KR20200099117A - Method for acquiring movement attributes of moving object and apparatus for performing the same - Google Patents

Abstract

The present invention relates to a method for obtaining movement attributes of a moving object using an image photographed by a camera installed in a ship and an artificial neural network. The method of the present invention comprises: a step of obtaining a plurality of images including a first image photographing the sea and a second image which is a subsequent frame of the first image; an object information obtaining step of obtaining first object information from the first image and obtaining second object information from the second image, using an artificial neural network learned based on output data including a first classification value reflected with object type information, wherein the first classification value corresponds to the sea and a second classification value reflected with object distance and type information, wherein the second classification value corresponds to an obstacle, and labeling data corresponding to the output data; and a step of calculating a moving direction of a moving object included in the photographed sea by comparing the first object information and the second object information.

Description

A method of acquiring movement properties of a moving object, and an apparatus that performs the same.

The present invention relates to a method of obtaining a moving property of a moving object. More specifically, the present invention relates to a method of acquiring a moving property of a moving object included in an image captured using a learned artificial neural network.

It is the era of artificial intelligence (AI). Since AlphaGo became a hot topic, attempts to apply artificial intelligence to various industrial fields have been actively conducted.

In recent years, artificial intelligence has been actively used in image processing technologies such as image recognition, analysis, generation, and synthesis, but recently, it is mounted on automobiles, ships, drones, etc., and is used for peripheral obstacle recognition and route planning.

On the other hand, when artificial intelligence is used in a ship or automobile to obtain information on surrounding obstacles, it is important to acquire movement properties for moving obstacles.

The problem to be solved according to an embodiment is to acquire object information of a surrounding environment using an artificial neural network that performs image segmentation.

A problem to be solved according to another exemplary embodiment is to obtain a moving property of a moving object using object information output from an artificial neural network.

Another problem to be solved according to an embodiment is to generate control signals for ships and vehicles by using the movement properties of a moving object.

The problem to be solved of the present invention is not limited to the above-described problems, and the problems that are not mentioned can be clearly understood by those of ordinary skill in the art from the present specification and the accompanying drawings. will be.

According to an embodiment, a method of acquiring movement properties of a moving object using an image captured by a camera installed on a ship and an artificial neural network, wherein a first image captured at sea and a second image that is a subsequent frame of the first image are Obtaining a plurality of images including; In order to output object information included in the image from the image, a first classification value reflecting the type information of the object-In this case, the first classification value corresponds to the sea-and distance information and type information of the object The reflected second classification value-In this case, the second classification value corresponds to an obstacle-is determined from the first image by using an artificial neural network learned based on the output data and labeling data corresponding to the output data. 1 object information obtaining step of obtaining object information and obtaining second object information from the second image; And calculating a moving direction of the moving object included in the imaged sea by comparing the first object information and the second object information.

According to another embodiment, a method of calculating a moving direction of a moving object using an artificial neural network, wherein a plurality of captured images including a first captured image and a second captured image that is a subsequent frame of the first captured image are obtained. step; A first classification value reflecting the type information of the object to output object information included in the captured image from the captured image-In this case, the first classification value corresponds to the sea-and distance information and type of the object The first image is captured using an artificial neural network learned based on output data including a second classification value reflecting information-In this case, the second classification value corresponds to an obstacle-and labeling data corresponding to the output data. Obtaining first object information from an image, and obtaining second object information from the second captured image; And calculating a moving direction of at least some of the moving objects included in the plurality of captured images by comparing the first object information and the second object information.

According to another embodiment, the camera is installed on the ship to capture the sea; And the object type information to obtain a plurality of images including a first image captured by the camera and a second image that is a subsequent frame of the first image, and to output object information included in the image from the image. -In this case, the first classification value corresponds to the sea-and a second classification value reflecting distance information and type information of the object-In this case, the second classification value corresponds to an obstacle -Obtaining first object information from the first image using an artificial neural network learned based on output data including and labeling data corresponding to the output data, obtaining second object information from the second image, and And a control unit that compares the first object information and the second object information to calculate a moving direction of the moving object included in the captured sea.

The solution means of the subject of the present invention is not limited to the above-described solution means, and solutions not mentioned will be clearly understood by those of ordinary skill in the art from the present specification and the accompanying drawings. I will be able to.

According to an embodiment, object information of a surrounding environment may be obtained using an artificial neural network that performs image segmentation.

According to another embodiment, a moving property of a moving object may be obtained using object information output from an artificial neural network.

According to another embodiment, a control signal for a ship and a vehicle may be generated by using the movement property of the moving object.

The effects of the present invention are not limited to the above-described effects, and effects that are not mentioned will be clearly understood by those of ordinary skill in the art from the present specification and the accompanying drawings.

1 is a block diagram illustrating a method of learning an artificial neural network according to an embodiment.
2 is a block diagram of an inference step of an artificial neural network according to an embodiment.
3 is a diagram illustrating a method of obtaining object information using an artificial neural network according to an embodiment.
4 is a diagram illustrating an image segmentation operation of an artificial neural network according to an exemplary embodiment.
5 is a table showing classification values according to an embodiment.
6 is a diagram for describing data expansion according to an exemplary embodiment.
7 is a diagram for describing a method of obtaining a movement attribute of a moving object according to an exemplary embodiment.
8 is a flowchart illustrating a method of obtaining movement properties of a moving object according to another exemplary embodiment.
9 is a diagram for describing a method of obtaining a movement attribute of a moving object according to another embodiment.
10 is a table showing classification values and movement properties of an object according to an exemplary embodiment.
11 is a diagram for describing a method of obtaining a movement attribute of a moving object according to another exemplary embodiment.
12 is a diagram for describing a method of obtaining movement properties of a moving object according to another exemplary embodiment.
13 is a diagram for describing a method of obtaining movement properties of a moving object according to another exemplary embodiment.

The above objects, features and advantages of the present invention will become more apparent through the following detailed description in conjunction with the accompanying drawings. However, in the present invention, various changes may be made and various embodiments may be provided. Hereinafter, specific embodiments will be illustrated in the drawings and described in detail.

In the drawings, the thicknesses of layers and regions are exaggerated for clarity, and elements or layers are referred to as "on" or "on" of other elements or layers. This includes not only directly above other components or layers, but also when other layers or other components are interposed in the middle. Throughout the specification, the same reference numerals represent the same elements in principle. In addition, components having the same function within the scope of the same idea shown in the drawings of each embodiment will be described using the same reference numerals.

If it is determined that a detailed description of known functions or configurations related to the present invention may unnecessarily obscure the subject matter of the present invention, a detailed description thereof will be omitted. In addition, numbers (eg, first, second, etc.) used in the description of the present specification are merely identification symbols for distinguishing one component from another component.

In addition, the suffixes "module" and "unit" for constituent elements used in the following description are given or used interchangeably in consideration of only the ease of preparation of the specification, and do not themselves have distinct meanings or roles.

According to an embodiment, a method of acquiring movement properties of a moving object using an image captured by a camera installed on a ship and an artificial neural network, wherein a first image captured at sea and a second image that is a subsequent frame of the first image are Obtaining a plurality of images including; In order to output object information included in the image from the image, a first classification value reflecting the type information of the object-In this case, the first classification value corresponds to the sea-and distance information and type information of the object The reflected second classification value-In this case, the second classification value corresponds to an obstacle-is determined from the first image by using an artificial neural network learned based on the output data and labeling data corresponding to the output data. 1 object information obtaining step of obtaining object information and obtaining second object information from the second image; And calculating a moving direction of the moving object included in the imaged sea by comparing the first object information and the second object information.

In another embodiment, the first object information and the second object information may be a segmentation image obtained through an image segmentation operation.

In another embodiment, the first object information is an image in which an object included in the first image has a color value corresponding to a classification value in which object information regarding at least one of the distance information and type information is reflected, and the The second object information may be an image in which an object included in the second image has a color value corresponding to a classification value in which object information about at least one of the distance information and type information is reflected.

In another embodiment, the first object information and the second object information may be an N×M matrix.

In another embodiment, the first object information includes only a classification value for a moving object included in the first image, and the second object information includes only a classification value for a moving object included in the second image. I can.

In another embodiment, the first object information and the second object information may not include the first classification value.

In another embodiment, the calculating comprises a first moving object classification value corresponding to a first moving object spaced apart from the camera by a first distance-In this case, the first moving object classification value is distance information of the first moving object And type information-a first moving direction calculating step of calculating a moving direction of the first moving object by comparing a plurality of first moving object object information including only at a first time interval, and the first distance from the camera A second moving object classification value corresponding to the second moving object spaced apart by a larger second distance-In this case, the second moving object classification value includes distance information and type information of the second moving object. 2 Comparing the moving object object information at a second time interval greater than the first time interval, and calculating a moving direction of the second moving object.

In another embodiment, the first moving object object information and the second moving object object information may be a two-dimensional matrix.

In another embodiment, the obtaining of the object information may include matching the first image and the second image in consideration of the attitude information of the ship before obtaining the second object information.

In another embodiment, the obtaining of the object information includes the first object information and the first object information using a classification value corresponding to a moving object included in the first object information and the second object information, and the attitude information of the ship. 2 It may include the step of matching object information.

In another embodiment, the calculating comprises comparing a first position according to the first object information and a second position according to the second object information, when the moving object moves along a horizontal axis direction, It may include calculating a moving direction.

In another embodiment, the calculating includes the number of components corresponding to the moving object included in the first object information and the moving object included in the second object information when the moving object moves in a vertical axis direction. It may include the step of comparing the number of components corresponding to the calculating the moving direction of the moving object.

In another embodiment, the first object information and the second object information are each a segmentation image, and the number of components may be the number of pixels of the segmentation image.

In another embodiment, an increase in the number of components may indicate a decrease in the distance from the ship to the moving body, and a decrease in the number of components may indicate an increase in the distance from the ship to the moving body. .

In another embodiment, in the calculating, the horizontal component of the moving direction is calculated by comparing a first position according to the first object information and a second position according to the second object information, and the first object And calculating a vertical component of the moving direction by comparing the number of pixels corresponding to the moving object included in information and the number of pixels corresponding to the moving object included in the second object information.

In another embodiment, the calculating includes the step of obtaining the absolute speed of the moving object based on the time interval between the first image and the second image, the speed information of the ship, and the attitude information of the ship. can do.

According to another embodiment, the method of obtaining the moving property of the moving object may further include generating a control signal of the ship based on the calculated moving direction of the moving object.

According to another embodiment, in a method of calculating a moving direction of a moving object using an artificial neural network, a plurality of captured images including a first captured image and a second captured image that is a subsequent frame of the first captured image are acquired. Step to do; A first classification value in which the type information of the object is reflected so as to output object information included in the captured image from the captured image-In this case, the first classification value corresponds to a drivable area-and distance information of the object And a second classification value reflecting the type information-In this case, the second classification value corresponds to an obstacle-using the artificial neural network learned based on the output data and labeling data corresponding to the output data. An object information acquisition step of obtaining first object information from a first captured image and second object information from the second captured image; And calculating a moving direction of at least some of the moving objects included in the plurality of captured images by comparing the first object information and the second object information.

In yet another embodiment, the drivable area may correspond to a water surface when the artificial neural network is used for a ship, and may correspond to a road when the artificial neural network is used for a vehicle.

According to another embodiment, the camera is installed on the ship to capture the sea; And the object type information to obtain a plurality of images including a first image captured by the camera and a second image that is a subsequent frame of the first image, and to output object information included in the image from the image. -In this case, the first classification value corresponds to the sea-and a second classification value reflecting distance information and type information of the object-In this case, the second classification value corresponds to an obstacle -Obtaining first object information from the first image using an artificial neural network learned based on output data including and labeling data corresponding to the output data, obtaining second object information from the second image, and And a control unit that compares the first object information and the second object information to calculate a moving direction of the moving object included in the captured sea.

According to another embodiment, a recording medium in which a program for performing any one of the above-described methods of obtaining movement properties of a moving object is recorded may be provided.

An artificial neural network (ANN) is a type of algorithm that mathematically models the learning method of the human brain.

The artificial neural network may include a plurality of nodes, which are artificial neurons, and a synapse connecting the plurality of nodes. The artificial neural network may include a layer including at least one or more nodes. For example, the artificial neural network may include an input layer, a hidden layer, and an output layer.

The input layer may receive input data from outside the artificial neural network and transmit the input data into the artificial neural network. The hidden layer may transmit input data transmitted from the input layer and data calculated based on a coupling strength of synapses to the output layer. The output layer may calculate output data based on the coupling strength of synapses and data transmitted from the hidden layer.

The artificial neural network may include various neural networks. For example, the artificial neural network may include a convolution neural network (CNN) that extracts features using a filter. Alternatively, the artificial neural network may include a recurrent neural network (RNN) having a structure in which an output of a node is fed back as an input. In addition, artificial neural networks are of various types such as Restricted Boltzmann Machine (RBM), Deep Belief Network (DBN), Generative Adversarial Network (GAN), and Relation Networks (RN). May contain neural networks of

On the other hand, artificial neural networks can be learned in various ways. For example, the artificial neural network may include supervised learning, unsupervised learning, reinforcement learning, and imitation learning. In addition, artificial neural networks can be learned through various types of learning methods.

1 is a block diagram illustrating a method of learning an artificial neural network according to an embodiment. Specifically, FIG. 1 may represent supervised learning according to an embodiment.

Referring to FIG. 1, the artificial neural network may receive training data and output output data. The artificial neural network may be learned through backpropagation of an error calculated based on comparison between output data and labeling data.

The labeling data may be related to the learning data. For example, the labeling data may include data calculated based on the learning data.

The labeling data may include ground truth. Alternatively, the labeling data may be data generated through a user or a program.

2 is a block diagram of an inference step of an artificial neural network according to an embodiment. Referring to FIG. 2, the learned artificial neural network may receive input data and output output data.

The input data may include various types of data. For example, the input data may include image data, audio data, and text data.

The output data may include various types of data. For example, the output data may include image data, audio data, and text data.

The accuracy of the output data may vary according to the learning degree of the learned artificial neural network. Specifically, as the learning degree increases, the accuracy of the output data may increase.

Hereinafter, a method of acquiring information on surrounding obstacles using an artificial neural network will be described.

3 is a diagram illustrating a method of obtaining object information using an artificial neural network according to an embodiment.

Referring to FIG. 3, the artificial neural network may receive input data and output output data. For example, the artificial neural network may receive the first image data 1000 and output the second image data 2000.

The first image data 1000 may be an image captured by a camera.

The second image data 2000 may be data generated based on the first image data 1000. For example, the second image data 2000 may include object information including at least one of type information and distance information of an obstacle included in the first image data 1000.

The artificial neural network may receive the first image data 1000 and perform an image segmentation operation.

The image segmentation operation is an image segmentation operation, and may mean an operation of dividing an image area according to attributes. The image segmentation operation may include a process of allocating a predetermined attribute value for each pixel of the image. For example, the property may mean the type of object included in the image. That is, the image segmentation operation may include a process of dividing an object included in an image for each pixel. Alternatively, the image segmentation operation may mean indicating which object a specific pixel corresponds to.

The attribute value can be expressed in various ways. For example, the attribute value may be expressed in color.

The image segmentation operation may be performed by a plurality of artificial neural networks. For example, the plurality of artificial neural networks may each perform the image segmentation operation and obtain object information by combining the results of the image segmentation.

Meanwhile, the artificial neural network may have various structures. For example, the artificial neural network may have an ENet structure.

Meanwhile, the first image data 1000 may be provided in various forms. For example, as illustrated in FIG. 3, the first image data 1000 may be provided as an image. Alternatively, the first image data 1000 may be provided as pixel data.

4 is a diagram for explaining an image segmentation operation of an artificial neural network according to an embodiment.

Referring to FIG. 4, the artificial neural network may receive a first image 1100 and output first output data 2100. The first output data 2100 may be an N×M matrix. In this case, N and M may be the same. The number of elements of the matrix may be the same as the number of pixels of the first image 1100. That is, the first image 1100 may include N×M pixels.

Each pixel of the first image 1100 may correspond to each element of the matrix. For example, each pixel of the first image 1100 may correspond to each element of the matrix one-to-one. Alternatively, a set of a plurality of pixels of the first image 1100 may correspond to each element of the matrix.

The first output data 2100 may include various types of information. For example, the first output data 2100 may include distance information of an object included in the first image 1100. Alternatively, the first output data 2100 may include information on the type of the object.

Meanwhile, each element of the matrix may have a predetermined classification value. Here, the classification value may be a value in which object information included in each pixel of the first image 1100 is reflected. Accordingly, each element of the matrix may include object information corresponding to each pixel of the first image 1100.

The classification value may be determined by various object information. For example, the classification value may be determined based on distance information and type information of an object.

5 is a table showing classification values according to an embodiment.

Referring to FIG. 5, the classification value may include object information. For example, the classification value may include distance information and type information of an object.

Meanwhile, the distance information may have a specific distance value in meters. For example, the distance information may have a distance value such as 10m.

Alternatively, the distance information may be classified into a category having a predetermined range. For example, the category may include a short distance, a medium distance, and a far distance according to the distance of the object. Specifically, 0-10m may be classified into a short distance, 10-20m a medium distance, and 20-30m a long distance.

The type information may include data on the type of object.

For example, the objects may be classified into topography, fixed obstacles, dynamic obstacles, and others according to their type. The terrain may include mountains. The fixed obstacle may include an island, a reef, and the like. The dynamic obstacle may include a ship.

Meanwhile, the classification value may be determined based on the distance information and the type information. For example, the first classification value may indicate a case where the object is a terrain located in a short distance.

Meanwhile, at least some of the classification values may not include distance information. For example, the first classification value corresponding to the sea may not include distance information.

That is, the classification value is not necessarily determined based on both distance information and type information. For example, the classification value may be determined based only on type information.

In addition, the classification value may be determined based on the distance information and the type information as well as additional information. For example, the additional information may include direction information of an object, speed information, route markers, and the like.

Meanwhile, the output data output from the artificial neural network may be image data. For example, the image data may include RGB data corresponding to the classification value.

Referring back to FIG. 3, the artificial neural network may receive first image data 1000 and output second image data 2000. Each pixel of the second image data 2000 may include object information included in the first image data 1000.

Each pixel of the second image data 2000 may include RGB data corresponding to a classification value in which the object information is reflected.

For example, the first object 100 is a dynamic obstacle located in a short distance and may correspond to the classification value 7 of FIG. 5. Also, the second object 200 is a terrain located at a distance and may correspond to the classification value 3 of FIG. 5.

The classification value 7 may correspond to the seventh color. The classification value 3 may correspond to a third color.

Accordingly, the first object 100 may be expressed in the seventh color. The second object 200 may be expressed in the third color.

Meanwhile, the artificial neural network may perform a pre-processing operation before performing the image segmentation operation.

For example, the artificial neural network may perform an operation of receiving a plurality of images and selecting some of the plurality of images.

Specifically, the artificial neural network may generate a third image by synthesizing the first image with the highest illuminance and the second image with the lowest illuminance among the plurality of images. The artificial neural network may obtain output data by performing the image segmentation operation based on the third image.

Accuracy of output data obtained from the artificial neural network may be improved through the selection operation.

In addition, the artificial neural network may sample one image from among a plurality of images. For example, the artificial neural network may sample an image having the largest focus value (or focus measure) among the plurality of images and the closest illuminance value to a predetermined value. The artificial neural network may perform the image segmentation operation based on the sampled image. Accordingly, accuracy of output data calculated through the image segmentation operation may be improved.

As another example of the preprocessing operation, the artificial neural network may perform an RGB normalization operation.

Meanwhile, the artificial neural network can be learned in various ways.

For example, the artificial neural network may receive training data and output output data. As an error calculated based on a difference between the output data and labeling data related to the training data is back-propagated to the artificial neural network, the artificial neural network may be trained.

The learning data may be image data. The image data may include a sea image obtained by capturing a sea image.

The training data may include a plurality of image data obtained through data augmentation from an arbitrary image.

6 is a diagram for describing data expansion according to an exemplary embodiment. Referring to FIG. 6, a plurality of generated images may be generated from an original image 1300. A plurality of generated images may be generated from the original image 1300 by reflecting various weather conditions or environmental noise.

For example, the first generated image 1310 may be an image in which fog is added to the original image 1300. Alternatively, the degree of fog of the first generated image 1310 may be greater than the degree of fog of the original image 1300. The second generated image 1320 may be an image in which a ratio is added to the original image 1300. The first generated image 1310 may be an image in which fog and rain are added to the original image 1300.

The artificial neural network may be learned based on a first generated image 1310, a second generated image 1320, and a third generated image 1330. Accordingly, the learning efficiency of the artificial neural network can be improved.

Meanwhile, the artificial neural network may be learned based on various labeling data. For example, when the artificial neural network outputs a classification value corresponding to an object included in an input image, the artificial neural network may be learned based on labeling data including a classification value of the object.

As another example, when the artificial neural network outputs RGB data corresponding to a classification value corresponding to an object included in an input image, the artificial neural network may be learned based on RGB data corresponding to the classification value related to the object. have.

The pre-processing operation may be performed in the learning step of the artificial neural network. Accordingly, the learning efficiency of the artificial neural network may be improved.

In the above, a method of obtaining object information of an obstacle through an artificial neural network that performs image segmentation has been described.

Hereinafter, a method of obtaining the moving property of an obstacle will be described.

7 is a diagram illustrating a method of obtaining movement properties of a moving object according to an exemplary embodiment.

Referring to FIG. 7, movement attribute data 1320 may be obtained based on a comparison between an image 1300 of a first frame and an image 1310 of a second frame that is a subsequent frame of the image 1300 of the first frame. I can.

For example, the controller 1 may obtain the moving direction of the moving object 10 by comparing the pixel value of the image 1300 of the first frame and the pixel value of the image 1310 of the second frame. In Fig. 7, it can be seen that the moving body 10 is moving from left to right.

In this case, the control unit 1 may obtain the moving direction of the moving object 10 by performing an inter-frame subtraction operation.

In addition, the controller 1 may obtain the moving speed of the moving object 10 based on the pixel value and the frame interval.

The controller 1 may obtain movement attribute data 1320. The movement attribute data 1320 may include a movement direction and a movement speed of the moving object 10. The movement attribute data 1320 may be image data as shown in FIG. 7, but is not limited thereto, and may be data in various forms.

Meanwhile, in the present specification, a series of operations in which the control unit 1 acquires the movement property of the moving object may be referred to as an operation of the control unit 1 obtains the movement property.

8 is a flowchart illustrating a method of obtaining a moving property of a moving object according to another exemplary embodiment.

Referring to FIG. 8, in the method of obtaining a moving property of a moving object according to an embodiment, the step of acquiring a first image and a second image captured of a sea (S1000), the first image and the first image using an artificial neural network. 2 Acquiring first object information and second object information from an image (S2000), and comparing the first object information and the second object information to obtain movement properties of the moving object included in the captured sea (S3000) may be included.

Hereinafter, each step will be described in detail.

First, the controller 1 may acquire a plurality of images including a first image and a second image captured by the sea (S1000). The first image and the second image may be images of different frames. For example, the second image may be an image of a frame subsequent to the first image.

Each of the first image and the second image may include a plurality of objects. For example, the plurality of objects may include various objects such as sea, ship, and buoy.

The controller 1 may obtain first object information and second object information, respectively, from the first image and the second image using an artificial neural network (S2000).

The artificial neural network may mean the artificial neural network described in FIGS. 1 to 6. For example, the artificial neural network may receive an input image and output object information included in the input image. Specifically, the artificial neural network may output distance information and type information of an obstacle included in the input image by performing the image segmentation operation.

In addition, the artificial neural network may perform the pre-processing operation.

The first object information and the second object information may refer to object information obtained through the above-described image segmentation operation. For example, the first object information and the second object information may mean distance information and type information about an object included in the first image and the second image, respectively.

The first object information and the second object information may be provided in an N×M matrix. Elements of the matrix may correspond to pixels of the first image and the second image.

The controller 1 may compare the first object information and the second object information to obtain a movement property of the moving object included in the captured sea (S3000).

For example, the controller 1 may obtain a moving direction of the moving object by comparing the first object information and the second object information.

The controller 1 may compare an element included in the first object information and an element included in the second object information to obtain a moving direction of the moving object. The element may include a classification value reflecting distance information and type information about an object included in the first image and the second image.

Meanwhile, the movement direction may mean a relative movement direction based on an installation position of a camera that has captured the plurality of images.

The controller 1 may calculate the moving direction of the object in consideration of type information of the object included in the first image and the second image. For example, the controller 1 may calculate the moving direction of the object only when the object is a dynamic obstacle. That is, when the object corresponds to a terrain or a fixed obstacle, the controller 1 may not calculate a movement direction for the object.

Alternatively, the period of the operation of calculating the moving direction of the controller 1 may vary according to the type of the object. For example, when the object is a dynamic obstacle, a period of the movement direction calculation operation may be smaller than when the object is a fixed obstacle.

The controller 1 may calculate the moving direction of the object in consideration of distance information of the object included in the first image and the second image. For example, when the object is located in a short distance, a period of the operation of calculating the movement direction of the controller 1 may be smaller than when the object is located in a far distance.

The controller 1 may obtain a moving speed of the moving object by comparing the first object information and the second object information. For example, the control unit 1 may obtain the moving speed of the moving object using the frame interval of the first image and the second image, and the first object information and the second object information.

In this case, the controller 1 may obtain the moving speed by considering distance information of an object included in the first object information and distance information of an object included in the second object information. For example, the movement speed may be obtained based on the distance information and the change in element values of the first object information and the second object information.

Meanwhile, the moving speed may mean a relative moving speed based on an installation position of a camera that has captured the plurality of images.

The controller 1 may calculate the absolute speed of the moving object based on the moving speed. For example, the control unit 1 may calculate the absolute speed in consideration of GPS information of a ship on which the camera is installed.

Meanwhile, the controller 1 may perform the movement attribute acquisition operation in consideration of the attitude information of the ship on which the camera is installed. For example, the control unit 1 may perform the movement attribute acquisition operation in consideration of the attitude information of the ship acquired from the IMU mounted on the ship. The attitude information may include a heading or yaw of the ship, a pitch, a roll, and the like.

Specifically, the controller 1 may match the first image and the second image in consideration of the posture information. For example, the positions of the first image and the second image may be matched in consideration of the pitch of the ship. In this case, the controller 1 may match the first image and the second image based on a predetermined reference position.

For example, the step S2000 and the step S3000 may include the image matching step.

9 is a diagram illustrating a method of obtaining a moving property of a moving object according to an exemplary embodiment.

Referring to FIG. 9, the controller 1 is based on a comparison of first object information 1400 obtained from a first image and second object information 1410 obtained from a second image that is a subsequent frame of the first image. Thus, the first movement attribute data 1420 may be obtained.

The first object information 1400 and the second object information 1410 may include distance information and type information about the moving object 10.

The controller 1 may obtain the moving property of the moving object 10 by comparing the first object information 1400 and the second object information 1410.

For example, the controller 1 compares the position of the moving object 10 according to the first object information 1400 and the position of the moving object 10 according to the second object information 1410 to determine the moving direction of the moving object 10 Can be obtained. Specifically, the controller 1 may obtain the moving direction of the moving object 10 based on a change in a position in a matrix to which a classification value corresponding to the moving object 10 belongs.

The controller 1 may obtain the moving speed of the moving object 10 by using the frame interval of the first image and the second image and distance information of the moving object 10.

The first object information 1400 and the second object information 1410 may be provided as various types of data. For example, the first object information 1400 and the second object information 1410 may be provided in the form of an image. Alternatively, the first object information 1400 and the second object information 1410 may be provided in a 2D matrix.

Meanwhile, the moving properties of the moving object 10 may be classified.

10 is a table showing classification values for movement attributes according to an exemplary embodiment.

Referring to FIG. 10, a classification value may be allocated according to each movement attribute. The movement attribute may include a movement direction and a movement speed.

For example, the moving direction may be classified into NOT MOVING, FORWARD, BACKWARD, LEFT, RIGHT, and the like. However, the present invention is not limited thereto, and the direction of movement may be classified in various ways. For example, the direction of movement may be classified into eight orientations such as east, west, south, north, northeast, southeast, southwest, and northwest.

On the other hand, the moving direction may indicate a direction based on a position in which the diffuser camera is installed. For example, when a plurality of cameras are installed on a ship, a direction indicated by each movement direction may be different depending on the installation position of the camera.

Alternatively, the moving direction may indicate a direction based on the moving direction of the ship on which the camera is installed. For example, the FORWARD may indicate the same direction as the moving direction of the ship. In addition, the backward may indicate a direction opposite to the moving direction of the ship.

The moving speed can be classified into three types of low speed (SLOW), medium speed (NORMAL), and high speed (FAST) based on a predetermined range. However, the present invention is not limited thereto, and the moving speed may be classified into various numbers according to the speed.

11 is a diagram illustrating a method of obtaining a moving property of a moving object according to another exemplary embodiment.

Referring to FIG. 11, the controller 1 may obtain second movement attribute data 1520 based on the third object information 1500 and the fourth object information 1510. The third object information 1500 may be output from the first image through the artificial neural network. The fourth object information 1510 may be output from the second image through the artificial neural network.

Meanwhile, the controller 1 may perform the operation of obtaining the moving property of the moving object 10 in the same manner as the operation of obtaining the moving property described in FIGS. 7 to 9. For example, the controller 1 may obtain a moving direction and a moving speed of the moving object 10 by comparing the third object information 1500 and the fourth object information 1510. Accordingly, a detailed description thereof will be omitted, and a description will be made focusing on the difference from the above-described movement attribute acquisition operation.

The control unit 1 may perform an operation of obtaining movement attributes only on objects having a predetermined classification value. For example, the controller 1 may obtain the moving property of the dynamic obstacle only when the third object information 1500 and the fourth object information 1510 include a classification value corresponding to the dynamic obstacle.

The controller 1 may perform an operation of selecting an object having a predetermined classification value. For example, the controller 1 may perform an operation of selecting a dynamic obstacle.

The controller 1 may perform an operation of discarding information other than the selected object from the third object information 1500 and the fourth object information 1510. For example, in the third object information 1500, the controller 1 may adjust values of the remaining elements to 0 except for an element corresponding to the selected object. In the fourth object information 1510, the controller 1 may adjust values of the remaining elements to 0 except for an element corresponding to the selected object.

For example, the control unit 1 may generate the third object information 1500 from the first object information 1400 from the first object information 1400 from which information other than the object information about the moving object 10 has been removed. I can. That is, the third object information 1500 may include only object information regarding the moving object 10 of the first object information 1400.

Likewise, the controller 1 may generate the fourth object information 1510 from the second object information 1410 from the second object information 1410 to which information other than the object information about the moving object 10 has been removed. . That is, the fourth object information 1510 may include only object information regarding the moving object 10 of the second object information 1410.

The controller 1 may obtain a movement property of the moving object 10 by comparing the third object information 1500 and the fourth object information 1510. For example, the controller 1 may obtain a moving direction and a moving speed of the moving object 10 based on a change in a position in a matrix of classification values corresponding to the moving object 10.

Accordingly, the total amount of calculation of the controller 1 for obtaining the moving property of the moving object may be reduced. That is, when the controller 1 acquires the movement attribute based on selected object information including only information on a predetermined object, the amount of data calculation of the controller 1 can be reduced.

Meanwhile, the execution cycle of the movement attribute acquisition operation performed by the control unit 1 may vary according to the distance information of the moving object. For example, when a near-distance dynamic obstacle is included in the first image, the controller 1 may obtain the moving property of the dynamic obstacle at a first time interval. On the other hand, when a distant dynamic obstacle is included in the first image, the controller 1 may acquire the moving property of the dynamic obstacle at a second time interval greater than the first time interval.

12 is a diagram for describing a method of obtaining movement properties of a moving object according to an exemplary embodiment.

Referring to FIG. 12, the artificial neural network may output information on a plurality of objects corresponding to the plurality of images from a plurality of images captured at sea.

For example, when the first moving object 11 is included in the captured sea, the control unit 1 may perform the operation of obtaining the moving attribute every N frames. On the other hand, when the second moving object 12 is included in the captured sea, the control unit 1 may perform the operation of obtaining the moving attribute for every M frame larger than N. The second moving body 12 may be located farther from the camera that has captured the sea than the first moving body 11. That is, the closer the distance from the moving object to the camera is, the more frequently the operation of obtaining the movement attribute of the controller 1 may be performed.

This is because the closer the moving object from the camera, the faster the position in the image may change. Alternatively, as the moving object closer to the camera is, the change in the element value in the object information may fluctuate rapidly.

On the other hand, depending on the moving direction of the moving object, the method of obtaining the moving attribute of the control unit 1 may vary. For example, when the moving object moves in the horizontal direction, the controller 1 may obtain the moving property of the moving object based on a change in the position of the classification value corresponding to the moving object as described above. On the other hand, when the moving object moves in the vertical direction, the controller 1 may obtain the moving property of the moving object based on a change in the number of pixels corresponding to the moving object.

13 is a diagram for describing a method of obtaining movement properties of a moving object according to an exemplary embodiment.

Referring to FIG. 13, the controller 1 may acquire third movement attribute data 1620 based on the fifth object information 1600 and the sixth object information 1610. The fifth object information 1600 may be output from the first image through the artificial neural network. The sixth object information 1610 may be output from the second image through the artificial neural network. Accordingly, the sixth object information 1610 may be information about an image of a frame following the fifth object information 1600.

The moving body 10 can move along a vertical direction. Here, the vertical direction may mean a vertical axis direction of the camera. Specifically, the vertical direction may mean a direction moving away from or close to the camera. Alternatively, the vertical direction may mean the FORWARD or BACKWARD direction of FIG. 10.

When the moving object 10 moves in the vertical direction, the controller 1 may determine the moving object 10 based on a change in the number of pixels corresponding to the moving object 10 included in the first and second images. You can acquire the movement property of.

Alternatively, the control unit 1 may obtain the moving property of the moving object 10 based on a change in the number of classification values corresponding to the moving object 10 included in the fifth object information 1600 and the sixth object information 1610. I can. The number of classification values may mean the number of elements having the classification values included in the fifth object information 1600 and the sixth object information 1610.

For example, when the number of classification values corresponding to the moving object 10 included in the sixth object information 1610 is greater than the number of classification values corresponding to the moving object 10 included in the fifth object information 1600 That is, when the number of classification values increases over time, it may mean that the moving object 10 is closer to the camera. Alternatively, it may mean that the moving body 10 is closer to the vessel.

Conversely, when the number of classification values corresponding to the moving object 10 included in the sixth object information 1610 is smaller than the number of classification values corresponding to the moving object 10 included in the fifth object information 1600, that is, , If the number of classification values decreases with time, it may mean that the moving object 10 is distant from the camera. Alternatively, it may mean that the moving body 10 is away from the ship.

On the other hand, it goes without saying that the control unit 1 can perform the movement attribute acquisition operation of the moving object 10 in the same manner as the movement attribute acquisition operation described in FIGS. 7 to 11. For example, the controller 1 may obtain the moving direction and the moving speed of the moving object 10 based on the change in the position of the moving object 10 included in the fifth object information 1600 and the sixth object information 1610. I can.

In this case, the control unit 1 may acquire the movement property of the moving object 10 in consideration of the attitude information of the ship on which the camera is installed. For example, the movement attribute may be obtained by considering the heading or yaw, pitch, and roll of the ship.

Meanwhile, the movement attribute of the moving object 10 obtained based on the change in the number of classification values may be more accurate than the movement attribute of the moving object 10 obtained based on the position change. For example, the number of classification values may be less affected by a change in attitude of the ship than the position. Specifically, the position may change as the attitude of the ship changes, while the number of classification values may be maintained even if the attitude of the ship changes. Alternatively, the position may change relatively largely as the attitude of the ship changes, while the number of classification values may change relatively little even if the attitude of the ship changes.

The control unit 1 may generate a control signal of the vessel based on the moving property of the moving body 10. For example, when the moving object 10 approaches the ship, the control unit 1 may generate a control signal that avoids the moving object 10. Alternatively, the control unit 1 may generate a control signal so that the ship avoids the moving body 10 when the moving body 10 enters a preset danger area based on the ship.

Meanwhile, in the above, it has been described that the control unit 1 acquires the movement property of the moving object included in the marine image from the sea image captured by the camera installed on the ship, but the control unit 1 obtains the movement property of the moving object from various images. can do. For example, the controller 1 may obtain the moving property of the moving object included in the captured image from a captured image captured by a camera installed in the vehicle. Alternatively, the controller 1 may obtain the moving property of the moving object included in the image from an image captured from a camera installed in the drone. Also, the controller 1 may generate a control signal for controlling a vehicle or a drone based on the movement property.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc. alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the embodiment, or may be known and usable to those skilled in computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. -A hardware device specially configured to store and execute program instructions such as magneto-optical media, and ROM, RAM, flash memory, and the like. Examples of the program instructions include not only machine language codes such as those produced by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like. The hardware device described above may be configured to operate as one or more software modules to perform the operation of the embodiment, and vice versa.

As described above, although the embodiments have been described by the limited embodiments and drawings, various modifications and variations are possible from the above description by those of ordinary skill in the art. For example, the described techniques are performed in a different order from the described method, and/or components such as a system, structure, device, circuit, etc. described are combined or combined in a form different from the described method, or other components Alternatively, even if substituted or substituted by an equivalent, an appropriate result can be achieved.

Therefore, other implementations, other embodiments, and claims and equivalents fall within the scope of the claims to be described later.

Claims (1)

A method of acquiring the moving property of a moving object using an image captured by a camera and an artificial neural network,
Acquiring a plurality of images including a first image captured at sea and a second image that is a subsequent frame of the first image;
In order to output object information included in the image from the image, a first classification value reflecting the type information of the object-In this case, the first classification value corresponds to the sea-and distance information and type information of the object The reflected second classification value-In this case, the second classification value corresponds to an obstacle-is determined from the first image by using an artificial neural network learned based on the output data and labeling data corresponding to the output data. 1 object information obtaining step of obtaining object information and obtaining second object information from the second image; And
Comprising: comparing the first object information and the second object information to calculate a moving direction of the moving object included in the captured sea.
How to obtain moving properties of moving objects.

Images