensional small amount of data, which is negligible compared to the mechanical operation time for alignment. Although the amount of data is small, the entire circumference of the container 60 is divided into 128, so the resolution at the outer periphery of the container 60 having a diameter of 53 mm corresponds to 1.3 mm, and there was no problem in the alignment accuracy. Even less 64 words (w
ord) is practically sufficient. Since this simple method uses the data of the entire circumference of the container 60 for both the registration data and the input data, it is not possible to judge the front direction only by looking at an arbitrary part as in the first and second embodiments. It is necessary to turn 60 one rotation or more each time to input data.

If the installation state of the photoelectric sensor 62 fluctuates, a determination error will occur. Although the correlation calculation unit and the registration unit and the like in the first, second and third embodiments are composed of semiconductor logic circuit elements, they can be processed by a general-purpose microcomputer as well. Further, an integrated circuit dedicated to image processing may be used to perform high-speed calculation. A conventional sequence control device may be used to control the container transport unit and the discharge device, and a rotation drive unit may be used that is capable of reverse rotation together with a normal pulse motor.

As the registered image, an image of the entire circumference of at least a part of the container body may be used. That is, as the registered image, an image of the entire circumference with a constant width in the axial (height) direction of the container body is used, and as an input image, an arbitrary part obtained by photographing with this electronic camera with this constant width. You may make it use the image of. Further, each device in the above-mentioned embodiments is not limited to this, and may be appropriately changed depending on the nature of the image of the inspection object. For example, if the feature is that it is printed with fluorescent ink, illumination with an ultraviolet lamp is effective, and even if there is a pattern with high similarity in visible light, if there is a difference in infrared absorption and reflection, the infrared camera The X-ray camera is effective when the use is effective and the internal structure is characteristic. Also,
A neurocomputer or the like can also be used for image retrieval.

[0025]

As described above, according to the container orientation alignment method of the present invention, a container such as a can supplied in an arbitrary direction is rotated accurately and automatically in a predetermined direction. It has the effect of enabling alignment. In addition, it is not necessary to modify the sensor position by changing the type and model of the container, which was conventionally done, and it is possible to shorten the time for changing the mold, greatly improving the productivity and further improving the appearance of the register mark. The thing that obstructs is also unnecessary, the degree of freedom of design is improved, and the product value is improved.

[Brief description of drawings]

FIG. 1 is a block diagram for explaining an outline of an embodiment of a container orientation alignment method of the present invention.

FIG. 2 is a block diagram showing a configuration according to a first embodiment of a container orientation alignment method of the present invention.

FIG. 3 is a diagram for explaining a distortion correction process in the first embodiment.

FIG. 4 is a diagram for explaining an image connection process in the first embodiment.

FIG. 5 is a block diagram showing a configuration according to a second embodiment.

FIG. 6 is a block diagram showing a configuration according to a third embodiment.

[Explanation of symbols]

 1 Container 2 Registered Image 3 Electronic Camera 4 Input Image 5 Pattern Matching Recognition Area 6 Position Number 10 Container 11 Electronic Camera 12 Illumination Unit 13 Container Sensor 14 Operation Control Unit 15 Input Image Memory 16 Registered Image Memory 17 Image Recognition Operation Unit 18 Rotation Drive Control unit 19 Rotating table 20 Pulse motor 21 Container transport control unit

Claims (8)

[Claims] 1. Image information of the entire circumference of a container for orientation alignment is registered in advance, and captured image information of a container supplied in an arbitrary orientation is obtained, and the registered image information and A container orientation alignment method for determining the orientation of a supplied container by collating with the photographed image information, and rotating the supplied container in a target predetermined direction based on this determination to perform alignment. 2. The collation between the registered image information and the photographed image information is performed by a normalized two-dimensional correlation method in which the area on the registered image having the maximum correlation value is determined to be the same portion as the input image. The described container orientation alignment method. 3. The container orientation alignment method according to claim 1, wherein the registered image information is collated with the photographed image information, and the orientation of the supplied container is determined by a one-dimensional cross-correlation coefficient. . 4. The container-orienting alignment method according to claim 1, wherein the image information of the entire circumference of the container to be registered is image information of one surface created by connecting a number of distortion-corrected screens. 5. The container orientation alignment method according to claim 1, wherein the image information of the entire circumference of the container to be registered is composed of a plurality of subdivided screens. 6. The container orientation alignment method according to claim 1, wherein the image information of the entire circumference of the container to be registered is one-dimensional brightness / darkness value data obtained by photographing the container to be registered from a predetermined height. 7. The container orientation alignment method according to claim 1, wherein the image information of the container supplied and the orientation alignment location of the container are separated. 8. The container-oriented alignment method according to claim 1, wherein when the amount of alignment rotation is more than half a rotation, the container is rotated by reverse rotation to perform alignment in a target predetermined direction.