JP7078236B1 - Ball appearance recognition device and ball appearance recognition method - Google Patents

Abstract

An imaging control unit 101 captures a captured image including a ball rolling on a predetermined surface. The detection control unit 102 uses artificial intelligence to detect a ball image, an essential image, and a defective image. When the essential image exists inside the ball image, the essential image integration control unit 103 integrates the number of times the essential image exists as the number of essential image recognitions. When the defective image exists inside the ball image, the defective image integration control unit 104 integrates the number of times the defective image exists as the number of defective image recognitions. The color information calculation control unit 105 calculates color information indicating the surface color of the ball from the ball image. The repetition control unit 106 repeats until the ball image passes through a predetermined recognition line provided in the captured image. The discrimination control unit 107 determines whether the ball of the ball image is a good product ball, a defective product ball, or a dissimilar product ball based on the required image recognition number, the defective image recognition number, and the color information. To determine whether or not. [Selection diagram] Fig. 1

Description

The present invention relates to a ball appearance recognition device and a ball appearance recognition method.

There is a technique of photographing a ball such as a golf ball with a camera and recognizing and selecting a good or defective ball. For example, Japanese Patent Application Laid-Open No. 6-54931 (Patent Document 1) discloses a golf ball device having a dark room, a first stopper, a second stopper, a light irradiation device, a color discrimination device, and a shutter. Has been done. The darkroom has a sloping bottom wall and entrances / exits that slope downward toward the front. The first stopper is provided and movable in the dark chamber, and the second stopper is provided and movable in the dark chamber so as to be located in front of the first stopper. The light irradiating device irradiates light that causes a luminescence effect provided in the dark room so as to be located near the second stopper, and the color identification device is located near the second stopper in the dark room. It is provided in. The shutter opens and closes the ball drop opening formed in the inclined bottom wall in front of the second stopper. In the golf ball sorting device, the shutter opens and closes in response to a color-coded signal from the color discrimination device. As a result, golf goals can be selected efficiently and reliably due to the luminescence action.

Further, in Utility Model Registration No. 3039134 (Patent Document 2), a golfer hits a golf ball at the turn at bat, flies over the lawn, is later collected, washed, dried, and then distributed to the turn at bat again. The equipment of the place is disclosed. In this equipment, the select guide is installed at an angle from the entrance to the exit of the main body of the sorter, the front part of the guide is a V-shaped groove or a U-shaped groove, and the rotation with a claw in the long hole notched at the bottom. Set up your body and send the golf ball from the groove to the central guide. This equipment detects a ball with dirt or scratches by an optical machine, and the remover is operated according to preset evaluation conditions, and the ball is removed on the side of the separate guide. As a result, dirt and scratches caused by the use cycle of the golf ball in the driving range do not reappear in the turn at bat, and after collection and automatic cleaning, by optical detection, mechanically and mechanically based on a certain judgment. , It is possible to sort automatically, reduce the workload of sorting workers, and follow certain criteria that do not differ from person to person.

Further, Japanese Patent Application Laid-Open No. 2004-230325 (Patent Document 3) discloses a golf ball sorting apparatus including an impact plate, a vibration detection sensor, a comparison unit, a vibration characteristic storage unit, and a sorting unit. There is. The impact plate collides with the golf ball under test, and the vibration detection sensor detects the vibration information of the impact plate. The comparison unit compares the vibration information with the vibration characteristics of the reference golf ball, the vibration characteristic storage unit stores the vibration characteristics, and the sorting unit selects the golf ball to be tested based on the comparison result. As a result, it is possible to sort golf balls by detecting the vibration of the impact plate, and it is possible to sort golf balls with high accuracy without being affected by the noise of the surrounding environment.

Further, in Japanese Patent Application Laid-Open No. 2013-34496 (Patent Document 4), a light irradiation means, a photographing means, a dividing means, a concentration measuring means, a scratch determining means, a non-defective product / defective product determining means, and a sorting means A golf ball sorter equipped with the above is disclosed. The light irradiating means irradiates the golf ball with light, and the photographing means photographs the golf ball. The dividing means divides the inspection area of the image data of the golf ball acquired by the photographing means into segment units composed of a plurality of pixels, and the density measuring means measures the density in segment units. The scratch determining means determines the presence or absence of scratches for each segment based on the concentration measurement value, and the non-defective / defective product determining means counts the number of scratches in the inspection area and based on the number of non-defective / defective golf balls. Determine defective products. The sorting means receives the determination result from the non-defective / defective product determining means and sorts the golf ball. As a result, it is possible to automatically sort out what should be reused and what should be discarded with high reliability.

In addition, there is also a technique for discriminating a good product with high accuracy from an image of a ball. For example, Japanese Patent Application Laid-Open No. 9-292349 (Patent Document 5) discloses a visual inspection method for detecting a defective portion generated on the surface of a golf ball having a large number of dimples on the surface. In this method, a predetermined portion of a golf ball is irradiated with light, and in this state, the golf ball is rotated at a constant speed in one direction, and the predetermined portion is rotated along a line orthogonal to the rotation direction of the ball with a line sensor camera. A two-dimensional image is obtained from the image data captured by the line sensor camera, and the change in brightness in the rotation direction of the ball appearing on the two-dimensional image is converted into a change amount for each predetermined unit, and the obtained change amount data is obtained. By setting a threshold value between the amount of change in the dimple portion and the amount of change in the defective portion and performing binarization processing, the presence or absence of the defective portion is detected. This makes it possible to reliably and accurately identify the dimple portion and the defective portion of the golf ball and accurately detect only the defective portion, and further, according to the second inspection method, the marking is made in the dimple. Even if the golf ball is marked with a mark, it is possible to reliably distinguish between the stamp and the defective part and accurately detect only the defective part.

Further, Japanese Patent Application Laid-Open No. 2002-767 (Patent Document 6) describes a seam position determination device for determining whether or not a seam is in a predetermined position in order to remove burrs existing on the seam of a golf ball after molding. Is disclosed. This device includes a camera for shooting a golf ball and a calculation means, and the calculation means is configured to determine the presence or absence of burrs in a predetermined area of the image taken by the camera. As a result, the determination accuracy is high and the defect rate is reduced.

Japanese Unexamined Patent Publication No. 6-54931 Utility Model Registration No. 3039134 Japanese Unexamined Patent Publication No. 2004-230325 Japanese Unexamined Patent Publication No. 2013-34496 Japanese Unexamined Patent Publication No. 9-292349 JP-A-2002-767

At driving ranges and golf courses, various golf balls launched by players are collected at ball collection sites. The collected golf balls include good golf balls with a specific logo specified by the golf practice field, golf balls with scratches due to the player's practice, golf balls that have faded due to long-term use, and the player's possession. Defective products such as golf balls with other logos and golf balls of different types are mixed.

Here, if a golf practice field or the like reuses a defective or different type of golf ball for a player, it is subject to a complaint. Therefore, at the ball collection place, a worker currently visually checks the collected golf ball. We are conducting golf ball sorting work by discarding defective and different types of golf balls and reusing good golf balls.

However, such sorting work of golf balls has a problem that it takes time and labor for the worker. In addition, since the selection of golf balls mainly relies on the visual inspection of workers, there is a possibility that defective or different types of golf balls may be misidentified or misselected, and a wide variety of golf balls may be selected. There is a problem that the accuracy is limited.

Since the technique described in Patent Document 1 is premised on the luminescence action, there is a problem that it is necessary to coat a good golf ball with a paint having the luminescence action. Although the technique described in Patent Document 2 detects dirt and scratches on a golf ball with an optical machine, the detection by the optical machine is performed only on a part of the golf ball, and there is dirt or the like on the other part of the golf ball. There is a problem that dirt and the like of the golf ball cannot be detected properly. The technique described in Patent Document 3 has a problem that it is not possible to optically detect the surface of a golf ball because the golf ball is selected based on the vibration characteristics of the reference golf ball.

Further, the technique described in Patent Document 4 has a problem that the logo of a golf ball cannot be determined because the presence or absence of scratches on the golf ball is determined based on the image data of the golf ball. In the technique described in Patent Document 5, since a defective portion on the surface of the golf ball is detected based on the image data of the golf ball, there is a problem that the logo of the golf ball cannot be determined as described above. Further, in the technique described in Patent Document 6, since the burr on the seam of the golf ball is determined based on the image data of the golf ball, it is not possible to determine the logo of the golf ball as described above. There is.

Therefore, the present invention has been made to solve the above-mentioned problems, and is a ball appearance recognition device and a ball capable of accurately discriminating various types of balls based on a ball image of a rolling ball. The purpose is to provide an appearance recognition method.

The ball appearance recognition device according to the present invention includes a photographing control unit, a detection control unit, an essential image integration control unit, a defective image integration control unit, a color information calculation control unit, a repetition control unit, and a discrimination control unit. , Equipped with. The shooting control unit shoots a shot image including a ball rolling on a predetermined surface. The detection control unit has artificial intelligence in which a learning ball image corresponding to the ball, a learning essential image showing a genuine product of the ball, and a learning defective image including scratches or stains on the ball are learned in advance. It is used to detect a ball image, an essential image, and a defective image included in the captured image. When the essential image exists inside the ball image, the essential image integration control unit integrates the number of times the essential image exists as the number of essential image recognitions. When the defective image exists inside the ball image, the defective image integration control unit integrates the number of times the defective image exists as the number of defective image recognitions. The color information calculation control unit calculates color information indicating the surface color of the ball from the ball image. The repeat control unit captures the captured image, detects the ball image, and integrates the number of required image recognitions until the ball image passes through a predetermined recognition line provided in the captured image. The integration of the number of defective images recognized and the calculation of the color information are repeated. When the repetition is completed, the discrimination control unit determines whether the ball of the ball image is a good product ball, a defective product ball, or a different type based on the required image recognition number, the defective image recognition number, and the color information. Determine whether it is a product ball.

Further, the ball appearance recognition method of the ball appearance recognition device according to the present invention is the same as each control unit of the ball appearance recognition device, that is, a shooting control step, a detection control step, an essential image integration control step, and a defective image integration control. It includes a process, a color information calculation control process, a repeat control process, and a discrimination control process.

According to the present invention, it is possible to discriminate various types of balls with high accuracy based on a ball image of a rolling ball.

It is a functional block diagram of the ball appearance recognition device which concerns on this invention. It is a flowchart which shows the execution procedure of the ball appearance recognition method which concerns on this invention. It is a figure (FIG. 3A) which shows an example of the relationship between a ball rolling on a predetermined surface, and a figure (FIG. 3B) which shows an example of the installation of the detection of a ball image and the required number of image recognition. A diagram showing an example when an essential image is detected by artificial intelligence and existing inside the ball image (FIG. 4A), and a diagram showing an example when an essential image is not detected by artificial intelligence (FIG. 4B). Is. A diagram showing an example when a scratch defect image is detected by artificial intelligence and exists inside the ball image (FIG. 5A), and an example when a stain defect image is detected by artificial intelligence and exists inside the ball image. It is the figure (FIG. 5B) which shows. A diagram (FIG. 6A) showing an example of calculating color information from a ball image of a relatively new ball, and a diagram (FIG. 6B) showing an example of calculating color information from a ball image of a relatively old ball. Is. An example of displaying an essential image recognition number of a ball image in the first captured image (FIG. 7A) and an example of detecting a ball image in the second captured image and displaying an essential image recognition number of the ball image. (FIG. 7B) and. A diagram (FIG. 8A) showing an example of detecting a ball image in the third captured image and displaying the required number of image recognitions of the ball image, and displaying the required number of image recognitions of the ball image in the nth captured image and a good product. It is a figure (FIG. 8B) which shows an example of the discrimination of a ball. A diagram (FIG. 9A) showing an example of displaying the required image recognition number of a ball image in a captured image and discriminating defective balls, displaying the required image recognition number of a ball image in a captured image, and discriminating between different types of balls. It is a figure (FIG. 9B) which shows an example. A diagram (FIG. 10A) showing an example of displaying the required image recognition number of the ball image in the captured image and discriminating the different product ball, and displaying the required image recognition number of the ball image in the captured image and discriminating the different product ball. It is a figure (FIG. 10B) which shows an example. It is a plan view (FIG. 11A) and a right side view (FIG. 11B) of a sorting apparatus using a rotary plate. A diagram (FIG. 12A) showing an example of displaying the required number of ball images recognized in the captured image of the embodiment, displaying the required number of recognized images of the ball image in the captured image of the embodiment, and determining the defective ball. It is a figure (FIG. 12B) which shows an example.

Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings for the purpose of understanding the present invention. It should be noted that the following embodiment is an example embodying the present invention and does not limit the technical scope of the present invention.

As shown in FIG. 1, the ball appearance recognition device 1 according to the present invention includes a camera 10 and a terminal device 11. Here, the configuration of the camera 10 is not particularly limited, and examples thereof include a normal camera capable of continuous shooting and a high-performance camera capable of continuous shooting at high resolution and high speed. The camera 10 of the present invention is basically composed of one camera, but may be composed of a plurality of (for example, two) cameras.

A predetermined surface 10a (for example, a rail) on which the ball B can roll is provided in the vicinity of the camera 10. The camera 10 is fixed toward a part of the surface 10a, and a part of the surface 10a is photographed as a photographing area A. When the ball B rolls on the surface 10a, the camera 10 photographs the ball B rolling on a part of the surface 10a in the photographing region A.

Here, the type of the ball B is not particularly limited, and for example, a golf ball can be mentioned. Golf balls are discriminated according to a predetermined number of types. Here, for example, a golf ball is discriminated as a non-defective ball, a defective ball, or a heterogeneous ball when a predetermined discrimination condition is satisfied (described later).

The configuration of the terminal device 11 is not particularly limited, and examples thereof include a desktop type terminal device, a portable terminal device, and a portable notebook computer. The terminal device 11 includes, for example, a display unit (output unit) for displaying a screen, a reception unit (input unit) for receiving input of a predetermined instruction by a user operation, and a control unit for executing each process. There is.

By the way, the ball appearance recognition device 1 may further include a sorting device 12 for sorting the discriminated balls B. When, for example, the ball appearance recognition device 1 discriminates the ball B into a good ball, a defective ball, or a different kind of ball, the sorting device 12 may have a plurality of balls B according to the type of the discriminated ball B. It is a device that discharges to a predetermined discharge part among the discharge parts of. Here, for example, the sorting device 12 changes the moving direction of the ball B rolling on the surface 10a, moves the ball B in another moving direction, and discharges the ball B to a discharging unit according to the type of the ball B.

Here, the configuration of the sorting device 12 is not particularly limited, but for example, it is provided in the rolling direction of the ball B in the shooting area A of the camera 10, and the ball appearance recognition device 1 sets the ball B as a good ball or a defective ball. When it is determined to be a different type of ball, the ball is blown out from the surface 10a toward the ball B rolling out from the photographing area A in a direction perpendicular to the moving direction of the ball B. Move B in another direction of movement. In FIG. 1, for example, when the ball B is determined to be a defective ball or a different kind of ball, the sorting device 12 is configured to blow the ball B with air in a direction perpendicular to the moving direction of the ball B. Note that this configuration is an example of a ball sorting method, and a wide variety of sorting methods can be adopted.

The terminal device 11 has a built-in CPU, GPU, ROM, RAM, HDD, SSD, etc. (not shown), and the CPU or GPU uses, for example, the RAM as a work area and stores it in the ROM, HDD, SSD, etc. Run the program that has been created. Further, for each control unit described later, each control unit is realized by the CPU executing a program.

Next, the configuration and the execution procedure according to the embodiment of the present invention will be described with reference to FIGS. 1 to 10. First, when the user turns on the power of the terminal device 11 of the ball appearance recognition device 1, the terminal device 11 is activated and the shooting control unit 101 of the terminal device 11 starts shooting.

Here, when the user connects one of the predetermined surfaces 10a in which the photographing region A of the camera 10 exists to the discharge port of the discharge device (for example, a ball cleaning device) from which a wide variety of balls B are discharged, this surface is connected. The ball B rolls from one side of 10a toward the other side. Therefore, the photographing control unit 101 photographs the ball B rolling on the surface 10a by using the camera 10 (FIG. 2: S101).

Here, the type of the surface 10a is not particularly limited, but for example, as shown in FIG. 3A, an inclined surface 10a inclined from above to downward may be used, or a flat surface 10a along the horizontal direction may be used. On the inclined surface 10a, the ball B rolls down from the upper side of the inclined surface 10a by connecting the upper part of the inclined surface 10a to the discharge port of the discharging device. On the other hand, in the plane 10a, one of the planes 10a is connected to the discharge port of the discharge device, and the ball B discharged from the discharge port is a plane depending on its own force or a moving means for moving the ball B from the outside. It rolls and moves from one side of 10a to the other side.

Further, the photographing method of the photographing control unit 101 is not particularly limited, but the camera 10 is installed directly above the surface 10a regardless of whether the surface 10a is an inclined surface or a flat surface, and a part of the surface 10a is covered with the photographing area A. Take a picture as. The shooting control unit 101 uses the camera 10 to shoot the ball B passing through the shooting region A of the surface 10a.

Here, lighting units for irradiating the photographing region A with light may be provided on either or both sides of the camera 10. As a result, the surface of the ball B existing in the photographing area A can be illuminated with light, and the photographed image of the ball B can be clearly photographed.

When the photographing control unit 101 photographs the ball B, the detection control unit 102 of the terminal device 11 receives a learning ball image corresponding to the ball B, a learning essential image showing a genuine product of the ball B, and a scratch on the ball. Alternatively, the ball image BP, the essential image RP, and the defective image NP included in the captured image C are detected by using artificial intelligence in which the learning defective image including dirt is learned in advance (FIG. 2: S102).

Here, the detection method of the detection control unit 102 is not particularly limited, but for example, when the first (first) captured image C1 is captured, the detection control unit 102 first captures the first captured image C1. The first captured image C1 is input to the artificial intelligence provided in advance by acquiring the image.

Artificial intelligence is an input image when the feature amount of the input input image is the same as or close to the feature amount of the learning reference image based on the feature amount (numerical data, parameters, etc.) of the reference image for learning. It is a tool that classifies and detects images as a reference image, so to speak, it functions as a classifier. For artificial intelligence, for example, machine learning or deep learning may be adopted, or a support vector machine, a neural network, a classification method based on a predetermined rule, or the like may be adopted.

In the present invention, as shown in FIG. 3B, the learning ball image 30 corresponding to the ball B to be detected is input in advance as teacher data in the artificial intelligence 300, so that the artificial intelligence 300 can use the learning ball image. 30 feature quantities are extracted and the feature quantities are learned. Then, when the detection control unit 102 inputs the first captured image C1 to the artificial intelligence 300, the artificial intelligence 300 extracts the feature amount of the target image from the predetermined target image inside the first captured image C1. When the feature amount of the extracted target image matches or approximates the feature amount of the trained learning ball image 30, the target image is classified as a ball image and detected. On the other hand, if the feature amount of the target image does not match or approximate the feature amount of the learning ball image 30, no object is detected. Then, the detection control unit 102 appropriately changes the size of the target image inside the first captured image C1, so that even if ball image BPs of various sizes are present in the first captured image C1. The ball image BP can be detected.

Here, when the detection control unit 102 detects the ball image BP, as shown in FIG. 3B, a quadrangle 31 circumscribing the outer peripheral circle of the ball image BP may be provided. Here, a column 32 for the number of required image recognitions is provided on the lower right side of the quadrangle 31, a column 33 for the number of defective images recognized is provided on the lower left side of the quadrangle 31, and a ball image is provided above the quadrangle 31. A BP color information column 34 is provided. Above the defective image recognition number column 33, a scratch defective image recognition number column 33a is provided among the defective image recognition numbers, and below the defective image recognition number column 33, among the defective image recognition numbers. A column 33b for the number of image recognition defects is provided. For example, "0" is set as an initial value in the required image recognition number column 32, the scratch defective image recognition number column 33a, the stain defective image recognition number column 33b, and the color information column 34. .. This makes it possible to check various numerical values at a glance. The positions and display forms of the required image recognition number column 32, the scratch defect image recognition number column 33a, the stain defect image recognition number column 33b, and the color information column 34 are not particularly limited. The position and display form may be used, and the design may be changed as appropriate.

When the detection control unit 102 detects the ball image BP, it determines whether or not the detected ball image BP is equivalent to the ball image in the immediately preceding captured image (FIG. 2: S103). As the immediately preceding captured image, for example, a captured image one frame before the first captured image C1 can be mentioned. Here, since the ball image BP was detected for the first time, the detection control unit 102 determined that it was not equivalent to the ball image in the immediately preceding captured image (FIG. 2: S103NO), and identified the ball image BP detected for the first time (FIG. 2: S103NO). An ID) (for example, "a001") is given (FIG. 2: S104). This makes it possible to identify a specific ball image BP. Further, the detection control unit 102 associates the assigned identification information with the required screen recognition number, the defective image recognition number (scratch defective image recognition number, stain defective image recognition number), and color information.

When the identification information (“a001”) is added to the ball image BP, the detection control unit 102 provides an identification information column 35 on the upper right side of the quadrangle 31, as shown in FIG. 3B, for identification. Information (“a001”) may be displayed. This makes it possible to confirm the identity of the ball image BP at a glance.

By the way, as shown in FIG. 4A, in addition to the learning ball image 30, the artificial intelligence 300 has previously learned the learning essential image 40 showing the genuine product of the ball B. Therefore, in S102, the detection control unit 102 Detects the essential image RP when the essential image RP is present in the first captured image C1. Here, the learning essential image 40 is an image showing essential items for the ball B to be a genuine product, and examples thereof include a company name, a logo, a name, and an identification number. In FIG. 4A, an image of the logo “Logo” of the ball B and an image of an identification number from “001” to “999” are shown as essential learning images 40. When the essential image 40 for learning is input to the artificial intelligence 300 as teacher data, the artificial intelligence 300 extracts the feature amount of the essential image 40 for learning and learns the feature amount in the same manner as described above.

Here, in FIG. 4A, since the ball image BP of the first captured image C1 shows the logo “Logo” of the ball B equivalent to the essential image 40 for learning, the detection control unit 102 uses artificial intelligence. By 300, the essential image RP that matches or approximates the feature amount of the trained essential image 40 for learning is detected.

Next, when the detection control unit 102 detects the essential image RP, the essential image integration control unit 103 of the terminal device 11 is required when the essential image RP exists inside the ball image BP (FIG. 2: S105YES). The number of times the image RP exists is integrated as the required number of image recognitions (FIG. 2: S106).

Here, there is no particular limitation on the integration method of the essential image integration control unit 103. For example, when the essential image RP is detected in the first captured image C1, the essential image integration control unit 103 determines the center coordinates of the detected essential image RP, the vertical and horizontal sizes of the essential image RP, and the detected ball image. Based on the center coordinates of the BP and the vertical and horizontal sizes of the ball image BP, it is determined whether or not the essential image RP exists inside the ball image BP (FIG. 2: S105). The center coordinates here are based on the origin of the camera coordinate system in the captured image, and the size is based on the number of pixels (number of pixels) constituting the captured image. Specifically, the essential image integration control unit 103 determines whether the vertical size of the essential image RP is within the vertical size of the ball image BP and the horizontal size of the essential image RP is within the horizontal size of the ball image BP. Is determined.

As a result of the determination, as shown in FIG. 4A, when the required image RP exists inside the ball image BP (FIG. 2: S105YES), the required image integration control unit 103 has the number of times the required image RP exists in the required image recognition number. "1" is added to add up (count) the number of required image recognitions (FIG. 2: S106).

Here, since "0" is set as an initial value for the initial required image recognition number, the essential image integration control unit 103 sets the required image recognition number "0" to "1" for the number of times the required image RP exists. The required image recognition number is "1", and the required image recognition number "1" is displayed in the required image recognition number column 32 in the square 31 of the ball image BP. As a result, the user can easily confirm the number of times the required image RP exists by looking at the column 32 of the required image recognition number of the first captured image C1.

On the other hand, as a result of the determination, when the essential image RP does not exist inside the ball image BP (FIG. 2: S105NO), in this case, the essential image RP is other than the ball image BP for some reason (erroneous detection of noise or the like). This is the case where the ball B is present in the first captured image C1 of the above, and cannot be used as an image for discriminating the ball B. Therefore, the essential image integration control unit 103 completes the process as it is without integrating the number of times the essential image RP exists. That is, the number of times the required image RP exists is not added to the number of required image recognition, and the number of required image recognition remains at the initial value of "0". As a result, the number of times the essential image RP for discriminating the ball B exists can be integrated with high accuracy.

On the other hand, as shown in FIG. 4B, when the ball image BP shows the ball B logo “Ball” completely different from the learning essential image 40, the detection control unit 102 artificially creates the first captured image C1. Even if it is input to the intelligence 300, the feature amount of the logo "Ball" does not match or approximate the feature amount of the essential image 40 for learning, so that the process is completed without detecting any object. In this case, since the essential image RP is not detected, the essential image integration control unit 103 determines that the essential image RP does not exist inside the ball image BP (FIG. 2: S105NO), and completes the process as it is. As described above, when the essential image RP does not exist, the number of essential image recognitions does not increase, so it is possible to determine whether or not the ball B is a desired ball by the degree of increase in the number of essential image recognitions.

Further, as shown in FIG. 5A, in the artificial intelligence 300, in addition to the learning ball image 30, a learning scratch defect image 50 showing a scratch provided on the ball B is pre-learned as a learning defect image. In S102, if the first captured image C1 has a scratch-defective image NP that is a defective image, the detection control unit 102 detects the scratch-defective image NP. Here, as the learning scratch defect image 50, for example, an image of a cut, an image of a scratch, or the like can be mentioned. When the learning scratch defect image 50 is input to the artificial intelligence 300 as teacher data, the artificial intelligence 300 extracts the feature amount of the learning scratch defect image 50 and learns the feature amount in the same manner as described above.

Here, in FIG. 5A, since the ball image BP of the first captured image C1 has the same scratches on the ball B as the learning scratch defect image 50, the detection control unit 102 uses the artificial intelligence 300. , The scratch defect image NP that matches or approximates the feature amount of the trained scratch defect image 50 for learning is detected.

When the detection control unit 102 detects the scratch defective image NP, the defective image integration control unit 104 of the terminal device 11 determines that the defective image NP exists inside the ball image BP (FIG. 2: S107YES). Is integrated as the number of defective image recognitions (FIG. 2: S108).

Here, there is no particular limitation on the integration method of the defective image integration control unit 104. For example, when the scratch defective image NP is detected in the first captured image C1 as a defective image, the defective image integration control unit 103 determines the center coordinates of the detected scratch defective image NP and the vertical and horizontal sizes of the scratch defective image NP. Based on the detected center coordinates of the ball image BP and the vertical and horizontal sizes of the ball image BP, it is determined whether or not the scratch defective image NP exists inside the ball image BP (FIG. 2: S107). This determination method is the same as described above.

As a result of the determination, as shown in FIG. 5A, when the scratch defect image NP exists inside the ball image BP (FIG. 2: S107YES), the defect image integration control unit 104 determines the scratch defect image NP in the number of scratch defect image recognitions. The number of scratch-defective image recognitions is integrated by adding "1" of the number of times of existence (FIG. 2: S108). Here, since "0" was set as the initial value for the initial number of scratch-defective image recognitions, the defective image integration control unit 104 sets the number of scratch-defective image recognition numbers to "0" and the number of times the scratch-defective image NP exists. “1”, which is the sum of “1”, is defined as the number of scratch-defective image recognitions, and “1” of the scratch-defective image recognition number is displayed in the field of scratch-defective image recognition number column 33a in the square 31 of the ball image BP. As a result, the user can easily confirm the number of times the scratch-defective image 50 exists by looking at the column 33a of the scratch-defective image recognition number of the first captured image C1.

On the other hand, as a result of the determination, when the scratch defective image NP does not exist inside the ball image BP (FIG. 2: S107NO), in this case, the scratch defective image is caused by some cause (erroneous detection of noise or the like) as described above. This is the case where the NP exists in the first captured image C1 other than the ball image BP, and cannot be used as an image for discriminating the ball B. Therefore, the defective image integration control unit 104 completes the process as it is without integrating the number of times the scratch defective image NP exists. That is, the number of times the scratch-defective image NP exists is not added to the scratch-defective image recognition number, and the scratch-defective image recognition number remains at the initial value of "0". As described above, when the scratch-defective image 50 showing scratches does not exist inside the ball image BP, the number of scratch-defective image recognitions does not increase. Can be determined.

By the way, in the above description, the scratch defective image is adopted as the defective image, but other defective images may be used. For example, as shown in FIG. 5B, the defective image may be an image of stains showing stains provided on the ball B. In FIG. 5B, in addition to the learning ball image 30, the artificial intelligence 300 has previously learned a learning stain defect image 51 showing stains provided on the ball B as a learning defect image, and is a learning stain defect image. Reference numeral 51 may include, for example, an image of light stains, an image of dark stains, and the like. When the learning stain defect image 51 is input to the artificial intelligence 300 as teacher data, the artificial intelligence 300 extracts the feature amount of the learning stain defect image 51 and learns the feature amount in the same manner as described above.

Here, in FIG. 5B, since the ball image BP of the first captured image C1 has the same dirt on the ball B as the learning dirt defect image 51, the detection control unit 102 uses the artificial intelligence 300. , The stain defect image NP that matches or approximates the feature amount of the trained stain defect image 51 for learning is detected.

Next, when the detection control unit 102 detects the stain defect image NP, the defect image integration control unit 104 determines that the stain defect image NP exists inside the ball image BP as described above (FIG. 2: S107YES). ), The number of times the dirty defective image NP exists is integrated as the number of recognized dirty defective images (FIG. 2: S108).

Here, in FIG. 5B, since the dirty defective image NP exists inside the ball image BP, the defective image integration control unit 104 adds “1” of the number of times the dirty defective image NP exists to the number of recognized dirty defective images. Then, the number of image recognitions for defective stains is integrated (FIG. 2: S108). Also in this case, since the initial value of the number of recognized stain-defective images is set to "0", the defective image integration control unit 104 sets the number of stain-defective image recognition numbers to "0" and the number of times the stain-defective image NP exists. “1”, which is the sum of “1”, is defined as the number of defective stain image recognition, and “1” of the number of defective stain image recognition is displayed in the column 33b of the number of defective stain image recognition in the square 31 of the ball image BP.

On the other hand, as a result of the determination, when the dirty defective image NP does not exist inside the ball image BP (FIG. 2: S107NO), the defective image integration control unit 104 processes as it is without accumulating the number of times the dirty defective image NP exists. Will be completed.

In S102, since the learning ball image 30, the learning essential image 40, and the learning defective images 50 and 51 are previously learned by the artificial intelligence 300, the captured image C includes the ball image BP and the essential image RP. If all the defective image NPs are present, the detection control unit 102 may detect all of the ball image BP, the essential image RP, and the defective image NP at once by the artificial intelligence 300. Is also good.

When the defective image integration control unit 104 completes the integration of the number of defective image recognitions, the color information calculation control unit 105 of the terminal device 11 calculates color information indicating the surface color of the ball B from the ball image BP (FIG. FIG. 2: S108).

Here, the calculation method of the color information calculation control unit 105 is not particularly limited, but for example, as shown in FIG. 6A, the color information calculation control unit 105 refers to the center image 60 of the detected ball image BP. The color information of each pixel constituting the referenced center image 60 is acquired.

Here, the color information is represented by, for example, the numerical values of the three primary colors R (red), G (green), and B (blue), and the white color information is R (255), G (255), B ( It is represented as 255), and the black color information is represented as R (0), G (0), and B (0). For example, white color information is converted from R (255), G (255), B (255) to "1", and black color information is converted into R (0), G (0), B (0). By converting from "0" to "0", the numerical values of the three primary colors of the color information can be expressed within the range of "0" to "1". When the normal color of the ball B is set to white, the higher the numerical value of the color information, the closer to "1", and the color of the color information approaches the white of the normal color of the ball B. As the color information, the numerical values of the three primary colors as they are may be used, or as described above, the numerical values of the three primary colors may be converted and calculated according to a predetermined rule.

Here, when the ball image BP shown in FIG. 6A is relatively new and is close to the normal color of the ball B, the color information calculation control unit 105 receives the color information “0” from the central image 60, which is close to the white color information. .82 ”is acquired. Then, the color information calculation control unit 105 displays the color information "0.82" in the color information column 34 of the ball image BP in the quadrangle 31 of the ball image BP.

On the other hand, when the ball image BP shown in FIG. 6B is relatively old and is dirty away from the normal color of the ball B, the color information calculation control unit 105 displays the color information close to the black color information from the central image 60. "0.28" is calculated, and the color information "0.28" is displayed in the color information column 34 of the ball image BP in the square 31 of the ball image BP. By providing the color information column 33 in this way, the user can easily confirm the color information of the ball image BP at first glance at the color information column 34 of the first captured image C1.

In the above description, the color information calculation control unit 105 acquires the color information only once from the central image 60, but as another method, for example, the color information may be acquired from the image of another part of the ball image BP. Color information may be acquired from the image of each part of the ball image BP, and the average value of the color information of the image of each part may be acquired.

Further, in the above, the numerical values of the three primary colors R, G, and B are adopted as the color information, but the numerical values of the L * a * b * color system and the numerical values of the XYZ color system are adopted as other color information. Alternatively, these numerical values may be combined.

When the color information calculation control unit 105 completes the calculation of the color information, the column 32 of the required image recognition number of the ball image BP in the first captured image C1, the column 33a of the number of scratch defective image recognition, and the stain defective image Numerical values are displayed in the recognition number column 33b and the color information column 34, respectively. In the ball image BP of the first captured image C1 shown in FIG. 3A, as shown in FIG. 7A, "1" is displayed in the required image recognition number column 32, and the scratch defect image recognition number column 33a and the stain defect image. "0" is displayed in each of the recognition number column 33b, and "0.82" is displayed in the color information column 34.

In the above description, the three processes are performed in the order of integration of the number of required image recognition (S106), integration of the number of defective image recognition (S108), and acquisition of color information (S109). The order is not particularly limited, and even if they are interchanged with each other, the same action and effect will be obtained.

When the above three processes are completed, the repeat control unit 106 of the terminal device 11 captures the captured image C until the ball image BP passes through a predetermined recognition line L provided in the captured image C. S101), detection of ball image BP (S102), integration of essential image recognition numbers (S106), integration of defective image recognition numbers (S108), and calculation of color information (S109) are repeated (FIG. 2: 2: S110).

Here, the repeating method of the repeating control unit 106 is not particularly limited, but for example, as shown in FIG. 7A, the captured image C corresponding to the photographing area A photographed by the camera 10 has a predetermined recognition set in advance. A line L is provided. The recognition line L is provided on the lower side of the inclined surface 10a, and is provided for recognizing that the ball B has rolled downward from the upper side of the inclined surface 10a.

Here, the repeat control unit 106 determines whether or not the ball image BP of the first captured image C1 has passed the recognition line L. At the time of FIG. 7A, the ball image BP is still located above the inclined surface 10a and has not passed through the recognition line L. Therefore, in the repeat control unit 106, the ball image BP of the first captured image C1 is the recognition line. It is determined that the ball has not passed through L (FIG. 2: S110NO). Then, the repetitive control unit 106 returns to S101 and causes the imaging control unit 101 to capture a captured image including the ball B (FIG. 2: S101).

Here, the photographing control unit 101 photographs the rolling ball B, and obtains the second (next) captured image C2 of the first captured image C1. Here, as shown in FIG. 7B, the ball B is By rolling, a surface different from the surface of the ball B appears. Then, the detection control unit 102 detects a ball image BP different from the previous ball image BP by the artificial intelligence 300 in the second captured image C2 (FIG. 2: S102). If the essential image RP is present in the second captured image C2, the detection control unit 102 will also detect this essential image RP by the artificial intelligence 300. Then, the detection control unit 102 determines whether or not the ball image BP detected by the second captured image C2 is equivalent to the ball image BP in the first captured image C1 immediately before (FIG. 2: S103). ..

Here, the determination method of the detection control unit 102 is not particularly limited, but for example, the number of overlapping pixels in which the ball image BP of the first captured image C1 and the ball image BP of the second captured image C2 overlap is calculated. , It is determined whether or not the calculated number of overlapping pixels is equal to or greater than a predetermined threshold value.

In FIG. 7B, since the number of overlapping pixels 70 between the ball image BP of the first captured image C1 and the ball image BP of the second captured image C2 is equal to or greater than a predetermined threshold value, the detection control unit 102 performs the second imaging. It is determined that the ball image BP of the image C2 is equivalent to the ball image BP of the first captured image C1 (FIG. 2: S103YES), and the ball image BP of the second captured image C2 is the first captured image C1. The identification information (“a001”) of the ball image BP is handed over (FIG. 2: S111). As a result, in a plurality of continuously shot images, the ball image BPs that partially overlap are treated as the same ball image BP, from the first shot until the predetermined recognition line L is passed. It becomes possible to track the same ball B. Further, by passing the same identification number to the ball image BP of the second captured image C2, the required image recognition number, the scratch defective image recognition number, the stain defective image recognition number, and the color information associated with this identification number are also obtained. It will be associated with the ball image BP of the second captured image C2 (data transfer). As a result, the data of the ball image BP of the first captured image C1 up to now is taken over by the ball image BP of the second captured image C2, so that the surface information of the same ball B can be left as a history.

On the other hand, when the number of overlapping pixels 70 between the ball image BP of the first captured image C1 and the ball image BP of the second captured image C2 is less than a predetermined threshold value, the ball B has some reason (accident or false detection). Since the identity may be impaired, the detection control unit 102 determines that the ball image BP of the second captured image C2 is not equivalent to the ball image BP of the first captured image C1 (FIG. 2: S103NO). ), A new identification information (ID) (for example, "0002") is added to the ball image BP of the second captured image C2 (FIG. 2: S104). As a result, when the identity of the ball B is impaired, the ball image BP can be identified by newly adding the identification information.

By the way, in FIG. 7B, since the image of the identification number “001” corresponding to the essential image RP appears inside the ball image BP in the second captured image C2 (FIG. 2: S105YES), the same as described above. The essential image integration control unit 103 integrates the number of required image recognitions (FIG. 2: S106), and adds "1", which is the number of times the required image RP exists, to "1", which is the number of required image recognitions. It is displayed in the recognition number column 32. On the other hand, since the ball image BP does not have a scratch defect image or a stain defect image, the defect image integration control unit 104 does not integrate the defect image recognition number (FIG. 2: 107NO), and the scratch defect image recognition number column 33a. "0" in the column 33b of the number of image recognitions with poor stains remains unchanged. Further, the color information calculation control unit 105 calculates the color information "0.82" (FIG. 2: S109) and displays "0.82" in the color information column 34.

Then, the repeat control unit 106 determines whether or not the ball image BP of the second captured image C2 has passed the recognition line L (FIG. 2: S110), but at the time of FIG. 7B, the ball image BP is determined. It is still located above the inclined surface 10a and has not passed through the recognition line L. Therefore, the repeat control unit 106 determines that the ball image BP of the captured image C has not passed through the recognition line L (FIG. 2: S108NO), and returns to S101.

Here, as shown in FIG. 8A, when the ball B rolls further, a surface different from the previous surface appears. Then, the detection control unit 102 detects a ball image BP different from the previous ball image BP by the artificial intelligence 300 in the third captured image C3. Since the rolling of one ball B is continuously photographed, the ball image BP detected by the detection control unit 102 in the third captured image C3 is equivalent to the ball image BP in the second captured image C2. (FIG. 2: S103YES), and the identification information (“a001”) of the ball image BP of the second captured image C2 is handed over to the ball image BP of the third captured image C3 (FIG. 2: S111).

Then, in FIG. 8A, since the image of the logo “Logo” corresponding to the essential image RP appears inside the ball image BP in the third captured image C3 (FIG. 2: S105YES), it is essential as in the above. The image integration control unit 103 integrates the number of required image recognitions (FIG. 2: S106), and adds "1", which is the number of times the required image RP exists, to "2", which is the number of required image recognitions, to recognize "3". It is displayed in the number column 32. Further, since the ball image BP does not have a scratch defect image or a stain defect image, the defect image integration control unit 104 does not integrate the defect image recognition number (FIG. 2: 107NO), and the scratch defect image recognition number column 33a. "0" in the column 33b of the number of image recognitions with poor stains remains unchanged. Further, the color information calculation control unit 105 acquires the color information "0.82" (FIG. 2: S109) and displays "0.82" in the color information column 34.

In this way, the processes from S101 to S109 are repeated until the ball image BP passes through the recognition line L. Here, since the ball image BP corresponding to various surfaces of the ball B appears by rolling the ball B, when the essential image RP appears with the rotation of the ball B, the essential image RP is required as many times as the number of times the essential image RP exists. The number of recognitions is accumulated. Further, when the scratch-defective image NP appears, the number of scratch-defective image recognitions is accumulated by the number of times the scratch-defective image NP exists, and when the stain-defective image NP appears, the number of stain-defective image recognitions is accumulated by the number of times the stain-defective image NP exists. Will be done. Color information is also calculated each time. Thereby, by inspecting various surfaces of the ball B by the rotation of the ball B, it is possible to discriminate various types of the ball B with high accuracy.

In the above description, the color information is acquired (S109) for each captured image, but the color information may be acquired (S109) for only one captured image and omitted thereafter. Then, color information may be acquired for a predetermined number of captured images at a predetermined timing, and the average value of the color information may be acquired (S109).

Now, as shown in FIG. 8B, the ball B rolls further, and the photographing control unit 101 photographs the nth photographed image Cn (n is a natural number, in this case n> 3) (FIG. 2: S101). The detection control unit 102 detects the ball image BP with the nth captured image Cn (n is a natural number) using the artificial intelligence 300 (FIG. 2: S102), and delivers the identification information (“a001”) (FIG. 2: S111). In FIG. 8B, since the ball image BP of the nth captured image Cn does not have the learning essential image 40, the learning scratch defect image 50, and the learning stain defect image 51 trained by the artificial intelligence 300, it is an essential image. The integration control unit 103 does not integrate the required image recognition number (FIG. 2: S105NO), the defective image integration control unit 104 does not integrate the defective image recognition number (FIG. 2: 107NO), and the color information calculation control unit 105 does not integrate. , The color information "0.82" is acquired (FIG. 2: S109).

Here, the repeat control unit 106 determines whether or not the ball image BP of the nth captured image Cn has passed the recognition line L (FIG. 2: S110), but at the time of FIG. 8B, the ball image BP is Since it is located below the inclined surface 10a and has passed the recognition line L, the repetition control unit 106 determines that the ball image BP of the nth captured image Cn has passed the recognition line L (FIG. 2: S110YES).

Then, the repetition control unit 106 completes the repetition, and the discrimination control unit 107 of the terminal device 11 determines the ball B of the ball image BP based on the number of required image recognitions, the number of defective image recognitions, and the color information. It is determined whether it is a non-defective ball, a defective ball, or a dissimilar ball (FIG. 2: S112).

Here, the discrimination method of the discrimination control unit 107 is not particularly limited, but for example, the number of required image recognition is "1" or more, the number of defective image recognition is less than the first threshold value (for example, "1"), and , The color information is within the color range (for example, "0.40" ≤ color information ≤ "1") using the second threshold value ("1") indicating that the surface color of the ball B is a normal color. In the case of the discrimination condition of, the discrimination control unit 107 discriminates the ball B of the ball image BP as a non-defective ball.

Specifically, when the repetition is completed, the discrimination control unit 107 acquires the number of essential image recognitions in the ball image BP, the number of defective images recognized (the number of scratched defective images recognized, the number of dirty defective images recognized), and the color information. do.

Here, as shown in FIG. 8B, the required image recognition number is "3", the scratch-defective image recognition number of the defective image recognition number is "0", and the dirt-defective image recognition number of the defective image recognition number is ". It is "0" and the color information is "0.82". Since the color information is calculated each time, for example, the discrimination control unit 107 may acquire the average value of the color information calculated so far, or may acquire the minimum value of the color information. , You may get the maximum value. Further, the setting of the color range is not particularly limited, but for example, if a second threshold value (“1”) indicating that the surface color of the ball B is a normal color is determined, this second threshold value is set to the maximum value and predetermined. The subtraction value obtained by subtracting the width value of (for example, "0.60") may be set as the minimum value to set the color range composed of the minimum value and the maximum value, or the second threshold value may be used as the center value. A color range composed of the subtracted value and the added value may be set by calculating an added value obtained by adding a predetermined width value and a subtracted value obtained by subtracting the width value from the center value.

Here, in the discrimination control unit 107, the ball image BP has an essential image recognition number of "1" or more, a defective image recognition number of "0" or less of the first threshold value, and color information "0" or less. It is determined that "0.82" corresponds to the first determination condition within the color range ("0.40" ≤ color information ≤ "1") (FIG. 2: S109). This first determination condition is a determination condition corresponding to whether the ball B is new or relatively new. Then, the non-defective product control unit 107 discriminates the ball B of the ball image BP as a non-defective product ball in accordance with the first discrimination condition (FIG. 2: S112). This makes it possible to discriminate the ball B as a non-defective ball with high accuracy.

Further, the number of required image recognition is "1" or more, the number of defective image recognition is equal to or more than the first threshold value ("1"), and the color information is in the color range ("0.40" ≤ color information ≤ "1". ), The discrimination control unit 107 discriminates the ball B of the ball image BP as a defective ball.

Specifically, as shown in FIG. 9A, when the logo "Logo" is present on the ball B as a result of the rotation of the ball B and scratches and stains are present on the ball B, the required image recognition number is "1". The number of scratches and defective images recognized in the number of defective images recognized becomes "1", the number of dirty defective images recognized in the number of defective image recognition becomes "1", and the color information becomes "0.82". In this case, the discrimination control unit 107 determines that the ball image BP of the ball B has an essential image recognition number of "1" or more, a defective image recognition number of "1" or more of the first threshold value, and color information. It is determined that "0.82" corresponds to the second determination condition within the color range ("0.40" ≤ color information ≤ "1") (FIG. 2: S109). This second discrimination condition is a discrimination condition corresponding to a ball in which the ball B is relatively used and has scratches or stains. Then, the discrimination control unit 107 discriminates the ball B of the ball image BP as a defective ball in accordance with the second discrimination condition (FIG. 2: S112). This makes it possible to discriminate the ball B as a defective ball with high accuracy.

Further, when the number of required image recognitions is less than "1" or the color information is outside the color range ("0.40" ≤ color information ≤ "1"), the discrimination control unit 107 sets the discrimination control unit 107. The ball B of the ball image BP is discriminated as a heterogeneous ball.

Specifically, as shown in FIG. 9B, when the logo "Ball" is present on the ball B as a result of the rotation of the ball B and the ball B is scratched and dirty, the required image recognition number is "0". The number of scratches and defective images recognized in the number of defective images recognized becomes "1", the number of dirty defective images recognized in the number of defective image recognition becomes "1", and the color information becomes "0.82". In this case, the discrimination control unit 107 determines that the ball image BP of the ball B corresponds to the third discrimination condition in which the required number of image recognitions is “0” (FIG. 2: S109). This third discrimination condition is a discrimination condition corresponding to a ball in which the ball B does not have a desired essential image. Then, the discrimination control unit 107 discriminates the ball B of the ball image BP as a different kind of ball according to the third discrimination condition (FIG. 2: S112).

Here, in the third determination condition, when the required image recognition number is "0", the ball image BP is desired regardless of the scratch defective image recognition number, the stain defective image recognition number, and the color information of the defective image recognition number. Since there is no required image, the ball B in the ball image BP is discriminated as a heterogeneous ball. As a result, the ball B without the desired essential image can be reliably identified as a different kind of ball.

Further, as shown in FIG. 10A, when the logo "Logo" is present on the ball B and the surface of the ball B is severely soiled, the required image recognition number is "1", and the scratch-defective image of the defective image recognition number is obtained. The recognition number is "0", the dirty image recognition number of the defective image recognition number is "0", and the color information is "0.28". In this case, the discrimination control unit 107 determines that the ball image BP of the ball B has "0.28" of the color information outside the color range ("0.40" ≤ color information ≤ "1"). It is determined that the condition is satisfied (FIG. 2: S109). This third discrimination condition is a discrimination condition in which the surface of the ball B is severely soiled. Then, the discrimination control unit 107 discriminates the ball B of the ball image BP as a different kind of ball according to the third discrimination condition (FIG. 2: S112). As a result, the heavily soiled ball B can be discriminated as a heterogeneous ball even if there is a desired essential image.

Further, as shown in FIG. 10B, if the ball B has a scratch with the logo "Ball" and the surface of the ball B is severely soiled, the required image recognition number becomes "0" and the number of defective image recognition scratches. The number of defective image recognition is "1", the number of defective image recognition is "0", and the color information is "0.28". In this case, the discrimination control unit 107 determines that the ball image BP of the ball B has "0.28" of the color information outside the color range ("0.40" ≤ color information ≤ "1"). It is determined that the condition is satisfied (FIG. 2: S109), and the ball B of the ball image BP is determined as a different kind of ball (FIG. 2: S112) in accordance with the third determination condition.

As described above, under the third determination condition, even if the required number of image recognitions is "1", if the color information is out of the color range ("0.40" ≤ color information ≤ "1"), the ball image Ball B of BP is discriminated as a different kind of ball. As a result, when the surface of the ball B is dirty, it can be reliably discriminated as a different kind of ball regardless of the presence or absence of a desired essential image. Even when the surface of the ball B has a significantly different color, it can be discriminated as a different kind of ball.

In the above description, the ball of the ball image BP is set by combining the number of required image recognition, the number of defective image recognition, and the color information to set three discrimination conditions of a good product ball, a defective product ball, and a different product product ball. The type of B is configured to be discriminated from a non-defective ball, a defective ball, and a dissimilar ball, but these are not particularly limited. A wide variety of ball B types can be discriminated by setting a threshold value for the number of essential image recognitions, a threshold value for the number of defective image recognitions, and a threshold value for color information.

For example, the first determination condition is that the required number of image recognition is "1" or more, the number of defective image recognition is less than "5", and the color information is in a color range different from the above (for example, "0.60"". ≤ Color information ≤ "1") is set in the discrimination condition, and in this discrimination condition, the ball is discriminated as a non-defective product. The second discrimination condition is that the required number of image recognitions is "1" or more, the number of defective image recognitions is "5" or more, and the color information is in a color range different from the above ("0.60" ≤ color information ≤". It is set in the discrimination condition in "1"), and in this discrimination condition, it is discriminated as a defective ball. The third discrimination condition is set to a discrimination condition in which the required number of image recognitions is less than "1", and in this discrimination condition, the ball is discriminated as a different kind of ball. With this setting, the recognition range of good balls will be wide, the recognition range of defective balls will be narrow, and the recognition range of dissimilar balls will be wide.

In addition, the first discriminant condition is set to a discriminant condition in which the required number of image recognitions is "1" or more, and under this discriminant condition, the ball is discriminated as a good product. The second discrimination condition is that the number of defective images recognized is "10" or more, and the color information is within a color range further different from the above (for example, "0.20" ≤ color information ≤ "1"). It is set as a condition, and under this determination condition, it is determined as a defective ball. The third discrimination condition is set to a discrimination condition in which the required number of image recognitions is less than "1" and the color information is outside the color range ("0.20" ≤ color information ≤ "1"). Under the conditions, it is discriminated as a heterogeneous ball. With this setting, the recognition range of good balls will be wider, the recognition range of defective balls will be narrower, and the recognition range of dissimilar balls will be narrower.

In this way, by changing the threshold value for the number of required image recognition, the threshold value for the number of defective image recognition, and the threshold value for the color information in various ways, it is possible to finely discriminate the type of the ball B. be. In particular, in golf driving ranges and golf courses, there are a wide variety of essential image RPs and defective image NPs, and there is also a wide variety of recognition of good, defective, and different types of products on site. It will be possible to respond.

By the way, in the above-mentioned, as the learning defective image, the learning scratch defective image 50 and the learning dirty defective image 51 are adopted, but either one may be adopted or another defective image may be used. For example, a yellowing defective image in which the ball is yellowed may be adopted, and the number of times the yellowing defective image exists may be configured to integrate the number of yellowing defective image recognitions. By discriminating the ball B as either a non-defective ball, a defective ball, or a different kind of ball by using the number of yellowing defective image recognitions, it is possible to select the ball B from different viewpoints.

When the discrimination control unit 107 completes the discrimination of the balls B, the discrimination control unit 107 can sort the balls B by using the sorting device 12. For example, when the discrimination control unit 107 discriminates the ball B as a defective ball or a dissimilar product ball, as shown in FIG. 1, the sorting device 12 is instructed to blow out air, and the sorting device 12 instructs the sorting device 12 to move the ball B. Blow out air in the direction perpendicular to. Then, air hits the ball B rolling downward, and the ball B moves in another moving direction. As a result, the defective ball or the dissimilar product ball can move in another moving direction without rolling in the normal moving direction, and the defective product ball and the dissimilar product ball can be removed.

On the other hand, when the discrimination control unit 107 determines that the ball B is a non-defective ball, no particular processing is performed and nothing is done. As a result, the ball B that rolls downward rolls in the normal movement direction as it is, so that it can be moved to the normal discharge destination.

In the above description, since the sorting device 12 changes the moving direction of the ball B by blowing air, the ball B is configured to sort the ball B into a non-defective ball, a defective ball, or a dissimilar ball. B may be configured to be sorted into either a non-defective ball, a defective ball, or a dissimilar ball.

For example, with respect to the sorting device 12, as shown in FIGS. 11A and 11B, a rotatable rotary plate 1101 is provided on a plate 1100 having an inclined surface 10a, and a circumference of the rotary plate 1101 is provided at the outer peripheral end of the rotary plate 1101. Ball grooves 1102 into which one ball B can enter are provided at predetermined intervals along the direction. The rotation axis of the rotary plate 1101 is provided in a direction perpendicular to the plane direction of the plate 1100. A first guide plate 1103 capable of guiding the ball B to the ball groove 1102 at the uppermost end of the rotating plate 1101 is provided on the left side above the inclined surface 10a, and the ball B is provided on the right side above the inclined surface 10a. A second guide plate 1104 that can be guided to the ball groove 1102 at the uppermost end of the rotary plate 1101 and that covers a predetermined area at the outer peripheral end on the right side of the rotary plate 1101 and extends downward is provided. By rotating the rotary plate 1101 clockwise from the left side to the right side, the ball B that has entered the ball groove 1102 at the uppermost end is on the inclined surface 10a, and the side surface of the ball B is the ball groove 1102 and the first. It is surrounded by the second guide plate 1104. Then, the ball B rolls and moves to the right side below the rotating plate 1101.

Here, the camera 10 is installed directly above a part of the predetermined area on the right side of the outer peripheral end of the rotating plate 1101, and a part of the predetermined area on the right side of the outer peripheral end of the rotating plate 1101 is photographed as a photographing area A. .. The camera 10 photographs a state in which the rotating plate 1101 rotates and the ball B surrounded by the ball groove 1102 rolls downward from above the inclined surface 10a. As a result, by moving the ball groove 1102 at a predetermined moving speed, the ball B that has entered the ball groove 1102 can be rolled at a predetermined moving speed.

The method of moving the ball B is not particularly limited, and for example, an endless rotating belt in which a plurality of protrusions are provided at intervals so that one ball B can enter may be adopted. Other movement methods may be adopted.

Further, the balls extend from a part below the outer peripheral end of the rotary plate 1101 to the lower part of the plate 1100, and are spaced apart from the lower portion of the second guide plate 1104 in the left-right direction of the plate 1100. A first discharge plate 1105 capable of guiding B downward is provided. Further, below the outer peripheral end of the rotary plate 1101, a first discharge portion between the second guide plate 1104 and the first discharge plate 1105 is provided. The first opening / closing shutter 1106 is provided at the entrance of the first discharging portion so as to be openable / closable along the lower side of the outer peripheral end of the rotating plate 1101. Further, the second discharge plate 1107 extends from another portion below the outer peripheral end of the rotary plate 1101 to the lower part of the plate 1100. The second discharge plate 1107 is provided at a predetermined distance in the left-right direction of the plate 1100 with respect to the first discharge plate 1105. A second discharge portion between the first discharge plate 1105 and the second discharge plate 1107 is provided below the outer peripheral end of the rotary plate 1101. The second opening / closing shutter 1108 is provided at the entrance of the second discharging portion so as to be openable / closable along the lower portion of the outer peripheral end of the rotating plate 1101. Further, a third discharge portion is provided on the left side below the rotary plate 1101 from the second discharge plate 1107.

Here, when the sorting device 12 discharges the ball B to the first discharging portion, the first opening / closing shutter 1106 may be opened, and when the ball B is discharged to the second discharging portion, the first opening / closing shutter 1106 may be opened. The opening / closing shutter 1106 may be closed and the second opening / closing shutter 1108 may be opened. When the ball B is discharged to the third discharging portion, the first opening / closing shutter 1106 and the second opening / closing shutter 1108 are closed. Just do it. As a result, the balls B can be sorted and moved to the three discharge portions. The sorting device 12 is not particularly limited and can be configured in a wide variety of ways.

Hereinafter, the effects of the present invention will be specifically described with reference to Examples and the like, but the present invention is not limited thereto.

First, a ball appearance recognition device 1 was prototyped based on FIGS. 1 to 2, and the ball appearance recognition device 1 was used as an example, and the appearance recognition of a golf ball was carried out using the ball appearance recognition device 1. The discriminant condition was based on the first discriminant condition to the third discriminant condition of FIGS. 8-10. As shown in FIG. 12A, when a predetermined golf ball B rolls from the right side to the left side in the photographing area A of the camera 10 of the ball appearance recognition device 1, the ball image BP of the golf ball corresponds to the logo. Since the required image RP appears, the number of required image recognitions is accumulated by "1" by the artificial intelligence 300. Further, the color information of the ball image BP is acquired as "0.42 ...". Further, when the golf ball rolls, a stain defect image NP corresponding to the stain appears on the ball image BP, and the artificial intelligence 300 integrates the number of stain defect image recognitions by "1". Further, when the essential image RP corresponding to the identification number appears in the ball image BP due to the rotation of the golf ball, the artificial intelligence 300 integrates the required image recognition number by "1".

Then, when the golf ball makes one rotation and the essential image RP corresponding to the logo appears in the ball image BP as shown in FIG. 12B, the artificial intelligence 300 integrates the required image recognition number by "1". Further, when the essential image RP corresponding to the logo and the essential image RP corresponding to the identification number appear in the ball image BP, the artificial intelligence 300 integrates the required image recognition number by "2". Then, when the ball image BP passes through the recognition line L, the required image recognition number becomes "1" or more, the stain defect image recognition number becomes "1", and the color information becomes "0.42 ...". In the ball image BP of this golf ball, the required image recognition number is "1" or more, the defective image recognition number is "1" or more of the first threshold value, and the color information is in the color range ("0.40" ≦). It corresponds to the second determination condition within the color information ≤ "1"). Therefore, this golf ball is determined to be a defective ball. Here, when the golf ball is determined to be a defective ball, a yellow image is displayed in the captured image as shown in FIG. 12B. If the golf ball is determined to be a good ball, a blue image may be displayed in the captured image, and if the golf ball is determined to be a heterogeneous ball, a red image may be displayed in the captured image. .. In this way, in order to alert the user, an image of a color corresponding to the discrimination result may be displayed.

Here, when a predetermined number of golf balls were rolled in the shooting area A and confirmed, a golf ball close to a new product was determined to be a good product, and a golf ball with scratches or stains was determined to be a defective product. Since the golf ball different from the above does not have the required image RP, it was determined to be a heterogeneous ball. When these discrimination accuracy was confirmed, the discrimination accuracy was almost 100%. As described above, in the embodiment, it was confirmed that the ball can be discriminated with high accuracy based on the ball image of the rolling ball.

In the embodiment of the present invention, the ball appearance recognition device 1 is configured to include each control unit, but the program for realizing each control unit is stored in a storage medium and the storage medium is provided. It doesn't matter. In this configuration, the program is read by a device, and the device realizes each part. In that case, the program itself read from the recording medium exhibits the effects of the present invention. Further, it is also possible to provide a method of storing the steps executed by each part in the hard disk.

As described above, the ball appearance recognition device and the ball appearance recognition method according to the present invention are effective as a device and method for recognizing the appearance of a ball in all fields of discriminating and selecting balls, and are effective for ball images of rolling balls. Based on this, it is effective as a ball appearance recognition device and a ball appearance recognition method capable of discriminating various types of balls with high accuracy.

1 Ball appearance recognition device 10 Camera 11 Terminal device 12 Sorting device 101 Imaging control unit 102 Detection control unit 103 Essential image integration control unit 104 Defective image integration control unit 105 Color information calculation control unit 106 Repeat control unit 107 Discrimination control unit

Claims (4)

A shooting control unit that shoots a shot image including a ball rolling on a predetermined surface,
The image taken using artificial intelligence in which a learning ball image corresponding to the ball, a learning essential image showing a genuine product of the ball, and a learning defective image including scratches or stains on the ball are learned in advance. A detection control unit that detects a ball image included in an image, an essential image, and a defective image,
When the essential image exists inside the ball image, an essential image integration control unit that integrates the number of times the essential image exists as an essential image recognition number, and
When the defective image exists inside the ball image, a defective image integration control unit that integrates the number of times the defective image exists as the number of defective image recognition, and a defective image integration control unit.
A color information calculation control unit that calculates color information indicating the surface color of the ball from the ball image,
The captured image is photographed, the ball image is detected, the required image recognition number is integrated, and the defective image recognition number is obtained until the ball image passes through a predetermined recognition line provided in the captured image. A repetitive control unit that repeats the integration of the above and the calculation of the color information,
When the repetition is completed, the ball of the ball image is either a good product ball, a defective product ball, or a dissimilar product ball based on the required image recognition number, the defective image recognition number, and the color information. A discrimination control unit that discriminates between
A ball appearance recognition device. The discrimination control unit has a second threshold value in which the required image recognition number is 1 or more, the defective image recognition number is less than the first threshold value, and the color information indicates that the surface color of the ball is a normal color. In the case of the first discrimination condition within the color range using In addition, in the case of the second determination condition in which the color information is within the color range, the ball of the ball image is determined as a defective ball, and the required number of image recognition is less than 1, or the color information is said. In the case of the third discrimination condition that is out of the color range, the ball in the ball image is discriminated as a different product ball.
The ball appearance recognition device according to claim 1. When the essential image is detected, the essential image integration control unit has the center coordinates of the detected essential image, the vertical and horizontal sizes of the essential image, the center coordinates of the detected ball image, and the ball image. Based on the vertical and horizontal sizes of the ball image, it is determined whether or not the essential image exists inside the ball image.
When the defective image is detected, the defective image integration control unit has the center coordinates of the detected defective image, the vertical and horizontal sizes of the defective image, the center coordinates of the detected ball image, and the ball image. It is determined whether or not the defective image exists inside the ball image based on the vertical and horizontal sizes of the ball image.
The ball appearance recognition device according to claim 1 or 2. A shooting control process that continuously shoots a shot image including a ball rolling on a predetermined surface, and
The image taken using artificial intelligence in which a learning ball image corresponding to the ball, a learning essential image showing a genuine product of the ball, and a learning defective image including scratches or stains on the ball are learned in advance. A detection control process for detecting a ball image included in an image, an essential image, and a defective image,
When the essential image exists inside the ball image, the essential image integration control step of integrating the number of times the essential image exists as the required image recognition number, and
When the defective image exists inside the ball image, a defective image integration control step of integrating the number of times the defective image exists as the number of defective image recognition, and a defective image integration control step.
A color information calculation control step for calculating color information indicating the surface color of the ball from the ball image, and
The captured image is photographed, the ball image is detected, the required image recognition number is integrated, and the defective image recognition number is obtained until the ball image passes through a predetermined recognition line provided in the captured image. Repeated control process that repeats the integration of the above and the calculation of the color information,
When the repetition is completed, the ball of the ball image is either a good product ball, a defective product ball, or a dissimilar product ball based on the required image recognition number, the defective image recognition number, and the color information. Discrimination control process to discriminate between
A ball appearance recognition method for a ball appearance recognition device comprising.

Images