KR20200094837A - Apparatus for controlling risk situation of robot work using artificial neural network, method thereof and computer recordable medium storing program to perform the method - Google Patents

Abstract

The present invention relates to a device for controlling the dangerous situation of robot work using an artificial neural network, a method therefor, and a computer-readable recording medium in which a program performing the method is recorded. The device for controlling the dangerous situation of robot work using an artificial neural network of the present invention includes: a camera unit that generates an original mixed image including an object area representing an object and a background area other than the object area by photographing a robot device and the object including and people or materials surrounding the robot device; a generation network that removes the background area by performing a plurality of operations to which weights are applied from the original mixed image, and generates a fake object image consisting of only the object area; and a processing unit that determines whether the object area of the fake object image is included within the working radius of the robot device, and controls to stop the work of the robot device if the object area of the fake object image is included within the working radius as a result of the determination.

Description

Apparatus for controlling risk situation of robot work using artificial neural network, method thereof and a method for controlling the risk situation of robot work using artificial neural network and computer recordable medium storing program to perform the method}

The present invention relates to a dangerous situation control technology, and more particularly, an apparatus for controlling a dangerous situation of a robot operation using an artificial neural network, a method therefor, and a computer-readable recording medium in which a program for performing the method is recorded. About.

Cases of personal injury due to safety accidents in the field where automated robots operate can be accessed through the media without difficulty. However, a simple response, such as hitting a fence outside the area where the automated robot operates, is the only way to avoid safety accidents in the field. However, this method cannot protect workers who are less sensitive to safety because they are accustomed to the work environment. That is, there may be a case where an operator who is said to know the robot's motion path inadvertently crosses the fence and enters the work space, causing an accident.

Registered in Korea Patent No. 1785598 on October 10, 2017 (Name: Access detection system for prevention of safety accidents)

An object of the present invention is to provide an apparatus for controlling a dangerous situation of a robot operation using an artificial neural network, a method therefor, and a computer-readable recording medium in which a program performing the method is recorded.

An apparatus for controlling a dangerous situation of a robot operation using an artificial neural network according to a preferred embodiment of the present invention for achieving the above-described object photographs a robot device and an object including workers or materials around the robot device. A camera unit that generates an original mixed image including an object region representing an object and a background region other than the object region, and the object excluding the background region by performing a plurality of operations to which weights are applied from the original mixed image It is determined whether a generation network for generating a dummy object image consisting of only an area, and whether the object area of the dummy object image is included within the working radius of the robot device, and if the determination result is included within the working radius, the It includes a processing unit that controls to stop the work.

The original mixed image and the dummy object image include a pixel value of each pixel and location information of each pixel, and the processing unit generates the dummy object image by inputting the original mixed image to the generation network, and generates the It is characterized in that it is determined whether the object is included within a working radius of the robot device by using position information of the object area of the dummy object image.

The apparatus is configured to set an expected value to determine whether the fake object image is an original or a fake through a plurality of calculations in which a weight is applied to the fake object image, and to determine that the fake object image is a fake. Thereafter, learning to modify the weight of the division network so that the difference between the set expected value and the output value of the division network is minimal, and after setting the expected value to determine that the fake object image is the original, the set expected value and the output value of the division network It further includes a learning unit that repeats so that learning of modifying the weight of the generation network to compete so that the difference of is minimal.

The learning unit is characterized in that during the learning, if there is no change in the weight of the generation network and the division network, the learning unit repeats the learning after setting by increasing the expected value.

A method for controlling a dangerous situation of a robot operation using an artificial neural network according to a preferred embodiment of the present invention for achieving the above object is an object including a robot device and a person or material around the robot device. Generating an original mixed image including an object region representing an object and a background region other than the object region by photographing, and the background region by performing a plurality of operations to which weights are applied from the original mixed image by a generation network And generating a dummy object image consisting of only the object region; determining whether the object region of the dummy object image is included within the working radius of the robot; and the processing unit as a result of the determination And, if included within the working radius, controlling to stop the work of the robot device.

In the method, before the step of generating the original mixed image, a division network for outputting whether the fake object image is the original or the fake object through a plurality of calculations in which a weight is applied to the fake object image by the learning unit is the fake object. After setting an expected value to determine that the image is a fake, learning to modify the weight of the partition network so that the difference between the set expected value and the output value of the partition network is minimal, and the expected value to determine that the fake object image is the original. After setting, the step of repeating the learning of correcting the weight of the generation network so that the difference between the set expected value and the output value of the division network is minimized to compete.

According to another aspect of the present invention, there is provided a computer-readable recording medium in which a program for performing a method for controlling a dangerous situation of a robot operation according to a preferred embodiment of the present invention is recorded.

By using a neural network, it is possible to detect that an operator enters the robot working radius and stop the robot work quickly, thereby ensuring the stability of the industrial site.

1 is a view for explaining a system for controlling a dangerous situation of a robot operation using an artificial neural network according to an embodiment of the present invention.
2 is a block diagram illustrating an apparatus for controlling a dangerous situation of a robot operation using an artificial neural network according to an embodiment of the present invention.
3 is a block diagram illustrating a detailed configuration of a control unit according to an embodiment of the present invention.
4 is a diagram illustrating a detailed configuration of a computing network according to an embodiment of the present invention.
5 is a diagram for explaining a detailed configuration of a generation network according to an embodiment of the present invention.
6 is a diagram for explaining a detailed configuration of a division network according to an embodiment of the present invention.
7 is a flowchart illustrating a method of learning a computing network according to an embodiment of the present invention.
8 is a flowchart illustrating initial learning of a computing network according to an embodiment of the present invention.
9 is a flow chart for explaining competitive learning of an artificial neural network according to an embodiment of the present invention.
10 is a flowchart illustrating a method for controlling a dangerous situation of a robot operation using an artificial neural network according to an embodiment of the present invention.

Prior to the detailed description of the present invention, terms or words used in the present specification and claims described below should not be construed as being limited to their usual or dictionary meanings, and the inventors will use their own invention in the best way. For explanation, based on the principle that it can be appropriately defined as a concept of terms, it should be interpreted as a meaning and concept consistent with the technical idea of the present invention. Therefore, the embodiments described in the present specification and the configurations shown in the drawings are only the most preferred embodiments of the present invention, and do not represent all the technical ideas of the present invention, and thus various equivalents that can replace them at the time of application It should be understood that there may be water and variations.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. In this case, it should be noted that the same components in the accompanying drawings are indicated by the same reference numerals as possible. In addition, detailed descriptions of well-known functions and configurations that may obscure the subject matter of the present invention will be omitted. For the same reason, in the accompanying drawings, some components are exaggerated, omitted, or schematically illustrated, and the size of each component does not entirely reflect the actual size.

First, a system for controlling a dangerous situation of a robot operation using an artificial neural network according to an embodiment of the present invention will be described. 1 is a view for explaining a system for controlling a dangerous situation of a robot operation using an artificial neural network according to an embodiment of the present invention.

Referring to FIG. 1, a system for controlling a dangerous situation of a robot operation using an artificial neural network (hereinafter, abbreviated as'control system') according to an embodiment of the present invention includes a control device 100 and a robot device 400. Includes.

The robot device 400 is an industrial robot, which has a memory device and is a general-purpose machine capable of replacing human tasks by automatically performing operations such as stretching, turning, and vertical movement of an arm with a tip that can hold an object. Say. The robot device 400 may be controlled by the control device 100.

The control device 100 photographs the surroundings of the robot device 400 through a plurality of camera units that perform a camera function, analyzes the captured image, When an object such as, is detected, the robot device 400 is immediately controlled so that the operation of the robot device 400 is stopped.

Next, a description will be given of a control device 100 for controlling a dangerous situation of a robot operation using an artificial neural network according to an embodiment of the present invention. 2 is a block diagram illustrating an apparatus for controlling a dangerous situation of a robot operation using an artificial neural network according to an embodiment of the present invention.

Referring to FIG. 2, the control device 100 according to an embodiment of the present invention includes a communication unit 110, a camera unit 120, an input unit 130, a display unit 140, an audio unit 150, and a storage unit 160. ) And a control unit 200.

The communication unit 110 is, for example, a means for communicating with the robot device 400. The communication unit 110 may include a radio frequency (RF) transmitter Tx for up-converting and amplifying a frequency of a transmitted signal, and an RF receiver Rx for low-noise amplifying and down-converting a received signal. In addition, the communication unit 110 may include a modem that modulates the transmitted signal and demodulates the received signal.

The camera unit 120 is for capturing an image and includes an image sensor. The image sensor receives light reflected from a subject and converts it into an electric signal, and may be implemented based on a Charged Coupled Device (CCD) or Complementary Metal-Oxide Semiconductor (CMOS). The camera unit 120 may further include an analog to digital converter, and may convert an electrical signal output from the image sensor into a digital sequence and output it to the controller 200.

In particular, the camera unit 120 includes a 3D sensor. The 3D sensor is a sensor for obtaining 3D coordinates of each pixel of an image in a non-contact method. When the camera unit 120 captures an image, the 3D sensor measures the actual distance from a predetermined reference point (eg, focus) of the camera unit 120 to the captured object, together with the captured image. Generate the three-dimensional coordinates of the pixel. As the 3D sensor, various types of sensors using laser, infrared, visible light, etc. can be used. These 3D sensors are laser type 3D scanners using any one of TOP (Time of Flight), phase-shift and Online Waveform Analysis, laser type 3D scanners using optical triangulation, and optics using white light or modulated light. Type 3D scanner, Handheld Real Time PHOTO, optical 3D scanner, optical method using Pattern Projection or Line Scanning, laser system full body scanner, photo scanner using photogrammetry, Kinect Fusion A real time scanner to be used may be exemplified.

The input unit 130 may receive a user's key manipulation for controlling the control device 100, generate an input signal, and transmit the input signal to the control unit 200. The input unit 130 may include various types of keys for controlling the control device 100. When the display unit 140 is formed of a touch screen, the input unit 130 may perform functions of various keys on the display unit 140, and when all functions can be performed only with the touch screen, the input unit 130 will be omitted. May be.

The display unit 140 is for screen display, and may visually provide a menu of the control device 100, input data, function setting information, and various other information to a user. In addition, the display unit 140 performs a function of outputting screens such as a boot screen, a standby screen, a menu screen, and the like of the control device 100. The display unit 140 may be formed of a liquid crystal display (LCD), an organic light emitting diode (OLED), an active matrix organic light emitting diode (AMOLED), or the like. Meanwhile, the display unit 140 may be implemented as a touch screen. In this case, the display unit 140 includes a touch sensor. The touch sensor detects a user's touch input. The touch sensor may be composed of a touch sensing sensor such as a capacitive overlay, a pressure type, a resistive overlay, or an infrared beam, or may be composed of a pressure sensor. . In addition to the above sensors, all types of sensor devices capable of sensing contact or pressure of an object may be used as the touch sensor of the present invention. The touch sensor detects a user's touch input, generates a detection signal, and transmits it to the controller 200. In particular, when the display unit 140 is formed of a touch screen, some or all of the functions of the input unit 130 may be performed through the display unit 140.

The audio unit 150 includes a speaker (SPK) for outputting an audio signal such as voice according to an embodiment of the present invention, and a microphone (MIKE) for collecting an audio signal such as voice. That is, the audio unit 150 may output an audio signal through the speaker SPK or transmit an audio signal input through the microphone (MIKE) to the control unit 200 under the control of the controller 200.

The storage unit 160 serves to store programs and data necessary for the operation of the control device 100. Each type of data stored in the storage unit 160 may be deleted, changed, or added according to a user's manipulation.

The controller 200 may control the overall operation of the control device 100 and a signal flow between internal blocks of the control device 100 and perform a data processing function of processing data. In addition, the control unit 200 basically performs a role of controlling various functions of the control device 100. The control unit 200 may be a central processing unit (CPU), a digital signal processor (DSP), or the like. Next, the configuration and operation of the control unit 200 will be described in more detail below.

First, the configuration of the control unit 200 according to an embodiment of the present invention will be described. 3 is a block diagram illustrating a detailed configuration of a control unit according to an embodiment of the present invention. Referring to FIG. 3, a control unit 200 according to an embodiment of the present invention includes a learning unit 210, a processing unit 220, and a computing network 300.

The computing network 300 is formed of a plurality of artificial neural networks, and the artificial neural network includes a generation network 310 and a division network 320. The learning unit 210 is for learning the computing network 300 through learning data.

The learning unit 210 is for learning the computing network 300 through learning data. That is, the learning unit 210 extracts only the object area excluding the background area from the original mixed image 10 including an object area representing an object including a worker or material and a background area other than the object area. The computing network 300 is trained to generate (30).

The processing unit 220 generates a simulated object image 30 through the computation network 300 from the original mixed image 10 photographed through the camera unit 120 in the computation network 300 on which the learning has been completed, and It is determined whether the object of the dummy object image 30 is included within the working radius of the robot device 400 through the location information of the object area of (30).

Then, a detailed configuration of the computing network 300 according to an embodiment of the present invention will be described. 4 is a diagram illustrating a detailed configuration of a computing network according to an embodiment of the present invention. Referring to FIG. 4, the computing network 300 includes a generating network 310 and a division network 320. Each of the generation network 310 and the division network 320 constitutes an artificial neural network (ANN). Each of the generation network 310 and the division network 320 will be described.

First, the generation network 310 will be described. 5 is a diagram for explaining a detailed configuration of a generation network according to an embodiment of the present invention. Referring to FIG. 5, the generation network 310 includes a plurality of layers including a plurality of operations to which weights are applied. Here, the plurality of layers including a plurality of operations include a convolution layer (CL) that performs a convolution operation, a pooling layer (PL) that performs a down-sampling operation, and It includes at least one each of an unpooling layer (UL) layer that performs an up-sampling operation and a deconvolution layer (DL) that performs a deconvolution operation. Each of the convolution, downsampling, upsampling, and deconvolution operations uses a filter composed of a predetermined matrix, and the values of the elements of the matrix become weights.

mm 떨어진 점이된다. The generation network 310 receives an original mixed image 10 including an object region representing an object and a background region other than the object region. In particular, the original mixed image 10 includes pixel values of each pixel and location information of each pixel. Here, the location information is a three-dimensional coordinate value (x, yz) having a predetermined reference point and a predetermined unit (eg, cm, mm, etc.). For example, if one pixel P1 of the original mixed image (10) has a three-dimensional coordinate value (a, b, c) and the unit is mm, this pixel P1 is from the reference point (0, 0, 0).

becomes a point away from mm.

When the original mixed image 10 is input, the generation network 310 generates a dummy object image 30 consisting of only the object region excluding the background region by performing a plurality of operations to which the weights of the plurality of layers described above are applied. . The generated dummy object image 30 includes a pixel value of each pixel and location information of each pixel. Here, the location information is a 3D coordinate value (x, y z) having the same reference point as the original mixed image 10 and the same unit (eg, cm, mm, etc.) as the original mixed image 10.

Accordingly, according to the location information of each pixel in the object area of the dummy object image 30 generated by the generation network 310, the location of the actual object, worker, or material in the three-dimensional space can be known.

Next, the division network 320 will be described. 6 is a diagram for explaining a detailed configuration of a division network according to an embodiment of the present invention.

As shown in FIG. 6, the division network 320 includes a plurality of layers including a plurality of operations to which weights are applied. Here, the plurality of layers including a plurality of operations are a convolution layer (CL) performing a convolution operation and a fully connected layer (FL) performing a soft-max operation. Layer). The convolution operation uses a filter composed of a predetermined matrix, and the values of the elements of the matrix become weights. In addition, the softmax operation is also performed by applying weights.

The division network 320 receives object images 30 and 40 composed of only object regions excluding the background region, and performs a plurality of operations to which weights are applied to the object images 30 and 40, and the input object image 30 , 40) is output whether it is the original (real) or the fake (fake). Here, the object images 30 and 40 include the dummy object image 30 and the original object image 40. As described above, the original mixed image 10 is an image including an object region representing an object and a background region other than the object region. The simulated object image 30 is an image generated by the generation network 310 in a simulated manner from the original mixed image 10. The original object image 40 is an image produced for learning, and is an image obtained by extracting only the object region from the original mixed image 10 through graphic work.

Next, a method of learning the computing network 300 according to an embodiment of the present invention will be described. 7 is a flowchart illustrating a method of learning a computing network according to an embodiment of the present invention.

Referring to FIG. 7, the learning unit 210 individually performs initial learning for each of the generation network 310 and the division network 320 of the computing network 300 in step S100. During initial learning, the learning unit 210 trains the generation network 310 to receive the original mixed image 10 and output the dummy object image 30. In addition, the learning unit 210 learns to determine that the identification network 320 is a fake object image 30 as a fake, and determines the original object image 40 as a real one during initial learning. Let it.

Then, the learning unit 210 performs competitive learning on the generation network 310 and the division network 320 in step S200. At this time, the learning unit 210 applies different expected values to perform learning so that the division network 320 and the generation network 310 compete with each other.

That is, the learning unit 210 trains the division network 320 so that the division network 320 determines that the dummy object image 30 generated by the generation network 310 is a fake. Correspondingly, the learning unit 210 trains the generation network 310 so that the division network 320 determines that the dummy object image 30 generated by the generation network 310 is real.

Then, the initial learning of the computing network 300 according to an embodiment of the present invention will be described in more detail. 8 is a flowchart illustrating initial learning of a computing network according to an embodiment of the present invention.

5 and 8, the learning unit 210 performs initial learning on the generation network 310 using the training data in step S110. During initial learning in step S110, the learning unit 210 uses the original mixed image 10 and the original object image 40 as the training data. The learning unit 210 inputs the original mixed image 10 as training data into the generation network 310. Then, the generation network 310 outputs the dummy object image 30 by performing a plurality of operations to which weights are applied to the input original mixed image 10. Then, the learning unit 210 compares the dummy object image 30 and the original object image 40 to each other, so that the dummy object image 30 is back-diffused so that the difference with the original object image 40 is minimized. propagation) algorithm to modify the weight of the generation network 310. This step S110 is repeatedly performed using a plurality of different training data, that is, the original mixed image 10 and the original object image 40 pair.

Next, referring to FIGS. 6 and 8, the learning unit 210 performs initial learning on the division network 320 in step S120. The learning unit 210 individually uses each of the dummy object image 30 and the original object image 40 as training data during initial learning of the division network 320. In this step S120, the learning unit 210 uses the learning data to determine that the division network 320 determines the fake object image 30 as a fake, and determines the original object image 40 as a real. Let's determine. In this step S120, the learning unit 210 uses the dummy object image 30 or the original object image 40 as the learning data in the division network 320. First, in response to the learning data, the learning unit 210 sets an expected value so that the classification network 320 determines the imitation object image 30 as being a imitation, and determines the original object image 40 as the original. That is, the learning unit 210 may set the expected values for the dummy object image 30 as'real = 0.40' and'fake = 0.60'. In addition, the learning unit 210 may set the expected values for the original object image 40 as'real = 0.60' and'fake = 0.40'.

Then, the learning unit 210 inputs the simulated object image 30 or the original object image 40 as training data to the division network 320. When the training data is input, the division network 320 outputs a probability that the training data input through a plurality of operations is a real and a fake probability. In other words, when the dummy object image 30 or the original object image 40 is input, the division network 320 includes the dummy object image 30 or the original object image 40 input through an operation to which a plurality of weights are applied. The probability of being real and the probability of being fake are output as output values. For example, the division network 320 may output a probability of a real and a probability of a fake as output values such as'real = 0.70' and'fake = 0.30'. Then, the learning unit 210 corrects the weight of the division network 320 by using a despreading algorithm so that the difference between the output value of the division network 320 and the expected value is minimized.

Competitive learning of the computing network 300 performed after the initial learning as described above is completed will be described. 9 is a flow chart for explaining competitive learning of an artificial neural network according to an embodiment of the present invention. To emphasize, FIG. 9 is an embodiment of step S200 of FIG. 7.

Referring to FIG. 9, the learning unit 210 sets an expected value so that the division network 320 determines that the dummy object image 30 generated by the generation network 310 is a fake in step S210. Learn (320). To this end, the learning unit 210 inputs the original mixed image 10 into the generation network 310 and obtains the dummy object image 30 from the generation network 310.

At this time, the learning unit 210 sets the expected values for the dummy object image 30 as'real = 0.40' and'fake = 0.60'. Then, the learning unit 210 inputs the dummy object image 30 acquired from the generation network 310 into the classification network 320. Accordingly, the division network 320 will output an output value through an operation to which a plurality of weights are applied. Then, the learning unit 210 uses a despreading algorithm to minimize the loss value, which is the difference between the output value of the division network 320 and the expected value, and the weight of the generation network 310 is fixed. Modify the weights.

In response to the learning of the segmentation network 320 in step S210, the learning unit 210 determines that the imitation object image 30 generated by the generation network 310 by the segmentation network 320 in step S220 is real. The generation network 310 is trained by setting an expected value to determine. To this end, the learning unit 210 inputs the original mixed image 10 into the generation network 310 to obtain the dummy object image 30 from the generation network 310. In this case, the learning unit 210 sets the expected values for the dummy object image 30 as'real = 0.60' and'fake = 0.40'. Then, the learning unit 210 inputs the dummy object image 30 generated by the generation network 310 into the classification network 320. Accordingly, the division network 320 will output an output value through an operation to which a plurality of weights are applied.

Then, the learning unit 210 uses the despreading algorithm so that the loss value, which is the difference between the output value of the division network 320 and the expected value, is minimized, and the weight of the division network 320 is fixed. Modify the weights. In this way, the competitive learning is a procedure of modifying the weight of the division network 320 so that the division network 320 determines the dummy object image 30 as a counterfeit, and the dummy object image 30 generated by the division network 320. The procedure of modifying the weight of the generating network 310 competes to determine that) is the original.

These steps S210 to S220 may be performed alternately and repeatedly.

If there is no change in the weights of the generation network 310 and the division network 320 in the repeated competitive learning process, the learning unit 210 may increase the expected value to continue the competitive learning. That is, when learning the division network 320, the learning unit 210 sets the expected values for the dummy object image 30 in'real = 0.40' and'fake = 0.60' with'real = 0.30' and'fake = 0.70'. Can be increased together. This means that an expected value for allowing the division network 320 to determine the simulated object image 30 as being real is increased. In response, the learning unit 210 sets the expected values for the dummy object image 30 when learning the generation network 310 in'real = 0.60' and'fake = 0.40', and'real = 0.70' and'fake = 0.30'. It can be increased like'. This means that the generation network 310 increases an expected value that causes the classification network 320 to determine the simulated object image 30 as being real. On the other hand, the learning unit 210 increases the expected value to the maximum (for example, the division network'real = 0.00' and'fake = 1.00', the generation network'real = 1.00' and'fake = 0.00') even after increasing the division network 320 ) And the weights of the generation network 310 do not change, the learning may be completed.

When learning is completed as described above, the control device 100 may determine whether an object enters within the working radius OR of the robot device 400 using the computing network 200. This method will be described. 10 is a flowchart illustrating a method for controlling a dangerous situation of a robot operation using an artificial neural network according to an embodiment of the present invention.

Referring to FIG. 10, the processing unit 220 captures an object including workers or materials around the robot device 400 and the robot device 400 through the camera unit 210 in step S310, and the original mixed image 10 ). Here, the original mixed image 10 includes an object region representing an object and a background region other than the object region. In this case, the original mixed image 10 includes pixel values of each pixel and location information (three-dimensional coordinates). As described above, the location information may be obtained through the 3D sensor of the camera unit 10.

Then, the processing unit 220 generates the dummy object image 30 through the generation network 310 of the computing network 300 in step S320. That is, the processing unit 220 inputs the original mixed image 10 into the generation network 310 of the computing network 300 on which the training has been completed. Then, the generation network 310 of the computing network 300 generates a dummy object image 30 by performing a plurality of operations to which the learned weight is applied.

As described above, each pixel of the original mixed image 10 includes location information as well as a pixel value. Accordingly, each pixel of the dummy object image 30 also includes location information as well as a pixel value. Therefore, the processing unit 220 derives the location information of the object area of the dummy object image 30 in step S330.

Then, the processing unit 220 compares the position information of the object area derived earlier in step S340 with the position information of the working radius of the previously stored robot device 400, so that the object of the dummy object image 30 is converted to the robot device 400. Determines whether it is within the working radius of ).

As a result of the determination, the processing unit 220 generates a control command for stopping the work of the robot device 400 and transmits it to the robot device 400 through the communication unit 110 when it is included within the working radius. Accordingly, the robot device 400 will stop working.

On the other hand, as a result of the determination, the processing unit 220 repeats steps S310 to S340 if it is not included within the working radius.

Meanwhile, various methods according to an embodiment of the present invention described above may be implemented in the form of programs that can be read through various computer means and recorded in a computer-readable recording medium. Here, the recording medium may include a program command, a data file, a data structure, or the like alone or in combination. The program instructions recorded on the recording medium may be specially designed and configured for the present invention, or may be known and usable to those skilled in computer software. For example, the recording medium includes magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic-optical media such as floptical disks ( magneto-optical media), and hardware devices specially configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of the program instructions may include not only machine language wires such as those made by a compiler, but also high-level language wires that can be executed by a computer using an interpreter or the like. Such a hardware device may be configured to operate as one or more software modules to perform the operation of the present invention, and vice versa.

Although the present invention has been described above using some preferred embodiments, these embodiments are illustrative and not limiting. As such, those of ordinary skill in the art to which the present invention pertains will understand that various changes and modifications can be made according to the equivalence theory without departing from the spirit of the present invention and the scope of the rights set forth in the appended claims.

100: control device 110: communication unit
120: camera unit 130: input unit
140: display unit 150: audio unit
160: storage unit 200: control unit
210: learning unit 220: processing unit
300: computing network 310: generating network
320: division network 400: robot device

Claims (7)

In an apparatus for controlling a dangerous situation of a robot operation using an artificial neural network,
A camera unit for generating an original mixed image including an object area representing the object and a background area other than the object area by photographing an object including a robot device and a worker or material around the robot device;
A generation network that performs a plurality of operations to which weights are applied from the original mixed image to generate a dummy object image composed of only the object region excluding the background region; And
And a processing unit that determines whether the object area of the simulated object image is included within the working radius of the robot device, and controls to stop the work of the robot device if it is included within the working radius as a result of the determination. With
A device for controlling dangerous situations in robot work. The method of claim 1,
The original mixed image and the dummy object image include a pixel value of each pixel and location information of each pixel,
The processing unit
Generating the fake object image by inputting the original mixed image into the generation network, and determining whether the object is included within the working radius of the robot device using the location information of the object area of the created fake object image Characterized by
A device for controlling dangerous situations in robot work. The method of claim 1,
A division network that outputs whether the dummy object image is the original or the dummy object image through a plurality of operations in which a weight is applied to the dummy object image;
After setting an expected value to determine that the fake object image is a fake, learning to modify the weight of the partition network so that the difference between the set expected value and the output value of the partition network is minimal, and to determine that the fake object image is the original After setting the expected value, a learning unit that repeats so that learning of modifying the weight of the generation network so that the difference between the set expected value and the output value of the division network is minimized competes; further comprising:
A device for controlling dangerous situations in robot work. The method of claim 3,
Reminder learning
When the weights of the generation network and the division network do not change during the learning, the learning is repeated after setting by increasing the expected value.
A device for controlling dangerous situations in robot work. In a method for controlling a dangerous situation of a robot operation using an artificial neural network,
Generating an original mixed image including an object area representing the object and a background area other than the object area by photographing the robot device and an object including people or materials around the robot device;
Removing the background region by performing a plurality of operations to which weights are applied from the original mixed image, and generating a dummy object image consisting of only the object region; And
Determining, by a processing unit, whether the object area of the fake object image is included within the working radius of the robot device;
And if the processing unit is included within the working radius as a result of the determination, controlling to stop the operation of the robot device.
A method for controlling hazardous situations in robotic work. The method of claim 5,
Before the step of generating the original mixed image,
After the learning unit sets an expected value so that the division network outputting whether the fake object image is the original or the fake object through a plurality of calculations in which weight is applied to the fake object image determines that the fake object image is a fake, Learning of modifying the weight of the division network so that the difference between the set expected value and the output value of the division network is minimum,
After setting an expected value to determine that the dummy object image is the original, learning to modify the weight of the generation network so that the difference between the set expected value and the output value of the division network is minimum
Repetitive step to compete; comprising a
A method for controlling hazardous situations in robotic work. A computer-readable recording medium in which a program for performing a method for controlling a dangerous situation of a robot operation according to claim 5 or 6 is recorded.

Images