JP2019105611A - Whole circumference image generation device and whole circumference image generation method - Google Patents

Abstract

To provide a whole circumference image generation device and a whole circumference image generation method capable of relatively easily generating a whole circumference image made to correspond to circumference-directional position information on an article.SOLUTION: In a label inspection device, an image acquisition processing part acquires photographic images of a container one after another (S11). A model acquisition part acquires a model based upon a pat of a container in a first photographic image (S13). A storage part performs matching processing using the model on a next photographic image, finds a front position of the container based upon the position information of a matching position and the position of the model when the match is obtained, stores the partial image at the front position of the container in the next photographic image in a memory part associatively with the position information (S14-S19). Then a model acquisition part repeats processing to acquire a model based upon a part of the container in a matching photographic image and the storage part repeats storage processing to store a partial image through matching processing using at least a latest model (S14-S20).SELECTED DRAWING: Figure 4

Description

  The present invention relates to an all-round image generating device and a all-round image generating an all-round image over the entire circumference of at least a part of an article in association with the circumferential position of the article using a plurality of images obtained by photographing the article. It relates to the generation method.

  BACKGROUND Conventionally, a container inspection apparatus for inspecting an inspection object such as a label attached to the outer surface of an article such as a container is known. For example, the container inspection apparatus described in Patent Document 1 acquires a partial image including a label by imaging the container and storing a storage unit that stores an all-round image indicating the entire outer surface of the non-defective label attached to the container. And an imaging unit. Then, the container inspection apparatus determines the position of the label part in the partial image using the all-round image, the position specifying unit, and the visual inspection of the part of the label according to the specified position, the all-round image and the part And a container inspection unit that compares the image with the image.

  Also, for example, Patent Document 2 discloses a container orientation detection device that detects the orientation of a container. A container direction detection apparatus is provided with the imaging device which has a camera which image | photographs the pattern of the outer peripheral surface of a container. The storage unit stores model registration data in which a plurality of symbol parts whose positions are different in the circumferential direction of the container are associated with position information as a model. The container orientation detection device performs matching processing for searching for a similar symbol portion similar to the model among partial images obtained by photographing the container by the camera, and detects position information of the similar symbol portion found and the similar symbol portion A container orientation detection unit is provided that detects the orientation of the container based on position information of the corresponding model.

JP, 2016-17810, A Unexamined-Japanese-Patent No. 2017-100796

  By the way, in the container inspection device described in Patent Literatures 1 and 2, it is necessary to store in advance in the storage unit the entire circumferential image to be compared with the partial image of the container in association with the circumferential position of the article. Further, in the container orientation detection device described in Patent Document 2, when creating a plurality of models, the image of the non-defective label which is the target for the operator to select the region to be the model is the circumferential direction of the article. It needs to be prepared in association with the position. For example, in the case where image data of an original (printed version) of a label can be obtained, it is possible to generate an all-round image associated with a circumferential position based on the original. However, when an original image can not be obtained, partial images of good containers taken with a camera are associated with the circumferential position and acquired for all the circumference, and the work of joining these partial images while correcting the position is performed manually. There is a need to do. For this reason, it is demanded to easily generate an all-round image correlated with positional information on the circumferential direction of the container. In addition, this kind of subject is common when articles | goods other than a container also perform various test | inspection and detection using the picked-up image image | photographed with the camera, and the perimeter image.

  An object of the present invention is to provide an all-round image generating device and a all-round image generating method capable of generating an all-round image in correspondence with a circumferential position of an article using a plurality of partial images obtained by photographing the article. It is to do.

Hereinafter, the means for solving the above-mentioned subject and its operation effect are described.
An all-round image generating device that solves the above-mentioned problem is a all-round image generating device that collects partial images obtained by photographing non-defective articles and generates an all-round image correlated with position information in the circumferential direction. An image acquisition unit for sequentially acquiring a photographed image obtained by photographing an article; and setting a front position of the article in one of the photographed images acquired by the image acquisition unit as a reference position; A storage unit that associates a partial image at a position with circumferential position information and stores the image in a storage unit, and a model based on a predetermined region of a part of the article in the one captured image acquired by the image acquisition unit A model acquisition unit for acquiring, the storage unit performs a matching process using the model on a captured image after the image acquisition unit has acquired the captured image after the acquisition source of the model; Position of the matching position The front position of the article based on the information and the position information of the model, and the partial image at the front position of the article in the subsequent captured image is stored in the storage unit in association with circumferential position information; If they do not match, storage processing is performed to shift to the next captured image, and the model acquisition unit acquires a model based on a predetermined region of a part of the article in the captured image at the time of the match. A plurality of partial images in which processing and the storage processing including the matching processing performed using at least the latest model, the partial processing associated with position information in the circumferential direction are collected at least corresponding to the entire circumference The above-mentioned all-round image is generated repeatedly. According to this configuration, it is possible to relatively easily generate the entire circumference image associated with the position information of the article in the circumferential direction.

  In the all around image generation device, the accumulation unit accumulates the partial image for each area to which a representative position obtained by dividing the circumference of the article is divided, and the area to which the current front position at the time of matching belongs first. When the partial image of is stored, if the current front position is closer to the representative position of the area than the front position of the previous partial image, the partial image to be accumulated in the area is It is preferable to update from the partial image to the current partial image. According to this configuration, the partial image accumulated during progress of the accumulation processing is updated to a more appropriate partial image, so it is possible to generate a more accurate all-round image with less distortion.

  In the all-around image generation device, the model acquisition unit is configured to move in one direction in the circumferential direction of the article with respect to the position in the circumferential direction of the model acquired first, and to move in the other direction. Preferably, the plurality of models are acquired at both positions, and the storage unit performs the matching process using the plurality of models.

  According to this configuration, matching processing is performed using a plurality of models located at positions moving in one direction in the circumferential direction of the article and in the other direction with respect to the position of the model acquired first. Frequency of Therefore, it is possible to generate an all-round image associated with the position information at relatively high speed.

  In the all-around image generation device, the model acquisition unit may set a plurality of the models such that positions in the circumferential direction of the model sandwich an empty area in which the partial image is not accumulated between the model and another model. Preferably, the storage unit performs the matching process using a plurality of the models. According to this configuration, the matching process is performed using a plurality of models in which positions in the circumferential direction of the model sandwich the empty region where partial images are not stored between other models. An all-round image associated with information can be generated relatively quickly.

  In the all-round image generation device, the partial image is not stored if the storage unit can not complete the all-round image even if the model acquisition process and the storage process are repeated until a predetermined condition is satisfied. The representative position of the unaccumulated area is acquired, and if there is an accumulated area at the position next to the unaccumulated area, the photographing which is the acquisition source of the partial image accumulated in the accumulated area Preferably, the partial image at a position corresponding to the representative position of the unaccumulated area is acquired from an image and stored in the unaccumulated area in association with position information.

  According to this configuration, even when the full circle image can not be completed, the full circle image associated with the position information can be completed using the existing captured image. Therefore, predetermined processing such as model registration processing, article orientation detection processing, or article inspection processing can be performed using the entire circumference image.

  The all-round image generation device creates a plurality of models based on a plurality of image areas having different positions in the circumferential direction selected from the all-round images, and associates the plurality of models with circumferential position information. It further comprises a model registration unit to register. According to this configuration, since it is possible to relatively easily generate an all-round image which is an acquisition source of a model necessary for detecting the direction of an article, it is possible to easily acquire a plurality of models whose positions are different in the circumferential direction.

  The all-around image generation device performs a matching process on a captured image obtained by capturing an article whose direction is to be identified and has a design on the outer surface, using the model created by the model registration unit, and matches similar symbol parts And a position specifying unit for specifying a front position of the article in the photographed image or a predetermined position in the circumferential direction of the article based on the position information of and the position information of the model. According to this configuration, it is possible to detect the orientation of the article consisting of the front position of the article or the predetermined position in the circumferential direction, and it is possible to relatively easily generate an all-round image necessary for detecting the orientation of the article. .

  The all-round image generating device is a non-defective image generating unit that generates a plurality of non-defective images that are partial images in the circumferential direction from the all-round image corresponding to all the circumferences and an inspection that And a position specifying unit for specifying a front position of the article in the photographed image obtained by photographing the article to be inspected, the inspection unit further comprising: an inspection image in the photographed image; The presence or absence of a defect of the article is inspected based on a comparison result comparing the non-defective image corresponding to the front position identified. According to this configuration, it is possible to inspect the presence or absence of a defect of an article, and to easily acquire a non-defective image necessary for the inspection.

  In the above-mentioned whole circumference image generation device, the article is preferably labeled on the outer surface, and the inspection unit preferably inspects the defect of the label. According to this configuration, it is possible to inspect the defects of the label attached to the article.

  An all-round image generating method for solving the above-mentioned problem is a all-round image generating method for collecting a partial image obtained by photographing a non-defective article and generating an all-round image correlated with positional information in the circumferential direction. An image acquisition step of sequentially acquiring a photographed image obtained by photographing an article, and model acquisition of acquiring a model based on a predetermined region of a part of the article in the one photographed image acquired in the image acquisition step Performing matching processing using the model on the captured image obtained after the captured image of the acquisition source of the model in the step of acquiring the image in the image acquiring step; Based on the position information of the model, the front position of the article is determined, and the partial image at the front position of the article in the subsequent captured image is stored in the storage unit in association with the circumferential position information. And an accumulation step of performing accumulation processing to shift to the next photographed image, and acquiring the model based on a predetermined region of a part of the article in the photographed image when the matching is performed; The accumulation step of performing the accumulation process including the matching process performed using at least the latest model is repeated until at least a plurality of partial images corresponding to circumferential positions are collected. And generate the all-round image. According to this method, it is possible to relatively easily generate the entire circumference image associated with the position information of the article in the circumferential direction.

  According to the present invention, it is possible to relatively easily generate an all-round image associated with position information of the article in the circumferential direction.

BRIEF DESCRIPTION OF THE DRAWINGS The model top view which shows the label inspection apparatus in 1st Embodiment. The front view which shows the container with a label. FIG. 2 is a block diagram showing an electrical configuration and a functional configuration of the label inspection device. 5 is a flowchart showing an all-round image generation processing routine. A panoramic view of the entire circumference of the label. The schematic diagram explaining the first process in all-rounds image generation processing. (A), (b) is a schematic diagram explaining the process which acquires a partial image. The schematic diagram explaining the correction arithmetic processing which acquires the front position (front angle) of a container. FIG. 5 is a schematic view showing a position storage unit, a first storage unit, and a second storage unit in a storage unit. FIG. 5 is a schematic view showing a partial image stored in a first storage unit. The schematic diagram explaining the 2nd process in a perimeter image generation process. (A), (b) is a schematic diagram explaining the process which acquires a partial image. The figure of the screen which shows the progress of an all-round image generation process. The schematic diagram which shows a perimeter image. The schematic diagram explaining the model for the matching process produced from a perimeter image. The schematic diagram which shows the non-defective goods image data for a test | inspection produced from a perimeter image. The flowchart which shows a model registration process routine. The flowchart which shows label inspection processing routine. The schematic diagram explaining the comparison process which compares a test | inspection image and a good quality image. The schematic diagram explaining the two-way system which adds the model in 2nd Embodiment. (A), (b) is a schematic diagram explaining the stepping stone system which adds a model. 10 is a flowchart showing an all-round image generation processing routine according to a second embodiment.

First Embodiment
Hereinafter, a label inspection device as a first embodiment in which an all-round image generation device is embodied will be described with reference to the drawings.

  The label inspection apparatus 11 shown in FIG. 1 photographs the container 20 as an example of the article shown in FIG. 2 which is placed on the conveyor 13 constituting the conveyance apparatus 12 as an example of the conveyance section and is conveyed. The label 22 attached to the outer peripheral surface 21 of the container 20 is inspected to inspect the presence or absence of the defect of the label 22. The inspection contents include defects (defective points) such as tearing of the label 22, adhesion of foreign matter, contamination, and label contraction failure. The label inspection apparatus 11 also inspects positional deviation (height deviation) of the label 22 with respect to the container 20 in the height direction.

  The label inspection device 11 according to the present embodiment generates an all-round image AG (see FIG. 14) of the label 22 attached to the container 20 by using a plurality of photographed images SG (FIG. 6, FIG. 11, etc.). By including the circumferential image generation unit 62 (see FIG. 3), the image processing apparatus functions as an all-round image generation device. The all around image AG is used to create each data of the model M (FIG. 15) and the non-defective item image RG (FIG. 16) used for the inspection of the container 20.

  The conveying device 12 shown in FIG. 1 conveys the container 20 to which the label 22 (shrink label) is attached to the outer peripheral surface 21 of the container 20 by a not-shown shrinker installed on the upstream side of the conveyor 13. The label 22 is attached in a state of being thermally shrunk and in close contact with the outer surface of the container 20 by being heated through a heating tunnel (not shown). The transport device 12 is provided with a known timing screw at a position upstream of the conveyor 13 and downstream of the insulin collector, and the container 20 is carried into the conveyor 13 with a predetermined interval in the transport direction X opened by the timing screw. Be done. The conveyor 13 is driven by the power of an electric motor, which is a power source of the conveyance device 12, and conveys the plurality of containers 20 at a predetermined conveyance speed with a predetermined interval. The containers 20 placed on the conveyor 13 are transported in any direction. If there is concern that the container 20 on the conveyor 13 may fall, the conveyor 13 is provided with a suction hole (not shown) for suctioning the bottom surface of the container 20, and the container 20 is subjected to negative pressure by the suction device. It is preferable to make it adsorb to the conveyor 13.

  As shown in FIG. 1, the label inspection device 11 generates an all-round image AG of the label 22 using the imaging device 30 for imaging the container 20 and the captured image obtained by imaging the container 20 by the imaging device 30. The controller 40 is provided with a generation process and an inspection process for inspecting the label 22. Further, the controller 40 creates the model M and the non-defective item image RG used for the inspection of the label 22 using the all-round image AG. The controller 40 also controls the conveyance device 12 and the imaging device 30.

  As shown in FIG. 1, the photographing device 30 photographs a plurality of containers 20 located at a predetermined photographing position located halfway along the conveyance path of the conveyor 13 from a plurality of directions (four directions in the example shown in FIG. 1). The camera 31 includes (for example, four) cameras 31 and an illumination device 32 for illuminating the container 20 to be photographed. The four cameras 31 shown in FIG. 1 are arranged at four positions shifted by about 90 degrees in the circumferential direction of the container 20 at the imaging position in plan view in the same drawing. The height position of each camera 31 is adjusted to the height of the label 22 of the container 20 transported on the conveyor 13. The camera 31 includes an optical system such as a lens and an imaging device such as a two-dimensional image sensor (such as a CCD imaging device or a CMOS imaging device).

  The illumination device 32 shown in FIG. 1 is disposed at a position where the specular reflection light of the light illuminating the container 20 does not directly enter the camera 31. The lighting apparatus 32 illuminates the container 20 from, for example, an obliquely upper position or an obliquely lower position with respect to the container 20. The lighting device 32 is, for example, a white light source, but light sources of colors other than white may be used as long as the label 22 can be inspected.

  The container sensor 33 shown in FIG. 1 detects the container 20 immediately before or when the container 20 on the conveyor 13 reaches the imaging position. The controller 40 instructs the four cameras 31 to perform imaging operations at the timing when the container sensor 33 detects the container 20. Based on this command, the four cameras 31 photograph the container 20 from the four directions substantially simultaneously. Therefore, an image of the label 22 around the entire side surface of the container 20 is taken by the four cameras 31.

  Each camera 31 sequentially outputs image data (image signal) of a photographed image obtained by photographing the label 22 of the container 20. Image data output from each camera 31 is input to the controller 40. The number of cameras 31 may be plural (for example, three or five) other than four, or may be one, as long as the configuration is capable of capturing the entire circumference of the container 20. For example, if the container 20 is configured to rotate (rotation) at least one rotation, or if the camera 31 is configured to rotate (revolution) with respect to the container 20, the entire circumference of the container 20 can be photographed even with one camera 31. . Further, in FIG. 1, the two cameras 31 arranged on both sides of the conveyance path of the conveyor 13 are disposed at different positions in the conveyance direction X, and the two cameras 31 have the containers 20 at different timings. It may be configured to shoot the

  The container 20 shown in FIG. 2 is, for example, a plastic container having a substantially cylindrical shape, and a label 22 is attached to the outer peripheral surface 21 thereof. The label 22 is attached to the container 20 at a predetermined height range in the height direction Z. A symbol 23 is drawn on the entire circumference of the label 22. In the example shown in FIG. 2, eight letters of upper case letters "A, B, C,..., H" of the alphabet are arranged in a line in a row. In addition, in FIG. 2, although it is set as an example of a simple symbol to simplify description, usually, the symbol 23 of the label 22 is a symbol such as a manufacturing company, a brand name or a logo mark, a commodity material / storage method -Includes one or more of a symbol part surrounded by a character string such as quality details of a manufacturing plant etc., a symbol part surrounded by a bar code, a symbol part such as an illustration or design (design) .

  Next, an electrical configuration and a functional configuration of the label inspection device 11 will be described with reference to FIG. As shown in FIG. 3, the controller 40 constituting the label inspection apparatus 11 incorporates a computer 50 composed of a chipset or the like mounted on a circuit board. Further, to the controller 40, the conveyance device 12, the camera 31 and the illumination device 32, which constitute the photographing device 30, the container sensor 33, the input device 41, and the display unit 42 are electrically connected. The controller 40 drives and controls the conveyor 13 at a constant speed by driving and controlling an electric motor (not shown), which is a power source of the conveyance device 12, and conveys the container 20 at a predetermined conveyance speed. Further, based on the instruction signal input from the input device 41, the controller 40 performs each process for generation of the all-round image AG, registration of the model M, generation of the non-defective item image RG, and label inspection. Further, the controller 40 performs display control to cause the display unit 42 to display an all-round image generation progress screen SC1 (FIG. 13), a model registration screen, a label inspection result notification screen (all not shown), and the like. The input device 41 and the display unit 42 may be provided in the control panel, or a part or all of the controller 40 is configured by a personal computer (PC), and the input device 41 is connected by a mouse and a keyboard connected to the PC. , And the display unit 42 may be configured by a monitor.

  As shown in FIG. 3, the computer 50 incorporates a CPU 51 (central processing unit) and a storage unit 52. The storage unit 52 is configured of a storage area of a hard disk, a non-volatile memory, or part of a RAM. The storage unit 52 includes an all-round image generation processing program (FIG. 4) for the CPU 51 to execute, a model registration processing program (FIG. 17), a non-defective image generation processing program (not shown) and a label inspection processing program A program PR including (FIG. 18) and the like is stored. The storage unit 52 also stores model data MD (see also FIG. 15), non-defective item image data GD (see also FIG. 16), feature image data FD and the like used in label inspection processing. In the present embodiment, the model data MD and the non-defective item image data GD are created based on the all-round image data. In addition, the feature image data FD is created based on the non-defective item image data GD.

  Further, the storage unit 52 shown in FIG. 3 is used as a storage area (storage area) to be used in the all-round image generation process, including the position storage unit 53, the first storage unit 54, the second storage unit 55, and the third storage unit. 56 is provided. The three storage units 53 to 55 are used to store data to be held in the process of the all-round image generation process, and the third storage unit 56 is an all-round image AG (FIG. 14) generated in the all-round image generation process. It is used to store image data. Each of the storage units 53 to 56 is configured by a predetermined storage area of a non-volatile memory or a RAM capable of rewriting data. In addition, the detail of each storage part 53-56 is mentioned later.

  The computer 50 illustrated in FIG. 3 performs an all-round image generation process, a label inspection process, and the like by the CPU 51 executing the program PR. The CPU 51 executes the program PR to function as the control unit 61, the all-round image generation unit 62, the model registration unit 63, the non-defective image generation unit 64, the feature image generation unit 65 and the inspection unit 66 shown in FIG. Do.

  The all around image generation unit 62 includes an image acquisition processing unit 71, a model acquisition unit 72, an accumulation unit 73, and an angle correction unit 74. The storage unit 73 includes a matching processing unit 75. The inspection unit 66 further includes an image acquisition processing unit 81, a position specifying unit 82, a label height inspection unit 83, and a detection unit 84. The detection unit 84 includes a comparison processing unit 85 and a determination unit 86.

  The control unit 61 illustrated in FIG. 3 is responsible for the entire control of the label inspection apparatus 11, and instructs each of the units 62 to 66 to perform predetermined processing and the like. Further, the control unit 61 performs setting of necessary data and reception of various instructions based on an operation signal from the input device 41 operated by the worker. The control unit 61 receives, for example, an instruction for an all-round image generation process, an instruction for a model registration process, and an instruction for a label inspection. The control unit 61 also controls the photographing operation of the camera 31 based on the detection signal of the container sensor 33. Furthermore, the control unit 61 performs display control to cause the display unit 42 to display various screens.

  Next, the all-round image generation unit 62 shown in FIG. 3 will be described. The image acquisition processing unit 71 performs processing for capturing and acquiring a captured image (image data) captured by the camera 31. The image acquisition processing unit 71 acquires four photographed images obtained by photographing the non-defective container 20 from four directions from the four cameras 31. The image acquisition processing unit 71 grasps the position of the container 20 in the image, and acquires an image of a predetermined area in which the container 20 is positioned at the center as a photographed image. In the present embodiment, the imaging device 30 and the image acquisition processing unit 71 constitute an example of an image acquisition unit.

  The model acquisition unit 72 acquires a model from the captured image. In this example, for example, a part of the image MG in a range of a predetermined angle in front of the container 20 in the captured image is acquired, and the features of the part of the image MG are extracted to acquire the model MF. As shown in FIG. 6, the model acquisition unit 72 acquires a part of the image MG in the front of the container 20 in the captured image SG, and extracts the feature (shape or brightness) of the part of the image MG. Get model MF. In the present embodiment, in order to use the pattern 23 of the label 22 of the container 20, a part of the image MG is an image including the front portion of the label 22. The width angle of a part of the image MG is a predetermined value within the range of 20 to 60 degrees, and is 45 degrees as an example. The model MF created by the model acquisition unit 72 is temporarily stored in a predetermined storage area of the storage unit 52. If the feature is different depending on the position of the container 20 in the circumferential direction, a portion other than the label 22 in the container 20 may be used as the model MF.

  The accumulation unit 73 specifies a front position, which is a position in the circumferential direction of the container 20 in the captured image during the entire circumference image generation processing, cuts out a partial image PG located in front of the container 20 from the captured image, and locates the partial image PG A process of accumulating data for all rounds in association with information is performed. The accumulation unit 73 includes a matching processing unit 75 used when specifying the front position of the container 20 in the photographed image. The matching processing unit 75 performs processing for extracting features to the captured image (another captured image) sequentially acquired to obtain a captured feature image, and the model MF is used for the captured feature image to resemble the pattern A matching process for searching for a similar symbol portion AM (FIG. 7A) is performed.

  Here, the details of the accumulation process performed by the accumulation unit 73 will be described with reference to FIGS. As illustrated in FIG. 5, the all-round image generation unit 62 generates an image of the whole circumference of 0 to 360 degrees in which the label 22 is panoramically expanded. The storage unit 73 cuts out a partial image PG at the front position C0 of the container 20 (particularly in this example, the label 22) from the captured image SG shown in FIG. 6 etc. sequentially acquired, and collects a plurality of partial images PG By doing this, an all around image AG (FIG. 14) of the label 22 is generated. When the first captured image SG (one captured image) shown in FIG. 6 is acquired, the accumulation unit 73 obtains a partial image PG indicated by a two-dot chain line in FIG. 6 at the front position C0 of the container 20 in the captured image SG. get. Here, the partial image PG is an image having a width value WP of, for example, a predetermined value (for example, 20 degrees) within a range of 10 to 30 degrees with respect to a representative angle RK (for example, 0 degrees) described later. First, in order to set the front position C0 (front angle) to the representative angle RK = 0 degrees, the accumulation unit 73 acquires a partial image PG of the width angle WP centered on the front position C0.

  The accumulation unit 73 collects a total of J partial images PG at each position (representative angle RK shown in FIG. 9) in 360 / J-degrees where the 360 ° of the entire circumference of the container 20 is divided by J. In this example, a total of 36 partial images PG are collected at every 10 degree position where 360 degrees is divided into 36 parts. In this case, a partial image PG of the width angle WP shown in FIG. 6 is obtained by adding an angle width of ± γ degrees on both sides in the circumferential direction to 10 degrees of width angle obtained by dividing 360 degrees into 36. γ is, for example, a value within the range of 0 to 10 degrees, and in this example, γ is set to 5 degrees. The accumulation unit 73 acquires, as a partial image PG, an area of the label 22 in a range in which the width angle WP is 20 degrees around the center position C0 in the front of the container 20 in the first captured image SG (FIG. 6). The storage unit 73 sets the front position of the container 20 in the first captured image SG (FIG. 6) to the reference position (temporary 0 degree) (see FIG. 5), and advances the subsequent storage processing. When the container 20 is photographed with the camera 31, a photographed image can be obtained at an angle of view near 170 degrees, but since both sides of the container 20 are photographed with the pattern 23 largely distorted, the photographed image SG An image in which the width angle of the container 20 is in a range of, for example, 90 to 170 degrees and which has a predetermined value (for example, 110 degrees) is used. However, the width angle of the container 20 that determines the range of the captured image SG can be appropriately changed according to the number of cameras 31.

  The matching processing unit 75 performs a matching process using the model MF with the area of the label 22 in the next captured image SG shown in FIG. 7A as a search target. The matching processing unit 75 shifts the model MF by a predetermined pitch (for example, one pixel or several pixels) in the photographed feature image in which the next photographed image SG is feature-extracted to change the position of the symbol part to be compared. Find the score of similarity with. If the score of the degree of similarity exceeds a predetermined threshold, the score of the degree of similarity and the position information of the similar pattern portion are temporarily stored in the storage unit 52. Then, when the search processing is finished, the matching processing unit 75 determines that the similar symbol portion AM having the highest score of similarity is matched. In the example shown in FIG. 7A, the model MF matches the similar symbol portion AM at the position shown in the figure.

  The storage unit 73 calculates a detection position C1 indicated by position coordinates in the circumferential direction of the similar symbol portion AM matched in the matching process, and detects the detection position C1 and a registered position Cm of the model MF used in the matching process The calculation using the geometrical relationship shown in FIG. 8 is performed on the basis of the reference position which is 0 degree of) to specify the front position C0 (front angle θc) of the container 20 in the captured image SG. The accumulation unit 73 acquires a partial image PG of the label 22 in the range of the width angle WP (20 degrees in this example) centered on the representative angle RK of the container 20 from the next captured image SG shown in FIG. 7B. . The contents of calculation using the geometrical relationship shown in FIG. 8 will be described later.

  Here, three storage units 53 to 55 used by the storage unit 73 in the storage process will be described with reference to FIG. The position storage unit 53 is J (36 in this example) corresponding to the representative angles RK = 0 degrees, 10 degrees, 20 degrees,..., 360 degrees obtained by dividing the entire circumference 360 degrees into J (36 in this example). The storage unit AK of The storage unit AK is reset to an initial value (for example, "99") prior to the start of the all around image generation process. A front angle θc at which the front position C0 of the partial image PG is indicated by an angle based on the reference position (temporary 0 degree) belongs to a region of any one of the J representative angles RK and a region to which the front angle θc belongs In the storage portion AK of the representative angle RK, an error between the representative angle RK and the front angle θc is stored. When the representative angle RK in steps of 10 degrees is RKi (where i = 0, 1,... 35), the condition RKi-5 <θc ≦ RKi + 5 indicating that the front angle θc belongs to the region of the representative angle RKi The storage unit AK of the representative angle RKi stores a value corresponding to an error between the representative angle RKi and the front angle θc. As described above, in the present embodiment, the position information of the partial image PG is indicated by the relative position (relative angle) with respect to the reference position (temporary 0 degree), and is indicated by the representative angle and the error.

  The first storage unit 54 includes J storage units PK corresponding to the representative angle RK. The storage portion PK stores image data of a partial image PG whose center is the width of the representative angle RK. Further, the second storage unit 55 includes J storage units QK corresponding to the representative angle RK. In the storage unit QK, image data of a captured image SG which is a cutout source of the partial image PG stored in the storage unit PK of the same representative angle RK is stored. When a value other than the initial value is stored in the storage unit AK of the position storage unit 53, the storage unit PK of the same representative angle RK is representative of the photographed image SG of the front angle defined by the representative angle RK and the error. A partial image PG in which a predetermined region centered at the angle RK is cut out is stored.

  For example, in the example shown in FIG. 9, since “0 (zero)” is stored in the storage unit AK whose representative angle RK is “0 degrees”, the representative angle RK is stored in the storage unit PK of the representative angle RK. And a partial image PG cut out from the photographed image SG having a front angle θc of “0 degree” without any error. Further, since “2” is stored in the storage unit AK of which the representative angle RK is “20 degrees”, the storage angle PK of the representative angle RK has a front angle with an error of 2 degrees from the representative angle RK. A partial image PG whose width center is 20 degrees of the representative angle RK is stored from the photographed image SG of “22 degrees”. Then, the storage unit 73 sequentially stores the partial images PG in the respective storage units PK, and finally the partial images PG for the entire circumference of the label 22 are stored in the first storage unit 54.

  FIG. 10 shows a partial image PG stored in the first storage unit 54 in the example shown in FIG. As shown in FIG. 10, the first storage unit 54 has a partial image PG with a front angle "0 degree" and a width angle of 20 degrees, and a representative angle of 20 degrees from the photographed image SG with a front angle of 22 degrees. And a partial image PG obtained by cutting out a predetermined area having a width angle of 20 degrees with the width center as the center. In the first storage unit 54, the partial image PG is stored so as to overlap the adjacent partial image PG by an angle width of γ degrees (5 degrees in this example).

  Then, as shown in FIG. 11, features are extracted from a part of the image MG in front of the current captured image SG to obtain a model MF, and the previous model MF is updated to a new model MF shown in FIG. . Then, similarly, as shown in FIG. 12A, the captured image SG is calculated by using a geometrical relationship based on the positional information of the similar symbol portion AM matched in the matching process and the positional information of the model MF. The front position C0 (front angle θc) of the container 20 inside is acquired, and a partial image PG of the label 22 centered on the representative angle RK of the container 20 is acquired from the photographed image SG (FIG. 12 (b)). Then, the error between the representative angle RK and the front angle θc is stored in the storage unit AK of the representative angle RK to which the front angle θc belongs, and the partial image PG is stored in the storage unit PK of the representative angle RK. Thereafter, by repeating the same storage processing, partial images PG shown by two-dot chain lines in FIG. 10 are sequentially stored in the first storage unit 54 shown in FIG. Partial images PG (36 sheets in this example) are accumulated.

  Here, as shown in FIG. 9, since the error with respect to the representative angle RK is known from the value stored in the position storage unit 53, the representative angle RK is obtained even if a value other than the initial value is already stored in the storage unit AK. It is preferable to capture the partial image PG from the photographed image SG, if the photographed image SG having a front angle θc smaller than the error of is obtained. Therefore, if a captured image SG having a smaller front angle θc with respect to the representative angle RK is obtained, the partial image PG cut out from the captured image SG is replaced with the partial image PG previously stored in the first storage unit 54. Update Thus, the storage unit 73 sequentially updates the partial images PG while updating to the partial images PG with a smaller error from the representative angle RK, thereby enhancing the accuracy of the generated all-round image AG.

  Returning to FIG. 3, the model registration unit 63 selects a plurality of models M based on a plurality of image areas selected by the operator operating the input device 41 from the all-round image AG generated by the all-round image generation unit 62. The model data MD including the plurality of models M is created and registered by writing in a predetermined storage area of the storage unit 52.

  The non-defective item image generation unit 64 shown in FIG. 3 generates non-defective item image data GD including a plurality of non-defective item images RG used in comparison processing at the time of inspection. The non-defective image generation unit 64 generates a predetermined width angle (for example, 90 to 150 degrees (for example, 10 degrees) per angle (for example, 10 degrees) obtained by dividing 1 round into m (for example, 36 divisions) based on the image data of the entire circumference image AG. Non-defective images RG0 to RG35 shown in FIG. 16 consisting of partial images of 110 degrees) are generated. The non-defective images RG0 to RG35 are associated with positional information of the container 20 in the circumferential direction. The division number m for determining the number of non-defective images RG is not limited to m = 36, and can be changed as appropriate. For example, m may be 12, 18, 24, 72, 120.

  Further, the feature image generation unit 65 shown in FIG. 3 generates feature image data FD used when acquiring an inspection image from a photographed image obtained by photographing the container 20 to be inspected. The feature image generation unit 65 extracts features from the non-defective images RG0 to RG35 and generates feature images FG0 to FG35. The characteristic images FG <b> 0 to FG <b> 35 are associated with positional information of the container 20 in the circumferential direction.

  Next, the inspection unit 66 shown in FIG. 3 will be described. The inspection unit 66 compares the inspection image KG in the image data captured and captured by the camera 31 with the container 20 to be inspected with the non-defective item image RG stored in advance, and sets the defect FT of the label 22 (both in FIG. Perform label inspection processing to check the presence or absence of

  The image acquisition processing unit 81 performs processing for capturing a captured image (image data) obtained by capturing a container 20 to be inspected by the camera 31. The image acquisition processing unit 81 of this example acquires four photographed images SG obtained by photographing the container 20 from four directions from the four cameras 31. The image acquisition processing unit 81 subjects the captured image SG to image processing such as position correction and clipping processing.

  The position specifying unit 82 specifies a front position that is a position in the circumferential direction of the container 20 in the captured image SG. Here, the front position is a value representing the position of the front of the container 20 to be inspected, which faces the camera 31 at the time of photographing, as a predetermined angle in the circumferential direction of the label 22 (see FIG. 5). The position specifying unit 82 performs a position specifying process of specifying the front position of the container 20 in the captured image SG using the model M shown in FIG. The position specifying unit 82 in this example performs an omni-matching process for searching for a similar pattern portion similar to the model M from the patterns 23 of the label 22 in the captured image using a plurality of models M, and performs the search The front position of the captured image SG is specified based on the position information of the similar symbol part and the registered position information of the model M.

  The label height inspection unit 83 shown in FIG. 3 inspects displacement (height displacement) of the position in the height direction Z where the label 22 is attached to the container 20. The label height inspection unit 83 of this example registers the position in the height direction Z of the similar pattern portion similar to the model M specified by the position specifying unit 82 in the omni-matching process, and registers the model M in the height direction Z The presence or absence of positional deviation (height deviation) in the height direction Z of the label 22 is inspected from the difference from the position (height deviation). In addition, the imaging | photography area of the camera 31 is set as the known predetermined height from the upper surface of the conveyor 13 in which the container 20 is mounted.

  The detection unit 84 performs a matching process on the captured image SG using the feature image FG corresponding to the front position identified by the position identification unit 82, and acquires a matched region as an inspection image KG. In this example, an image in a partial region including the label 22 in the captured image SG is acquired as an inspection image KG (see FIG. 19). The detection unit 84 detects the defect FT of the label 22 based on the comparison result of the inspection image KG and the non-defective item image RG. The comparison processing unit 85 performs comparison processing for comparing the inspection image KG and the non-defective product image RG, obtains the difference between the inspection image KG and the non-defective product image RG as a comparison result, and acquires a difference image DG (see FIG. 19). . Then, the determination unit 86 determines the presence or absence of the defect FT of the label 22 based on the difference image DG. Here, the defects FT include tears on the label 22, adhesion of foreign matter, stains (not shown), and the like.

  Next, the operation of the label inspection device 11 will be described. The all-round image generation processing performed by the computer 50 executing the program PR will be described with reference to the flowchart shown in FIG.

  The CPU 51 executes an all-round image generation process shown by the flowchart in FIG. Before executing this process, the CPU 51 initializes the storage units 53 to 56 in the storage unit 52. When the position storage unit 53 shown in FIG. 9 is initialized, the value of each storage unit AK becomes an initial value “99”.

  First, in step S11, the CPU 51 acquires a photographed image of the container 20. That is, the camera 31 acquires image data obtained by photographing the container 20. In this example, the image data of the photographed image SG of the entire circumference (four) taken by the four cameras 31 from four directions is acquired. In the present embodiment, the process of step S11 corresponds to an example of the image acquisition step.

  In step S12, the CPU 51 sets the front position of the container 20 in the photographed image to a reference position (for example, 0 degrees). Further, as shown in FIG. 6, a partial image PG at the front position of the container 20 in the captured image SG is acquired. The CPU 51 stores the difference between the representative angle RK and the front angle θc in the storage unit AK having a representative angle RK of 0 degrees, stores the partial image PG in the storage unit PK of 0 degrees, and further stores the storage unit QK of 0 degrees. The photographed image SG is stored in.

  In step S13, the CPU 51 creates a model. An image of a partial region of the front of the container 20 in the image is cut out, and feature extraction is performed on the partial image to create a model MF. In this example, as shown in FIG. 6, the model MF is a range including at least a part of the label 22 in the height direction Z at a predetermined angle (for example, 45 degrees) range in the front of the container 20 in the captured image SG. The image MG of is used. In the model MF, the registration position is set to the reference angle (0 degree). The position, shape, and size of the model MF can be changed as appropriate.

  In step S14, the CPU 51 reads the next captured image and performs a matching process after position correction. That is, the matching processing unit 75 performs a matching process on the next captured image to search for similar symbol parts using the model MF. In the matching process, the matching process is performed while moving a pattern area of the comparison object to be compared with the model MF little by little in the processing target area, and a similar pattern portion having a score of similarity with the model MF equal to or more than a threshold is searched Do.

  In step S15, the CPU 51 determines whether or not the matching process can match. If it does not match, the process returns to step S14, the next captured image is read, and a matching process is performed to search for a similar symbol part using the model MF. On the other hand, if it matches, it will progress to step S16.

In step S16, the CPU 51 calculates the front position of the container 20 in the captured image using the position information of the matched model. The calculation of the front position is performed as follows. When one similar symbol portion is determined, a model registration position Cm (coordinates of a front angle of the model M) shown in FIG. 8 of the model MF used at that time is acquired. Furthermore, the CPU 51 calculates the front position C0 (front angle θc) of the container 20 in the captured image using the model registration position Cm and a known value such as the container width view angle α. The front angle θc of the container 20 is given by the following equation (1) from the geometrical relationship shown in FIG.
θc = θmc +
(Arcsin ((C0-C1) / r)-arcsin ((C 0-Cm) / r))) 180 / pi ... (1)
Here, θmc is a front angle corresponding to the registration position of the corresponding model MF. C0 is the center position (front position) of the container 20 in the photographed image. Cm is a registration position (model registration position) of the corresponding model MF. C1 is a detection position at which a similar symbol similar to the corresponding model MF is detected. r is the radius of the container 20. The radius r is given by the following equation (2) using the container width view angle α and the container view angle width W (see FIG. 8).
r = W / 2 ÷ sin (α / 2) · 180 / π (2)
Thus, the matching processing unit 75 of the storage unit 73 calculates the front angle θc of the container 20 in the photographed image, using the equations (1) and (2).

  In step S17, the CPU 51 determines whether the representative angle to which the front angle θc belongs has been registered. That is, the CPU 51 determines whether or not the previous partial image PG has been registered in the storage unit PK which is an area of the representative angle RK to which the front angle θc of this time belongs. In this example, the CPU 51 determines whether or not the value is other than “99” based on the value of the storage unit AK of the corresponding representative angle RK, to thereby obtain the partial image PG in the corresponding storage unit PK. It is judged whether or not is registered. If the representative angle is already registered, the process proceeds to step S18, and if not registered (that is, if it is not registered), the process proceeds to step S19.

  In step S18, the CPU 51 determines whether the front position is to be updated. That is, in the present embodiment, in order to manage the circumferential position of the container 20 by an angle, the CPU 51 uses the storage angle AK corresponding to the representative angle RK at this time more than the front angle θc already stored (registered). It is determined whether or not the front angle θc of is closer to the representative angle RK. In this example, since the error with the representative angle RK is stored as the front position information in the storage unit AK, it is determined whether the absolute value of the current error is smaller than the absolute value of the previous error. . The CPU 51 determines that the front position to be updated is the front position to be updated if the current front angle θc is closer to the representative angle RK, and the registered front angle θc is the representative angle RK than the current front angle θc. If it is close, it is judged that it is not the front position which should be updated. If the front position is to be updated, the process proceeds to step S19. If the front position is not to be updated, the process proceeds to step S21.

  In step S19, the CPU 51 registers a partial image and front position information. That is, the storage unit 73 stores the current partial image PG cut out from the current captured image SG in the storage unit AK of the corresponding representative angle RK in the first storage unit 54 and the corresponding representative angle in the position storage unit 53. The front position information is stored in the storage unit AK of RK. With this storage, the current partial image PG and front position information are registered. In this case, when the front position information has already been registered in the storage unit AK of the corresponding representative angle RK, the registration information is updated to the current partial image PG and the front position information. Here, the front position information is stored in the storage unit AK as a value of an error between the front angle θc and the representative angle RK. In addition, in the storage unit QK of the corresponding representative angle RK in the second storage unit 55, the current captured image SG is stored.

  In step S20, the CPU 51 updates the model. That is, the model acquisition unit 72 extracts features of an image of a predetermined region at a front position of the container 20 in the current captured image SG to create a new model MF, and generates a new model MF of the previous model. Replace with MF. In the present embodiment, the processes of steps S13 and S20 correspond to an example of the model acquisition step.

  In step S21, the CPU 51 determines whether or not the all around image has been completed. If the all-round image AG is not completed, the process returns to step S14, and the generation process (S14 to S21) of the all-round image AG is repeated until it is determined that the all-round image AG is completed in step S21. On the other hand, if the all around image AG is completed, the process proceeds to step S22. In the present embodiment, the processes of steps S14 to S19 correspond to an example of the accumulation step.

  Here, during the all-round image generation process of repeating the all-round image AG generation process (S14 to S21), the all-round image generation progress screen SC1 shown in FIG. 13 is displayed on the display unit 42. On the screen SC1, an annular progress scale PS is superimposed and displayed on the front surface of the photographed image SG being processed as a background. The progress scale PS is unlearned that the partial image PG has not been stored yet, or the partial image PG is not yet stored for each area corresponding to each representative angle RK divided into 360 (one divided 360 in this example) into 360 degrees. It is displayed in a distinguishable manner by the difference in color whether stored or already learned. In addition, the region of the learned representative angle RK is color-coded stepwise in accordance with the value of the error between the front angle θc of the stored partial image PG and the representative angle RK. For this reason, the worker can confirm the generation progress status and the completion accuracy of the all-around image AG by viewing the progress scale PS. When the unlearned area on the progress scale PS disappears, the all around image AG is completed, and the CPU 51 proceeds to the process of step S22.

  In step S22, the CPU 51 determines whether to perform reinforcement learning. If reinforcement learning is to be performed, the process proceeds to step S23. If reinforcement learning is not performed, the process proceeds to step S24. Here, reinforcement learning is processing that is repeated a plurality of times until reinforcement learning end conditions are satisfied, which is a predetermined condition for ending the generation processing (S14 to S23) of all-round images, to improve the completion accuracy of all-round images AG. .

  In step S23, the CPU 51 determines whether a reinforcement learning termination condition is satisfied. If the reinforcement learning termination condition is not established, the process returns to step S14, and reinforcement learning (S14 to S23) is repeated until it is determined that the reinforcement learning termination condition is established in step S23. On the other hand, if the reinforcement learning termination condition is satisfied, the process proceeds to step S24. The CPU 51 determines that the reinforcement learning end condition is satisfied when the predetermined determination parameter used to determine the reinforcement learning end has reached the threshold value. For example, the number of photographed images read in reinforcement learning, the number of updates of partial images, the average error, the occupancy rate of the area less than the error threshold, or the number of epochs is used as the determination parameter.

  In step S24, the CPU 51 corrects the angle of the entire circumference image. The operator designates the 0-degree position of the all-round image AG in the computer 50 by operating the input device 41 prior to the start of the all-round image generation process or after the all-round image is completed. The CPU 51 confirms the designated 0 degree position, and sets the angle of the all-round image AG associated with the angle on the basis of the provisional 0 degree position shown in FIG. Correct the angle so that Further, as shown in FIG. 14, the CPU 51 creates the all-round image AG by extending it from the range of 0 to 360 degrees to a predetermined value exceeding 0 to 360 degrees. In this example, the full circumference image AG is extended, for example, to a range of 0 to 540 degrees (one and a half). Here, to create the all-round image AG over a range exceeding 360 degrees, a model registration screen (not shown) is displayed on the display unit 42 at the time of model registration performed by the operator activating the model registration unit 63 thereafter. This is for the purpose of smoothly performing an operation of displaying the entire circumference image AG on the model registration screen and designating a region for registering the model M (FIG. 15) by the operation of the input device 41. If the all-around image AG is created over a range exceeding 360 degrees, for example, in the all-around image AG shown in FIG. 14, an area across 0 or 360 degrees between the characters “G” and “H” is It can be specified. In addition, if it is not necessary in particular, all the circumference image AG may be 0 to 360 degrees.

  If there is at least one region of the representative angle RK of the sky in which the partial image PG is not stored because the entire-circle image AG can not be completed due to a lack of the captured image SG, etc. The full circumference image AG may be completed by supplementing the partial image of the representative angle RK. That is, even if the model acquisition process and the accumulation process are repeated until the predetermined condition is satisfied, if the all-around image AG can not be completed, the accumulation unit 73 sets the unregistered representative angle RKi in which the partial image PG is not accumulated. get. If there is an area of the registered representative angle RKi + 1 or RKi-1 at a position next to the area of the unregistered representative angle RKi, the photographing which is the acquisition source of the partial image PG accumulated in this adjacent area An image SG is obtained from the storage unit QK of its representative angle RKi + 1 or RKi-1. A partial image PG at a position corresponding to the representative angle RKi of the unregistered area is acquired from the captured image SG, and the acquired partial image PG is associated with position information and accumulated in the empty area of the unregistered representative angle RKi. Do. In this case, the compensated partial image PG has an error (for example, 12 degrees) in which the error between the front angle and the representative angle exceeds a prescribed ± 5 degree. Although the accuracy of the complete circumference image AG completed in this way is partially reduced at the filling point, model registration processing, non-defective image generation processing, container orientation detection processing and inspection performed using the total circumference image AG The process can be performed.

  Next, model registration processing will be described with reference to FIG. When the operator performs model registration, the operator operates the input device 41 and instructs the label inspection device 11 to start model registration processing. When receiving the instruction, the CPU 51 of the computer 50 executes model registration processing shown by the flowchart in FIG.

  First, in step S31, the CPU 51 displays an all around image on the display unit 42. That is, the CPU 51 displays on the display unit 42 a model registration screen (not shown) including the all around image AG. The operator designates a model area GA (see FIG. 15) which is an image area having a feature to be used as the model M in the all-round image AG on the model registration screen by operating the input device 41.

  In the next step S32, the CPU 51 determines whether or not the model area GA is designated. If the model area GA is not specified, the process waits until the model area GA is specified. If the model area GA is specified, the process proceeds to step S33.

  In step S33, the CPU 51 registers the model in association with the position information. Since the all-round image AG is data associated with position information (horizontal coordinate in the container circumferential direction and vertical coordinate in the height direction), position information of the designated model area GA is calculated from the coordinates of the all-round image AG Identified above.

  In the next step S34, the CPU 51 determines whether all models have been registered. If registration of all models is not completed, the process returns to step S32, and thereafter, the processing of steps S32 to S34 is repeated until it is determined in step S34 that registration of all models is completed. Thus, the operator designates model regions GA in order, and finishes registration of all (for example, eight) models M, whereby the storage unit 52 stores model data MD.

  In addition, the non-defective item image generation unit 64 sets a predetermined width angle (for example, 110 degrees) for each angle (for example, 10 degrees) obtained by dividing one round into m (for example, 36 divisions) based on the image data of the entire circumference image AG. The non-defective images RG0 to RG35 shown in FIG. 16 composed of partial images are acquired. Then, the non-defective item image generation unit 64 stores the non-defective item images RG0 to RG35 in the storage unit 52 in association with the circumferential position (front position) of the container 20. The feature image generation unit 65 also extracts features from the non-defective images RG0 to RG35 to generate feature images FG0 to FG35. Then, the feature image generation unit 65 stores the feature images FG0 to FG35 in the storage unit 52 in association with the circumferential position (front position) of the container 20. As described above, since the computer 50 automatically performs creation and registration of the model data MD, the non-defective image data GD and the feature image data FD from the all-around image AG, the operator compares the specification of the condition and the area by the input device 41 You just need to do a simple operation.

  Next, the label inspection process will be described with reference to FIG. The computer 50 performs label inspection processing. The CPU 51 of the computer 50 executes the label inspection process shown by the flowchart in FIG. 18 in a state where the necessary data GD and FD are stored in the storage unit 52. In the storage unit 52, the user operates the input device 41 from the entire circumferential image of the non-defective container 20, and one of the partial images for each angular range of 0 to 360 / N degrees (for example, 45 degrees). The area of the part is registered as models M1 to M8. The models M1 to M8 are feature images (for example, shape images) from which features (shapes or brightness) are extracted, and are associated with circumferential positions (angles) of the container 20 (or label 22).

  In the storage unit 52, partial images at different positions (angles) in the circumferential direction of the non-defective container 20 or portions obtained by dividing the original image of the label 22 (see FIG. 5) at different positions (angles) in the circumferential direction A plurality of non-defective images RG, which are images, and a feature image FG in which the features of each non-defective image RG are extracted are stored in association with positions (angles) in the circumferential direction. In the present embodiment, the non-defective item image RG and the characteristic image FG are generated by the non-defective item image generation unit 64 automatically based on the plurality of images of the non-defective container 20 taken. It is generated by dividing the printing plate) and is associated with the circumferential position (angle) of the container 20.

  First, in step S41, the CPU 51 acquires image data of the container 20. That is, the camera 31 acquires image data obtained by photographing the container 20. At this time, image data of the photographed image SG of the entire circumference (four) taken by the four cameras 31 from four directions is acquired. In the following, it is assumed that the photographed image SG has a defect such as breakage or adhesion of a foreign substance or the like to the label 22 to be inspected attached to the container 20.

  In step S42, the CPU 51 specifies the front position of the container in the photographed image. The position specifying unit 82 executes the omni-matching process to specify the front angle θc and the height of the label 22. The position specifying unit 82 performs an omni-matching process for searching for a similar symbol portion similar to the model M with respect to the area of the label 22 in the captured image SG obtained by capturing the container 20 to be inspected by the camera 31. In this omni-matching process, the position specifying unit 82 performs a matching process while moving a symbol area to be compared with the model M little by little, and sets a similar pattern part whose score of similarity with the model M is equal to or more than a threshold Explore. Moreover, when the similar symbol part whose score is more than a threshold value is not detected, it changes to the following model M. Thus, while sequentially changing the model M, a matching process is performed to search for similar symbol parts for all the models M1 to M8. In this example, in the position specifying process, the omni matching process may be performed on one of the four captured images SG, or the omni matching process may be performed on all the four captured images SG. When a plurality of similar symbol parts are detected, one similar symbol part having the highest score among them is determined. In the omni-matching process, for example, the matching process may be stopped when a similar symbol portion whose score of similarity exceeds the target value is searched.

  In step S43, the CPU 51 performs comparison processing between the inspection image and the non-defective image. Specifically, as shown in FIG. 19, the comparison processing unit 85 performs an operation of taking the difference between the inspection image KG and the non-defective item image RG, and acquires a difference image DG shown in the figure as a comparison result. The difference image DG is an image in which the normal part of the label 22 is white and the part of the defect FT in the label 22 is black or gray. In the difference image DG, a region (black or gray region) whose density (for example, luminance) is equal to or less than the density threshold is detected as the defect FT. For example, a broken label 22 or a foreign substance is detected as a defect FT. The difference between the two images KG and RG may be reversed from the example of FIG. In this case, the normal part is black and the defect FT is white or gray.

  In step S44, the CPU 51 outputs an inspection result. Specifically, the determination unit 86 determines the presence or absence of the defect FT based on the difference image DG. Then, the control unit 61 causes the display unit 42 to display the inspection result. The determination unit 86 may determine the type of the defect FT in addition to the presence or absence of the defect FT. For example, in the difference image DG, the density, the shape, the size, and the like may be used as parameters, and the type of the defect FT may be determined using a threshold or the like set for each parameter. In this case, in addition to the presence or absence of the defect, the type of the defect FT is displayed on the display unit 42 in the case of the defect.

As described above, according to this embodiment, the following effects can be obtained.
(1) The all-round image generation unit 62 that constitutes an example of the all-round image generation device is an image acquisition processing unit 71 that sequentially acquires the photographed image SG of the non-defective container 20; The storage unit 73 sets the front position of the container 20 in the captured image SG as a reference position and associates the partial image PG at the front position of the container 20 with position information in the circumferential direction and stores the partial image PG in the storage unit 52. Furthermore, the all-around image generation unit 62 includes a model acquisition unit 72 that acquires a model MF based on a predetermined region of a part of the container 20 in one captured image SG. As the accumulation process, the accumulation unit 73 first performs a matching process using the model MF on the next photographed image SG acquired after the photographed image SG of the acquisition source of the model MF. Further, as the accumulation processing, if the storage unit 73 matches, the accumulation unit 73 determines the front position C0 of the container 20 based on the matched detection position C1 and the registered position Cm of the model MF, and the front surface of the container 20 in the next captured image SG. The partial image PG at the position C0 is stored in the storage unit 52 in association with circumferential position information (representative angle RK and an error). If the storage unit 73 does not match, the storage unit 73 shifts to the next captured image SG. Model acquisition processing for acquiring the model MF based on a predetermined region of a part of the container 20 in the captured image SG when the model acquisition unit 72 matches, and matching performed by the accumulation unit 73 using the latest model MF The all-round image AG is generated by repeating the accumulation process including the process until a plurality of partial images PG associated with circumferential position information are collected corresponding to the entire circumference. Thus, it is possible to relatively easily generate the all-round image AG associated with the position information of the container 20 in the circumferential direction. For example, the operator photographs the container 20 while manually positioning the photographing position of the container 20, and manually operates the input device of the PC from a plurality of photographed images in association with the partial image of the container 20 with the position information It is possible to eliminate the troublesome work of acquiring a plurality corresponding to one rotation.

  By the way, although it is possible to obtain an image for one rotation using a line camera, in this case, it is necessary to separately prepare a line camera for acquiring an all-round image, and if the cameras are different, Since the image quality and the like of the photographed image and the model M and the non-defective image RG generated based on the all-round image AG are different, they affect the inspection accuracy. However, in the present embodiment, the all-around image AG can be generated using the captured image SG of the container 20 captured by a normal camera such as the camera 31 used for the inspection. Therefore, high inspection accuracy can be obtained.

  (2) The accumulation unit 73 accumulates the partial image PG for each storage unit AK (region) to which the representative angle RK, which is obtained by dividing the circumference of the container 20 into a plurality, belongs. If the previous partial image PG is already stored (registered) in the storage unit AK of the representative angle RK to which the current front angle θc (front position C0) at the time of matching belongs, the storage unit 73 has the current front angle If θc is closer to the representative angle RK than the front angle θc of the previous partial image PG, the storage data of the storage unit AK of the representative angle RK is updated from the previous partial image PG to the current partial image PG. Therefore, since the partial image PG accumulated during progress of the accumulation processing is updated to a more appropriate partial image PG, it is possible to generate a more accurate all-round image AG with less distortion.

  (3) Even if the accumulation unit 73 can not complete the all-round image AG even if the model acquisition process and the accumulation process are repeated until the predetermined condition is satisfied, a representative of the unregistered area in which the partial image PG is not accumulated Get the angle RK. Then, if there is a registered area at a position next to the unregistered area, the accumulation unit 73 determines that the registered image is not registered from the photographed image SG which is the acquisition source of the partial image PG accumulated in the registered area. A partial image PG at a position corresponding to the representative angle RK of the area is acquired, and the acquired partial image PG is stored in an unregistered area in association with position information. Therefore, even if the full circle image AG can not be completed, the full circle image AG can be completed using the existing captured image SG. Therefore, predetermined processing such as model registration processing, article orientation detection processing, and article inspection processing can be performed using the all-round image AG.

  (4) The label inspection device 11 generates a plurality of all-round image generation unit 62 and a plurality of model regions GA having different positions in the circumferential direction selected from the all-round image AG generated by the all-round image generation unit 62. A model registration unit 63 is provided that creates a model M and registers the plurality of models M in association with circumferential position information. Therefore, a plurality of matching processing models M used for processing such as specification of the front position of the container 20 can be generated relatively easily using the easily generated all-round image AG.

  (5) The label inspection apparatus 11 performs a matching process on the photographed image obtained by photographing the container 20 of the direction identification target having the symbol 23 on the outer surface using the model M, and the position information of the matched similar symbol portion and the corresponding The position identification unit 82 identifies the front position C0 of the container 20 in the captured image SG based on the position information of the model M. Therefore, it is possible to detect the orientation that is the front position where the container 20 faces the camera 31, and furthermore, it is possible to relatively easily generate the all-round image AG necessary for detecting the orientation of the container 20.

  (6) The label inspection device 11 inspects the all-round image generation unit 62, a non-defective item image generation unit 64 that generates a plurality of non-defective images RG that are partial images in the circumferential direction from the all-round image AG. The inspection unit 66 inspects the presence or absence of the defect FT of the target container 20 (in particular, the label), and the position specifying unit 82 specifying the front position C0 of the container 20 to be inspected in the photographed image SG. The inspection unit 66 inspects the presence / absence of the defect FT of the container 20 based on the comparison result of comparing the inspection image KG and the non-defective product image RG corresponding to the front position C0. Therefore, the presence or absence of the defect FT of the container 20 can be inspected, and the non-defective item image data GD necessary for the inspection can be easily acquired. In particular, in the present embodiment, the position specifying unit 82 performs the matching process on the captured image SG using a plurality of models M to specify the front position C0 of the container 20 to be inspected in the captured image SG. The plurality of models M are generated based on the model area GA selected from the all-round image AG generated by the all-round image generation unit 62. Therefore, a plurality of models M necessary for inspection can be easily acquired.

  (7) The label 22 is attached to the outer surface of the container 20 to be inspected by the label inspection apparatus 11, and the inspection unit 66 inspects the defect FT of the label 22. Thus, the defect FT of the label 22 attached to the container 20 can be inspected.

  (8) The all around image generation method includes an image acquisition step (S11), a model acquisition step (S13, S20), and an accumulation step (S14 to S19). According to this method, it is possible to relatively easily generate the all-round image AG associated with the positional information of the container 20 in the circumferential direction.

Second Embodiment
Next, a second embodiment will be described with reference to FIGS. In the second embodiment, the matching process is performed using a plurality of models MF to shorten the time required for the all-round image generation process. When the computer 50 receives the all-round image generation instruction, the computer 50 executes the all-round image generation process shown by the flowchart in FIG.

  The processes of steps S51 to S55 shown in FIG. 22 are the same as the processes of steps S11 to S15 in FIG. 4 of the first embodiment. That is, the front position of the container 20 in the captured image SG acquired first is set to the reference position (for example, 0 degree), and the partial image PG at the front position is obtained (S51, S52). The model M is created from an image of a part of a predetermined area at the front position of the container 20 in the photographed image SG (S53). Then, matching processing using the model MF is performed on the next captured image SG (S54), and if it matches (affirmative determination in S55), the process proceeds to step S56. On the other hand, if it does not match (negative determination in S55), the photographed image SG is sequentially read until it matches, and if it matches, the process proceeds to step S56.

  Then, in step S56, the front position of the container is calculated based on the position of the most matched model. Here, since the matching processing is initially performed using one model MF, the front position of the container 20 is calculated based on the registration position of one model MF. In the subsequent processing, the number of models MF is sequentially increased to the maximum number.

  The processes of the next steps S57 to S59 correspond to the processes of steps S17 to S19 in FIG. That is, if the front position of the container 20 in the captured image SG has been registered (YES at S57), it is determined whether it is the front position to be updated (S58). Then, if the front position is not registered (NO in S57) or if it is a front position to be updated (YES in S58), the partial image and the front position information are registered by storing in the storage unit 52 ( S59). That is, the representative angle RK to which it belongs is determined from the front position, and the storage portion 73 stores the front position information in the storage portion AK of the corresponding representative angle RK of the position storage portion 53, and the partial image PG is stored in the first storage portion 54. The corresponding representative angle RK is stored in the storage unit PK. The front position information stored in the storage unit AK is an error between the front angle θc and the representative angle RK. Further, the storage unit 73 stores the captured image SG in the storage unit QK of the corresponding representative angle RK of the second storage unit 55.

  In step S60, the CPU 51 determines whether a model addition condition is satisfied. In the present embodiment, there are two methods in which the model acquisition unit 72 adds up to the maximum number of models MF. One is a two-way system, and the other is a stepping stone system. The model addition condition is determined according to the method of adding the model MF. For example, the area of the representative angle RK to which the front angle θc of the container 20 in the captured image SG this time belongs must be adjacent to the area of the representative angle RK to which the front angle θc of the captured image SG of the acquisition source of the existing model MF belongs. For example, if one or more unregistered areas are sandwiched between them, the model addition condition is satisfied. If the model addition condition is not satisfied, the process proceeds to step S62, and if the model addition condition is satisfied, the process proceeds to step S61. The details of the two methods for adding the model MF will be described later.

  In step S61, the CPU 51 adds a model. The model acquisition unit 72 extracts features based on an image of a part of a predetermined area at the front position of the container 20 in the captured image SG this time, and creates and adds a model MF. Therefore, when the area of the representative angle RK to which the front angle θc of the present captured image SG belongs is positioned with one or more unregistered areas in between with respect to the area corresponding to the existing model MF, A new model MF is created from the captured image SG this time and added.

  In step S62, the CPU 51 determines whether a model deletion condition is satisfied. For example, if the area on both sides of the area corresponding to the existing model MF is registered, the model elimination condition is satisfied. Further, if one or both of both sides of the region corresponding to the model MF are unregistered, the model MF usable for the matching process for acquisition of the partial image PG to be accumulated in the adjacent unregistered region. Is left, the model elimination condition is not satisfied. If the model deletion condition is not satisfied, the process proceeds to step S64, and if the model deletion condition is satisfied, the process proceeds to step S63.

In step S63, the CPU 51 deletes the model MF for which the model deletion condition is satisfied. The model acquisition unit 72 also performs deletion of the model MF.
At step S64, the CPU 51 determines whether or not the all around image has been completed. If the all-round image is not completed, the process returns to step S54, and the all-round image generation processing (S54 to S64) is repeated until it is determined in step S64 that the all-round image is completed. On the other hand, if the all around image is completed, the process proceeds to the same processing as step S22 in FIG.

  Hereinafter, the same processing as steps S22 to S24 in FIG. 4 of the first embodiment is performed. That is, when the all-round image is completed (affirmative determination in S64), it is determined whether or not reinforcement learning is to be performed (S22). When reinforcement learning is to be performed (affirmation determination in S22), steps S54 to S64 and S22, S23. Are repeated until the reinforcement learning termination condition is satisfied. Further, regardless of the presence or absence of reinforcement learning, when the all-round image AG is completed, the angle correction of the all-round image AG is performed (S24). As a result, as shown in FIG. 14, the image data of the all-round image AG associated with the angle at which the designated position is 0 degrees is generated.

  Next, two methods using a plurality of models MF will be described. First, the two-way method will be described with reference to FIG. The two-way system is a system in which the model MF is increased in two directions, clockwise and counterclockwise, in the circumferential direction of the article. That is, first, the model MF is created at the reference position (0 degree), and the angle region to which the front position of the captured image SG determined from the position of the matched model MF belongs as a result of matching processing corresponds to the model MF under registration. When it is the position which opened one or more unregistered angle areas with respect to an angle area (for example, 0 degree), model MF is created from this photography picture SG. This condition is satisfied when one or more unregistered angle areas are opened clockwise on the angle area (for example, 0 degree) corresponding to the existing model MF, and counterclockwise There may be cases where one or more unregistered angle areas are open. In this method, the model MF is propagated in both the clockwise direction propagating model MFc and the counterclockwise direction propagating model MFk. Therefore, the model MF increases in the clockwise direction and also in the counterclockwise direction. As a result, since the matching process is performed by the plurality of models MF in the middle of the accumulation process, the whole circumference image can be completed earlier than the method of the first embodiment in which there is one model MF. .

  Next, the stepping stone system will be described with reference to FIG. The stepping stone method is a method of increasing the model MF to a position where one or more unregistered angle regions are opened. First, a model MF is created at the reference position (0 degrees). As a result of the matching process using the model MF, the angle area to which the front position of the container 20 in the captured image SG determined from the position matched with the model MF belongs with respect to the angle area (for example, 0 degrees) corresponding to the model being registered. In the case where one or more unregistered (“99”) angle regions are opened, a model MF is created from the captured image SG of this time and added. Since both sides of the angle area corresponding to the model MF are angle areas of the unregistered “99”, the frequency of matching the model MF in the matching process for the next captured image SG is high. In the example shown in FIG. 21 (a), one unregistered angle region is opened with respect to the first left model ML, and the middle model MM is increased, and the right model MR is further as shown in FIG. 21 (b). Increase. Thereafter, matching processing is performed using three models ML, MM, and MR.

  As a result, since the matching process is performed by a plurality of models MF in the middle after the start of the accumulation process, the whole circumference image AG is completed earlier compared to the method of the first embodiment in which there is one model MF. can do. In the matching process using a plurality of models MF, the plurality of models MF may be switched in order and the search process for each model MF may be performed by serial processing, but for example, a plurality of models MF are processed using a multi-core CPU 51 It is preferable to perform each search process used in parallel processing.

  (9) In the two-way system, the model acquisition unit 72 moves in one direction (clockwise direction) and the other direction (counterclockwise direction) in the circumferential direction of the container 20 with respect to the position of the model MF acquired first In order to do so, a plurality of (for example, two) models MF are acquired. The accumulation unit 73 performs the matching process using the plurality of models MF located at positions moving in one direction in the circumferential direction and in the other direction with respect to the position of the model MF acquired first. As a result, the frequency of matching in the matching process increases. Therefore, the all-round image AG associated with the position information can be generated relatively quickly.

  (10) In the stepping stone method, the model acquisition unit 72 sets the plurality of models MF such that the position in the circumferential direction of the model MF interposes an empty area where the partial image PG is not accumulated between the model MF and another model MF. get. The accumulation unit 73 performs the matching process using a plurality of models MF in which positions in the circumferential direction of the model MF sandwich an empty area where the partial image PG is not accumulated between the model MF and another model MF. As a result, the frequency of matching in the matching process increases. Therefore, the all-round image AG associated with the position information can be generated relatively quickly.

The embodiment is not limited to the above, and may be modified as follows.
The image acquisition unit is replaced with the imaging device 30, and from the server (not shown) via the external storage device 91 (shown by a two-dot chain line in FIG. 3) in which the image data obtained by imaging the container 20 is stored It may be an input unit 57 (for example, a USB port or a line connection unit) shown in FIG. 3 that reads image data. As described above, the image acquisition unit is the input unit 57, and acquires image data (examination image) from the external storage device 91, or acquires image data (examination image) from the server via the input unit 57. Even in this case, the all-round image AG associated with the position information can be generated relatively easily.

  The partial image PG at the front position of the container 20 in the captured image SG is not limited to the partial image centered on the representative angle RK, and may be a partial image directly on the front around the front angle θc. In the embodiment, the width center of the container 20 in the captured image SG is the front position, and the partial image PG of the width angle WP centered on the representative angle RK is accumulated if there is an error with the representative angle. A partial image PG centered on the front angle θc in SG may be stored. Also in this case, the error between the representative angle RK and the front angle θc may be stored, and the partial image PG centered on the front angle θc may be updated so as to reduce the error. Also in this configuration, if the error is reduced by, for example, reinforcement learning, it is possible to obtain the all-round image AG with relatively high accuracy.

  -At least one of the model data MD and the non-defective item image data GD may not be generated based on the all-round image AG. At least one of the data MD and GD may be generated based on the photographed image SG photographed by the camera 31 while the operator manually adjusts the circumferential position of the container 20.

  The number J of divisions for determining the number of partial images PG constituting the all-round image AG is not limited to 36, which is 36 divisions of 10 degrees each, and may be a natural number of 6 or more. Here, the reason why the division number J should be a natural number of 6 or more is as follows. In order to acquire the position (angle) by matching and create the all-around image AG, it is necessary to ensure that both the master image and the next image are captured. That is, when the master image is toward the end, the limit value when the next image is in front is the image that can be used for inspection in the image acquired from one direction (effective) is 120 degrees, so the front is centered It is ± 60 degrees, and the number obtained by dividing 360 degrees all around by this 60 degrees becomes “6”. Thus, the division number J is preferably a natural number of "6" or more. For example, J = 6, 12, 18, 24, 72, 120, 180, and 360 may be used. For example, 12 divisions divided by 30 degrees, 18 divisions divided by 20 degrees, 24 divisions divided by 15 degrees, 72 divisions divided by 5 degrees, and 120 divisions divided by 3 degrees may be used. Furthermore, it may be 180 divisions or 360 divisions. An appropriate division number J can be selected in consideration of required accuracy and processing speed.

  In the two-way method according to the second embodiment, as in the first embodiment, the angle area to which the front position of the captured image SG determined from the position of the matched model MF belongs is the position of the adjacent unregistered angle area. In this case, a new model MF may be created from the captured image SG of this time.

  The inspection unit 66 of the all-round image generation device may inspect a portion other than the label 22 of the article (for example, the container 20). The pattern of prints, grooves or recesses applied to the outer surface of the article may be inspected.

  The container is not limited to the container on which the label 22 (the design thereof) is attached, but may be a container on which the design is printed on the outer surface of the container, a container in which the design is integrally formed by two-color molding or the like. It can be applied to various containers having a graphic on the outer surface. In this case, the entire circumferential image AG of the entire region in the height direction Z of the container 20 may be generated without being limited to the partial circumferential image AG in the height direction Z of the container 20. In addition, a plurality of all-round images AG having different positions in the height direction Z of the container 20 may be generated.

The matching process is not limited to the shape matching process, and may be another known matching process. For example, correlation matching processing such as normalized correlation matching may be used.
The all-around image generating device may be applied to the article orientation detection device described in Patent Document 2 that detects the direction in which the predetermined position of the article is directed in order to print the predetermined position in the circumferential direction of the article. In this case, the model registration unit 63 creates a plurality of models M from the all-round image AG generated by the all-round image generation unit 62. As a result of the matching process using the model M for the photographed image, the direction of the article is detected based on the position information of the matched similar symbol part and the position information of the model M. Then, the article is rotated until the printed portion, which is a predetermined position of the article, faces the printing device, and printing is performed on the printed portion of the article by the printing device.

  The inspection unit 66 of the all-round image generation apparatus may perform an appearance inspection of the item by comparing a partial image obtained by photographing the article such as the container 20 with the camera 31 with the all-round image AG. Even with this configuration, the all-round image AG used for the appearance inspection can be easily obtained by the all-round image generation unit 62.

  · A conveying device that conveys the container 20 while rotating the container 20 using the speed difference between a pair of belts capable of holding the container 20, or a rotary table method in which the container is placed on the rotary table and rotated, a chuck or a negative pressure adsorption unit Alternatively, the container lifted by the holding portion may be rotated, or one camera 31 may be used to take a plurality of shots during one rotation of the rotating container 20.

  The position specifying unit 82 may not be configured to perform the omni matching process. A plurality of marks may be attached to the container 20 or the label 22 at different positions in the circumferential direction by printing or the like, and the position specifying unit 82 may specify the front position of the container 20 from the positions of the marks in the captured image SG.

  -Instead of the illumination device 32, for example, an illumination device that emits infrared light (infrared radiation) is disposed on the opposite side of the imaging target container 20 to the camera 31, and both reflected light and transmitted light are used. The appearance of the label 22 may be inspected.

  The label is not limited to the shrink label, and may be adhesively bonded to the outer surface of the article. In this case, the label is not limited to resin but may be paper, metal or laminate. Also, the label 22 may not be attached to the entire circumference of the article. The label 22 may be attached, for example, only to a part of the circumferential direction of the article.

  The shape of the container is not limited to a bottomed cylindrical shape, and may be a conical shape, a shape in which a part of the outer peripheral surface is narrowed, a round pot or a square pot. Further, the material of the container is not limited to the synthetic resin, and may be metal or ceramic (for example, glass or ceramic).

  The container may be a cup. As a cup, the cup for instant foods is mentioned, for example. Further, the container is not limited to one intended for containing a beverage liquid or food, but may be one intended for containing a liquid other than food, solid matter, particles or flour. In addition, an empty container may be inspected, or a container containing contents may be inspected.

  -An article is not limited to a container, but it is sufficient if it is an article which has a pattern on the outer surface. For example, it may be an article to which a shrink label is attached, or an article to which a graphic is printed, a groove or a recess or the like. Also, the article may be a part or a product.

  11 Label inspection device as an example of all-round image generation device 12 Transport device 13 Conveyor 20 Container 21 Outer peripheral surface (outer surface) 22 Label 23 Symbol 30 of image acquisition unit An imaging apparatus constituting an example, 31: camera constituting an example of an image acquisition unit, 32: illumination device, 33: container sensor, 40: controller, 41: input device, 42: display unit, 50: computer, 51: CPU , 52: storage unit, 53: position storage unit, 54: first storage unit, 55: second storage unit, 56: third storage unit, 57: input unit constituting an example of an image acquisition unit, 61: control unit 62: all-round image generation unit 63: model registration unit 64: non-defective image generation unit 65: feature image generation unit 66: inspection unit 71: image acquisition processing unit constituting an example of the image acquisition unit 72 ... model acquisition unit, 73 ... storage unit, 4 angle correction unit 75 matching processing unit 81 image acquisition processing unit 82 position specifying unit 83 label height inspection unit 84 detection unit 85 comparison processing unit 86 determination unit 91 ... external storage device, PR ... program, MD ... model data, GD ... non-defective image data, FD ... feature image data, SG ... photographed image, MF, MFc, MFk, ML, MM, MR ... model, AM ... similar symbol part , C0 ... front position, RK ... representative angle, PG ... partial image, AG ... whole circumference image, PS ... progress scale, KG ... inspection image, RG, RG0 to RG35 ... non-defective image, FG, FG0 to FG35 ... characteristic image, DG: Difference image, GA: Model region as an example of image region, M, M1 to M8: Model, FT: Defect, X: Transport direction, Z: Height direction.

Claims (10)

An all-round image generating device for collecting a partial image obtained by photographing a non-defective article and generating an all-round image correlated with position information in the circumferential direction,
An image acquisition unit that sequentially acquires photographed images obtained by photographing an article;
The front position of the article in the one captured image acquired by the image acquisition unit is set as a reference position, and a partial image of the article at the front position of the article is stored in the storage unit in association with circumferential position information. An accumulation unit,
And a model acquisition unit that acquires a model based on a predetermined area of a part of the article in the one captured image acquired by the image acquisition unit.
The storage unit performs a matching process using the model on the captured image after the image acquisition unit acquires the captured image after the acquisition source of the model, and if it matches, the position information of the matched position The front position of the article is determined based on the position information of the model and the model, and a partial image at the front position of the article in the subsequent captured image is stored in the storage unit in association with circumferential position information. If it does not match, it performs accumulation processing to shift to the next captured image,
The model acquisition unit performs model acquisition processing for acquiring a model based on a predetermined region of a part of the article in the captured image when the matching is performed, and the storage unit performs at least using the latest model An all-round image is generated by repeating the accumulation process including the matching process until a plurality of partial images corresponding to circumferential position information are collected at least for the entire circumference. Image generator.   The accumulation unit accumulates the partial image for each area to which a representative position obtained by dividing the circumference of the article is divided, and the partial image is accumulated in the area to which the current front position at the time of matching belongs. If the current front position is closer to the representative position of the area than the front position of the previous partial image, the partial image to be accumulated in the area is the partial image from the previous partial image to the current partial image. The full circle image generation device according to claim 1, wherein the full circle image is updated. The model acquisition unit is configured to generate the plurality of models at positions obtained by moving in one direction in the circumferential direction of the article with respect to positions in the circumferential direction of the model acquired first and at positions moved in the other direction. Get
The all-around image generation device according to claim 1, wherein the storage unit performs the matching process using a plurality of the models. The model acquisition unit acquires a plurality of the models such that a circumferential position of the model sandwiches an empty area in which the partial image is not accumulated between the model and another model.
The all-around image generation device according to claim 1, wherein the storage unit performs the matching process using a plurality of the models.   The storage unit is configured to set a representative position of an unaccumulated area in which the partial image is not accumulated, when the all-round image can not be completed even if the model acquisition process and the accumulation process are repeated until a predetermined condition is satisfied. If there is an accumulated area at a position next to the unaccumulated area, the acquired image from the photographed image which is the acquisition source of the partial image accumulated in the accumulated area is The whole image according to any one of claims 1 to 4, wherein the partial image at a position corresponding to the representative position is acquired, correlated with position information, and accumulated in the unaccumulated area. Round image generator.   A model registration unit is further provided, which creates a plurality of models based on a plurality of image areas having different positions in the circumferential direction selected from the circumferential image, and registers the plurality of models in association with circumferential position information. The all-around image generating device according to any one of claims 1 to 5, characterized in that:   A matching process is performed on a photographed image obtained by photographing an article whose direction is to be specified and having a design on the outer surface, using the model created by the model registration unit, and position information of the matched similar symbol part and the position of the model 7. The all-around image generating device according to claim 6, further comprising a position specifying unit for specifying a front position of the article in the photographed image or a predetermined position in the circumferential direction of the article based on information. . A non-defective image generation unit configured to generate a plurality of non-defective images that are partial images in the circumferential direction from the all-circumferential image corresponding to all the circumferences;
An inspection unit which inspects the presence or absence of a defect of an article to be inspected;
And a position specifying unit for specifying a front position of the article in the photographed image obtained by photographing the article to be inspected.
The inspection unit inspects the presence or absence of a defect of the article based on a comparison result of comparing the inspection image in the photographed image and the non-defective image corresponding to the front position specified by the position specifying unit. The all-around image generation device according to any one of claims 1 to 6, which is characterized by the above. The article is labeled on the outer surface,
The apparatus according to claim 8, wherein the inspection unit inspects the defect of the label. A method for generating an all-round image, which collects a partial image obtained by photographing a non-defective article and generates an all-round image correlated with position information in the circumferential direction,
An image acquisition step of sequentially acquiring photographed images obtained by photographing an article;
A model acquisition step of acquiring a model based on a predetermined area of a part of the article in one of the photographed images acquired in the image acquisition step;
A matching process using the model is performed on the captured image acquired after the captured image of the model acquisition source in the image acquisition step, and if it matches, the position information of the matched position and the position of the model Based on the information, the front position of the article is determined, partial images at the front position of the article in the subsequent captured image are stored in the storage unit in correspondence with circumferential position information, and if they do not match, the next And an accumulation step for performing an accumulation process to shift to a photographed image of
Performing the storage process including the model acquisition step of acquiring a model based on a predetermined region of a part of the article in the captured image when the match is made, and the matching process performed using at least the latest model A method of generating an all-round image, comprising repeatedly repeating the accumulating step until a plurality of partial images corresponding to circumferential position information are collected at least for the entire circumference.