JPH05237213A - Image processing method and device for 'go' game device - Google Patents

Abstract

PURPOSE:To determine whether an intersection zone in a panel surface is unchanged, of whether a white or black stone is placed on with high precision even a game is played at an optional place where brightness of the panel surface may be changed partially or by time in forming 'go' pattern data based on image signal for the panel surface of a 'go' panel taken by an image pick-up device. CONSTITUTION:Image signals of a panel surface 2a detected by a CCD camera 3 are stored in an image memory 14, A specified number of image signals within a sampling zone smaller than a stone around an intersection in the panel surface 2a are read from the image memory 14 to normalize a level and supply it to a neutral network 15. Whether the sampling zone is the panel surface 2a, or a black or white stone, is determined by the neutral network 15, and 'go' pattern data are formed based on addresses of intersections in the image memory 14 and result of determination by the neutral network 15 for the sampling zone.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image processing method and apparatus for a Go device for imaging a board surface of a board to determine the stone position and the black and white of the stone to create game data.

[0002]

2. Description of the Related Art According to Japanese Unexamined Patent Publication No. 62-127081, there is known a device capable of remotely playing a game on a board using a robot so as to give an actual playing atmosphere. The game record data created by judging the position and the black and white of the struck stone based on the image signal of the captured image is transmitted.

[0003]

However, since these judgments are made on the basis of the level discrimination of the image signal level with an allowable range, it is necessary to give special consideration to room lighting. It is a prerequisite. Further, even if the room brightness is set to the standard state, if it is continued for several hours, the brightness of the room is affected by the outdoor brightness that changes naturally, or if there is a human figure, there is a problem that an erroneous determination is made. Furthermore, even if the image signal changes due to scratches, peeling off, or dirt of the reflection state of the board surface or its ruled lines, there is a possibility that it may lead to erroneous judgment. That is, there is a problem in that the method of creating a game record data that simply analyzes a ruled line pattern from an image signal or discriminates black and white is unreliable especially when playing a game in an ordinary room.

In view of such a point, the present invention determines the stone position and the black and white of the stone from the image signal detected by picking up the board surface of the board with the image pickup device to generate the game data. An image processing method and device for a learning device, a game device, or a go device that allows a remote partner to respond,
Whether the intersection area of the board remains the board or is struck by a white stone or a black stone, even if it is played in any place where the brightness of the board may change in whole or in part or over time. The purpose is to make the judgment with high accuracy.

[0005]

SUMMARY OF THE INVENTION The present invention relates to a go device for creating game data by determining a stone position and stone black and white from an image signal detected by imaging the board surface of a board with an image pickup device. In order to achieve the above-mentioned object in an image processing method, according to claim 1, the image signal of the board detected by the image pickup device is stored in an image memory, and a predetermined number within a sampling area smaller than stones around the intersection of the board is stored. The image signal of is read from the image memory and the level is normalized and supplied to the neural network, and the neural network is made to judge whether the sampling area is the board surface, black stone or white stone, and the address of the intersection of the image memory and the vicinity of this address. It is characterized in that the game data is created from the judgment result of the neural network for the sampling area of. In order to improve the judgment accuracy, according to claim 2, before the image signal is supplied to the neural network with its level being normalized, the image of the ground color of the board surface with respect to the image signal level of the ruled line or black stone of the board surface is provided. It is preferable to perform a contrast process for increasing the signal level.

As an apparatus for carrying out such a method,
According to claim 3, an image memory for storing the image signal of the board detected by the image pickup device and a predetermined number of image signals are input, and it is determined whether these image signal areas are the board or black stone or white stone. A predetermined number of image signals within a sampling area smaller than a stone around the intersection of the ruled line on the board in the image memory and the neural network are read out, and level normalization processing is performed at the reference level stage that matches the neural network to create a neural network. Image data control means for supplying the image data is provided, and the judgment result of the neural network for the intersection address of the image memory and the sampling area around this intersection address is used as game record data.

[0007]

The image signal of the board detected by the image pickup device is stored in the image memory. The image signal of the sampling area around the intersection of the board surfaces is read from this memory and supplied to the neural network. The neural network learns various board conditions such as the brightness of the board surface, uneven brightness, variation of stones placed, scratches or stains on ruled lines, and learns them based on a predetermined number of image signals. Judge whether it is the surface of the board, black stone or white stone. At this time, when the image signal level of the ground color of the board is made higher than the image signal level of the ruled line of the board or the black stone, the black and yellow discrimination, that is, the ruled line pattern, and thus the black stone and the white stone are surely judged.

[0008]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 shows a schematic external structure of a device for carrying out an image processing method for a Go device according to an embodiment of the present invention and a structure of a built-in circuit portion. A normal board 2 is arranged on the table 1, a player sits on one end 1a of the board, and the opposite robot 4 and white and black stone bowls 5 and 5a are arranged on the opposite side. There is. Robot 4
By providing the go stone gripping device 4b at the tip of the arm 4a, a stone is hit or raised. Further, a CCD camera 3 for picking up an image of the board surface 2a of the board 2 is attached to the opposite end, and the image processing apparatus 1 is installed in the table 1.
0 and the robot controller 20 are stored. CCD
The camera 3 has an auto iris function, and the aperture amount can be manually adjusted, and the aperture amount can be automatically controlled so as to maintain a predetermined average image signal level in response to the brightness instruction signal. There is.

The image processing apparatus 10 includes an A / D converter 16 for digitizing an image signal from the CCD camera 3 and an R / D converter 16.
A CPU 11 that operates according to a program stored in the OM 12 and that has a work area RAM 13 attached thereto, and an image memory 14 that stores an image signal for one screen of the CCD camera 3.
And a neural network 15 for determining whether the sampling area around each intersection of the board 2a is the board surface, black stone, or white stone.

As a result, the CPU 11 causes the ROM 12,
In cooperation with the RAM 13 and the like, as shown as a functional block diagram in FIG. 2, the image signals of a predetermined number of pixels in each sampling area of the image memory 14 are sequentially read out to obtain the ruled line of the board 2a or the image signal level of black stone. On the other hand, an image signal having a contrast processing unit 11a for increasing the image signal level of the ground color of the board 2a and a level normalization processing unit 11b for allocating the contrast-processed image signal to any one of reference level steps matching the neural network 15. It constitutes the control means 11c. That is, the CPU 11 detects the maximum and minimum image signal levels when capturing an image signal for one screen for the contrast processing, and the background color image signal of the board 2a between these maximum and minimum levels. A level correction formula that relatively raises the intermediate level near the level is created. Then, the intersection 8 or the star intersection, the corner L intersection and the peripheral T intersection shown in FIG. 3 are smaller than the Go stone 9 and are sufficiently larger than the star mark, which corresponds to a pixel in the sampling area 8a, for example, 6 × 6 image signals are sequentially read. Incidentally, the size of the sampling area 8a and the ruled line pattern viewed from the CCD camera 3 change according to the change in the distance between the position of the board 2a and the CCD camera 3.
Further, the read image signal is subjected to contrast processing according to the level correction formula described above. Then, the contrast-processed image signal level is normalized into 100 levels between 0 and 1 to match the input level condition of the neural network 15. The CPU 11 also cooperates with the ROM 12, RAM 13, etc. to configure the input / output control means 15a (FIG. 2) of the neural network 15.

The CPU 11 includes a ROM 12 and a RAM.
The brightness instruction signal generating means 11d and the intersection address analyzing means 11e for controlling the diaphragm amount of the CCD camera 3 are also configured in cooperation with the camera 13 and the like. That is, as shown in FIG. 1, with the board 2, the robot 4, the black and white bowls 5 and 5a, and the CCD camera 3 set on the table 1 installed at the place of use, the board surface 2a is monochrome with a predetermined pattern. The stone is placed, the image is taken by the CCD camera 3, and the aperture amount is gradually opened. Then, after confirming the aperture range that the neural network 15 normally determines, the ROM 12 stores the brightness instruction signal corresponding to the average value of the signal levels of all the images on the board 2a with the aperture set to the intermediate value. I'll do it. At the time of use, the read brightness instruction signal is sent to the CCD camera 3 to perform automatic aperture control as a reference value. Further, when the power is turned on, the image signal of the board 2a is fetched, the ruled line pattern is analyzed, and the addresses of 361 intersections of the image memory 14 are determined.

The neural network 15 has as many units as the input layer corresponding to the number of pixels of 6 × 6, and has as output layers two units for transmitting three kinds of output data of black stone, white stone and board. Input / output of these so that the normal + intersection, star intersection, L intersection, and T intersection pattern recognition on the board 2a and the stones and black and white placed on them are surely identified by backpropagation. An intermediate layer is connected between the layers. That is, the shape of each intersection, the placement position of the stone, the surrounding brightness, the uneven brightness, the shadow that may partially occur, etc. are changed variously, and the stone is placed at the intersection in various patterns or the position thereof. Assuming 1, by slightly varying
Approximately 000 to several thousand or so of the surface condition of the board is simulated and imaged.
For 6 × 6 contrast-processed and level-normalized input signals to, the learning signal is given by any one of the three kinds of teacher signals and learned by back propagation so that a normal output signal can be obtained. ..

The robot controller 20 uses a CPU, and sequentially stores the game record data while updating the game record data based on the intersection address of the image memory 14 and the output data output from the neural network 15 each time. It has a game data memory 20a and a game software memory 20b. Then, based on the intersection address sequentially output from the image memory 14 and the associated output data output from the neural network 15, it is determined that a stone has been hit and the game record data in the game record data memory 20a is updated. At the same time, a responder is determined based on the game software in the memory 20b and a robot control signal is sent to the robot 4.

The operation of the image processing apparatus thus configured is as follows. As shown in FIG. 1, table 1
A board 2, a robot 4 and a black and white board 5, 5a and C
With the CD camera 3 set and the initial settings completed,
When the player wants to play against the robot 4, when the power of the device is turned on, the stored brightness indicating signal is supplied to the CCD camera 3, and the aperture amount is adjusted so that the indicated brightness, that is, the standard average image signal level can be secured. Automatically set. Also, 361 intersection point addresses of the image memory 14 are automatically set.

The image memory 14 determines how to perform contrast processing when successively capturing the image signals of the board 2a, and then sequentially reads out the 6 × 6 image signals of the sampling area 8a at each intersection. , Contrast and level normalization processing is performed, and the result is sequentially supplied to the neural network 15. After the contrast processing, the level difference of the ruled line with respect to the yellow ground color of the board 2a is emphasized by normalizing to the reference level step that matches the neural network 15, and the ruled line pattern is surely recognized. Therefore, the white stone or the black stone that changes the ruled line pattern can be recognized with high accuracy. The robot controller 20 sequentially takes in the addresses of the intersections and the judgment data of the neural network 15 and sequentially updates the game record when it judges that the stone has been struck, and also causes the robot 4 to respond or play the stone. To raise.

In such a state, when the lighting condition of the room, the brightness outside the room change, or a figure appears on the board 2a, the CCD camera 3 automatically adjusts the aperture so as to eliminate the error with respect to the brightness indicating signal. Control. The learning function of the neural network 15 makes accurate judgments for various changes in the board surface. With repeated use, even if the ruled line is damaged or dirty, it can be reliably judged. In addition, with the generalization function of the neural network 15, the board 2a
It is possible to make a judgment even if the brightness of the image changes greatly and is out of the follow range of the aperture of the CCD camera 3 or the surface of the board changes unpredictably.

FIG. 4 shows another embodiment of the present invention.
The address of the image memory and the game data of the neural network are transmitted from the image processing device 30 configured as described above to the player at the remote location from the communication device 31 which transmits and receives signals through the modem, while the game data of the other party is also communicated. It is supplied to the robot controller 32 through the device 31.

In the above-described embodiment, the contrast processing is performed so that the image signal level of the white stone is completely different from the image signal level of the ruled line because the stone pattern and the patterns of various ruled lines are completely different. Not high but
Depending on the case, linear contrast processing may be performed together with the yellow region. The number of pixels in the sampling region is 6 × 6, and it has been tested that highly reliable game data can be created within a practical processing time. Further, it may be considered that the number is set to 7 × 7, but there is an upper limit in terms of processing time and circuit configuration. The image signal CCD camera 3 does not need to have the auto iris function, but the neural network 1
By including 5, the allowable change state of the board surface 2a is somewhat limited and can be put to practical use. The image processing device 10 and the robot control device 20 can also be configured by using a common personal computer.

[0019]

As described above, according to the present invention, by using the learning function and the generalization function of the neural network, the whole and partial brightness of the board can be variously varied and variously changed. It is possible to create game record data with extremely high reliability in a normal room with. In other words, the board surface state can be reliably identified regardless of the position of the lighting equipment in the room, the illuminance, the situation outside the room, and the unpredictable change. Even if the ruled line is damaged or dirty, it can be reliably judged. By performing contrast processing so that the image signal level of the ground color of the board surface is higher than that of ruled lines or black stones, pattern recognition of ruled lines is performed with high accuracy, and white stones and black stones that change the pattern are surely recognized. Done.

[Brief description of drawings]

FIG. 1 is a diagram showing a configuration of an image processing device of a Go device for carrying out the method of the present invention.

FIG. 2 is a functional block diagram for explaining the configuration of the same apparatus.

FIG. 3 is a board plan view for explaining the operation of the device.

FIG. 4 is a block diagram showing a circuit configuration of a Go device according to another embodiment.

[Explanation of symbols]

 2 Go board 2a Board surface 3 CCD camera 4 Robot

Claims (4)

[Claims] 1. An image processing method of a go device for creating a game record data by determining a stone position and stone black and white from an image signal detected by imaging a board surface of a board with an imaging device, The detected image signal of the board is stored in the image memory, a predetermined number of the image signals in a sampling area smaller than a stone around the intersection of the board are read from the image memory, and the level is supplied to the neural network after normalization, The neural network is caused to judge whether the sampling area is the board surface, the black stone or the white stone, and game record data is created from the judgment result of the neural network for the address of the intersection of the image memory and the sampling area around this address. An image processing method for a Go device characterized by the above. 2. A contrast process for increasing the image signal level of the ground color of the board surface with respect to the image signal level of the ruled lines or black stones of the board surface before supplying the image signal to the neural network with the level being normalized. The image processing method for a Go device according to claim 1, wherein 3. An image processing device of a go device that determines the stone position and the black and white of the stone from image signals detected by imaging the board surface of the board with the imaging device, and creates a game record data. An image memory for storing the image signal of the board surface, a neural network for inputting a predetermined number of image signals, and determining whether these image signal areas are the board surface or the black stone or the white stone, and the above-mentioned in the image memory An image signal to be supplied to the neural network by reading out the predetermined number of image signals in a sampling area smaller than a stone in the vicinity of the intersection of the ruled lines on the board and performing a level normalization process at a reference level step that matches the neural network. Control means, and the intersection address of the image memory and the sampler around the intersection address. The image processing apparatus of Go device, characterized by the determination result of the neural network to the area and game record data. 4. The image signal control means, in addition to a level normalization processing unit for performing level normalization processing at a reference level step matching a neural network, prior to the level normalization processing, the ruled lines or black stones on the board surface 4. The image processing device for a Go device according to claim 3, further comprising a contrast processing unit that performs a contrast process for increasing the image signal level of the ground color of the board with respect to the image signal level.