KR102070956B1 - Apparatus and method for processing image - Google Patents

Abstract

The present specification relates to an image processing apparatus and a method. The image processing apparatus according to the present embodiment includes a feature extractor which extracts visual information by applying a convolution filter to an input frame, a patch splitter which obtains a feature patch by dividing a feature map including the visual information by pixel, and a feature A temporal information processor for extracting temporal information on the patch, a probability classifier for extracting one probability result per pixel from the temporal information, and a background separator for separating the foreground and background of the input frame based on the extracted probability result; .

Description

Image processing apparatus and method {APPARATUS AND METHOD FOR PROCESSING IMAGE}

Embodiments disclosed herein relate to an image processing apparatus and a method. More specifically, the present invention relates to an image processing apparatus and method for separating a foreground and a background of a video using an artificial neural network.

The existing image processing method uses a comparison value of the input image to compare the previous image with the current image to separate the foreground and the background. In the related art, Korean Patent No. 10-2013-0063963 discloses a cumulative image change value by accumulating change values between images by comparing a current image with a previous image, and using the accumulated image change value. An image processing method for estimating a change value of an image due to object movement is described.

Conventional image processing methods including this set the front part of the video as the learning section of the background model to learn all the areas as the background model without foreground separation, and separate the foreground and background for the input image after the learning section, and update the background model. And so on. In this case, various techniques are used to reduce noise and improve performance in learning, separating, and updating the background, and the user directly designs for noise reduction and uses different parameters for each image characteristic. As a result, all parts for image processing must be designed by the user, complicated memory update operations are performed, and the speed can be limited because the CPU operation is focused on.

As such, existing image processing methods have a problem that performance is limited due to inefficiency of hardware resource use because it is not easy to implement because the user needs to design the image by using a non-learning based image processing method.

Therefore, in recent years, an apparatus and method for solving such a problem are required.

On the other hand, the background art described above is technical information that the inventors possessed for the derivation of the present invention or acquired in the derivation process of the present invention, and is not necessarily a publicly known technique disclosed to the general public before the present application. .

Embodiments disclosed herein are provided to provide an image processing apparatus and method for separating foreground and background of a video using an artificial neural network.

Embodiments disclosed herein are intended to provide an image processing apparatus and method for extracting temporal information of a video using an artificial neural network and separating the foreground and the background using the extracted temporal information.

Embodiments disclosed herein are provided to provide an image processing apparatus and method that can automatically learn, separate, or update a background using an artificial neural network without user intervention.

Embodiments disclosed herein are intended to provide an image processing apparatus and method capable of detecting motion information by learning an artificial neural network so as to utilize not only visual information but also temporal information.

As a technical means for achieving the above technical problem, according to an embodiment, the image processing apparatus is a feature extractor for extracting visual information by applying a convolution filter to the input frame, the feature map including the visual information in units of pixels A patch partition unit for dividing into a feature patch to obtain a feature patch, a temporal information processing unit for extracting temporal information about the feature patch, a probability classifier for extracting one probability result for each pixel from the temporal information, and based on the extracted probability result And a background separator for separating the foreground and the background of the input frame.

According to another embodiment, an image processing method performed by an image processing apparatus may include extracting visual information by applying a convolution filter to an input frame, dividing a feature map including the visual information into pixel units and patching the feature Acquiring a symbol, extracting temporal information on the feature patch, extracting one probability result for each pixel from the temporal information, and separating the foreground and the background of the input frame based on the extracted probability result. Steps.

According to any one of the aforementioned problem solving means, an image processing apparatus and method for separating the foreground and the background of a video using an artificial neural network can be proposed.

In addition, according to any one of the aforementioned problem solving means, it is possible to provide an image processing apparatus and method for extracting the temporal information of the video using an artificial neural network, and separating the foreground and the background by using the extracted temporal information.

In addition, according to any one of the above-described problem solving means, it is possible to provide an image processing apparatus and method that can automatically learn the learning, separation, or update of the background without user intervention using an artificial neural network.

In addition, according to any one of the above-described problem solving means, it is possible to provide an image processing apparatus and method that can detect the motion information by learning the artificial neural network to utilize not only visual information but also temporal information.

Effects obtained in the disclosed embodiments are not limited to the above-mentioned effects, and other effects not mentioned above are clearly understood by those skilled in the art from the following description. Could be.

1 is a block diagram illustrating an image processing apparatus according to an exemplary embodiment.
2 is a diagram for describing a process of extracting visual information from an input frame, according to an exemplary embodiment.
3 is a diagram for describing a process of extracting a feature patch from a feature map, according to an exemplary embodiment.
4 is a diagram for describing a process of processing temporal information, according to an exemplary embodiment.
5 is a view for explaining the internal structure of the circulatory neural network shown in FIG.
6 is a diagram for describing a process of outputting a probability result for each pixel, according to an exemplary embodiment.
7 is a flowchart illustrating an image processing method, according to an exemplary embodiment.

DETAILED DESCRIPTION Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those skilled in the art may easily implement the present invention. As those skilled in the art would realize, the described embodiments may be modified in various different ways, all without departing from the spirit or scope of the present invention. In the drawings, parts irrelevant to the description are omitted in order to clearly describe the present invention, and like reference numerals designate like parts throughout the specification.

Throughout the specification, when a part is "connected" to another part, it includes not only "directly connected" but also "electrically connected" with another element in between. . In addition, when a part is said to "include" a certain component, this means that it may further include other components, except to exclude other components unless otherwise stated.

Hereinafter, the present invention will be described in detail with reference to the accompanying drawings.

1 is a block diagram illustrating an image processing apparatus according to an exemplary embodiment.

As illustrated in FIG. 1, the image processing apparatus 100 may include a feature extractor 110, a patch divider 120, a temporal information processor 130, a probability classifier 140, and a background separator 150. It includes.

The feature extractor 110 may extract visual information that may be configured as a feature map from an input frame. Here, the input frame may be one of a plurality of frames constituting the video.

The feature map extracted by the feature extractor 110 includes visual information about at least a portion of an area in the input frame. The feature extractor 110 may use a convolution filter to obtain a feature map. Here, the convolution filter may be implemented using a convolutional neural network (hereinafter, referred to as 'CNN'). Accordingly, the feature extractor 110 may learn a convolution filter for background separation using a CNN from an input frame, that is, an RGB image, and extract a feature map using the learned convolution filter. .

The patch dividing unit 120 may obtain a feature patch by dividing the feature map by a pixel unit. When the feature map includes a plurality of layers, the patch splitter 120 may divide the feature map into feature patches including all visual information about each of the plurality of layers. Since the patch dividing unit 120 divides the feature map in pixel units, the patch divider 120 may obtain a plurality of feature patches from one feature map.

The temporal information processing unit 130 may extract temporal information for each feature patch. Here, the temporal information processing unit 130 may use a recurrent neural network (hereinafter, referred to as 'RNN') for extracting temporal information. In this case, the temporal information processor 130 extracts temporal information included in the video by calculating the current feature patch using information extracted from the previous feature patch. In addition, the temporal information processor 130 may update an internal state of the cyclic neural network, that is, a cell state.

The probability classifier 140 may extract a probability result for each pixel from the temporal information. To this end, the probability classifier 140 may use a Fully Connected Neural Network (FCNN).

The background separator 150 may detect a movement of a specific object in the input frame based on the extracted probability result. In this way, the background separator 150 may separate the foreground and the background in the input frame.

As such, the proposed image processing apparatus 100 may separate an image by using an artificial neural network. As described above, the image processing apparatus 100 may use a structure in which artificial neural networks of CNN, RNN, and FCNN are combined in order to detect movement of a specific object in a video. In particular, the image processing apparatus 100 enables the learning of an artificial neural network utilizing temporal information, instead of relying only on visual information of a video using an RNN structure.

In addition, the image processing apparatus 100 may automatically learn learning, separation, or updating of the background using an artificial neural network without a user's direct design, and there is little performance change due to parameter selection.

2 is a diagram for describing a process of extracting visual information from an input frame, according to an exemplary embodiment.

As shown in FIG. 2, the feature extractor 110 may extract visual information from an input frame using a convolution filter, and may use a CNN for this purpose.

The feature extractor 110 may use the convolution filter 220 for the partial region 211 of the input frame 210, and may generate visual information (or CNN feature) that may be configured as the feature map 230. Can be extracted. In FIG. 2, the feature map 230 is illustrated as including four layers 231, 232, 233, and 234, but is not limited thereto. The feature map 230 may have one or more layers using a convolution filter. It can be set to various numbers.

The feature extractor 110 may learn a convolution filter required for background separation from the input image 210. Here, the feature extractor 110 may serve as a filter bank as a convolution filter to learn and generate a plurality of feature maps 230. Through this, the feature extractor 110 may automatically learn and extract information necessary for background separation.

Meanwhile, the above-described CNN may support various functions for searching, classifying, and understanding images for image processing through learning about images as one of artificial neural networks. Since the CNN has a different structure from other artificial neural networks that connect and learn all areas of the input, one parameter can be used in various areas of the input frame. As a result, the CNN can process the image even with a small number of parameters, and since the characteristic does not change even if the image is spatially different, the CNN can be used to extract visual information from the image analyzing apparatus 100.

3 is a diagram for describing a process of extracting a feature patch from a feature map, according to an exemplary embodiment.

As illustrated in FIG. 3, the patch division unit 120 may perform patch division on each of the plurality of layers 231, 232, 233, and 234 through patch division of the feature map 230. In this way, the patch division unit 120 may obtain the feature patch 240 in units of pixels from the feature map 230. Here, the feature patch 240 may include visual information in pixel units of each of the plurality of layers 231, 232, 233, and 234. In addition, the number of layers of the feature map 230 may be divided into L (here, L = 4), where L is a natural number of 1 or more.

The patch divider 120 may generate a plurality of feature patches in units of pixels by dividing the feature map 230. In addition, the patch splitter 120 may divide the data at a predetermined interval in consideration of the image processing speed.

Meanwhile, as the number L of layers of the feature map 230 divided by the patch dividing unit 120 increases, the visual information for separating the foreground and the background may be further included. For this reason, when the number of layers in the feature map 230 is increased, performance may be improved due to separation of the foreground and the background.

4 is a diagram for describing a process of processing temporal information, according to an exemplary embodiment.

As shown in FIG. 4, the temporal information processor 130 may extract the temporal feature 260 from the feature patch 240. Here, the temporal information processor 130 may extract temporal features 260, which are information on the flow of visual information over time, and are prediction values for separating the foreground and the background. Here, the temporal information processor 130 may be implemented using the RNN 250.

In addition, the temporal information processor 130 may output temporal information 260 in pixel units, and may receive a cell state of a previous input as a next input. In this way, the temporal information processor 130 may extract temporal information 260, which is a prediction value for separation of the foreground and the background, by using the feature vector extracted from the feature patch 240 in units of pixels.

5 is a view for explaining the internal structure of the circulatory neural network shown in FIG.

As shown in FIG. 5, the RNN 250 is a neural network configured to receive and process information of a previous input by using an internal loop for a series of sequential inputs. Accordingly, the RNN 250 may perform different operations for each input by using internal state information on sequential inputs, and may learn a matrix for updating cell state information through learning.

The RNN 250 may be implemented as long and short term memory (hereinafter, referred to as 'LSTM'). The LSN RNN 250 may include a cell state operation line 251, a forget gate layer 252, an input gate layer 253, and an update cell state ( Update cell state 254, and an output gate layer 255. Here, the description will be made based on the RNN 250-2, and the RNN 250-1 and the RNN 250-3 may have a structure similar to that of the RNN 250-2. Are shown for each representation.

The cell state line 251 includes a multiplication operation and an addition operation. The cell state line 251 may receive C _{t- 1} , which is a previously output internal state (ie, a cell state), and output a current cell state C _t . The cell state 251 determines whether to reflect visual information (ie, visual information included in the feature patch) by using the forge gate layer 252, the input gate layer 253, and the output gate layer 255. Can be.

) 연산을 이용하여, h_t-1과 x_t를 입력으로 수신한다. 여기서, x_t는 현재 프레임의 특징 패치(240)이다. 따라서, x_{t -1}은 이전 프레임의 특징 패치이고, x_{t +1}은 다음 프레임의 특징 패치이다. 또한, h_{t- 1}는 히든 스테이트로서, 이전 RNN(250-1)의 출력값을 나타낸다. 포겟 게이트 레이어(252)는 시그모이드 연산을 사용하여 0에서 1사이의 값을 출력(0<

<1)할 수 있으며, 시그모이드 연산의 출력값은 1에 가까울수록 정보의 반영을 많이 하고, 0에 가까울수록 정보의 반영을 적게하는 것을 의미한다.Forget gate layer 252 is a sigmoid (

Operation, we receive h _t-1 and x _t as inputs. Where x _t is the feature patch 240 of the current frame. Thus, x _{t -1} is the feature patch of the previous frame and x _{t +1} is the feature patch of the next frame. In addition, h _{t- 1} is a hidden state, and represents the output value of the previous RNN 250-1. Forget gate layer 252 outputs a value between 0 and 1 using sigmoid operation (0 <

<1), the output value of the sigmoid operation means that the information is reflected more as the value is closer to 1, and the reflection of information is less as the value is closer to 0.

The operation of the forge gate layer 252 may be represented by Equation 1 below.

Here, f _t represents the output of the forge gate layer 252. W _f is a weight of the forge gate, b _f is a bias value of the forge gate, and W _f and b _f are values set by learning of the RNN.

) 연산과 쌍곡 탄젠트(tanh) 연산을 포함할 수 있다. 여기서, 시그모이드 연산은 어떤 값을 업데이트할지를 결정하며, 쌍곡 탄젠트 연산은 셀 상태에 더해질 후보값들의 벡터를 만든다. 이때, 입력 게이트 레이어(253)에서 수행되는 시그모이드 연산과 쌍곡 탄젠트 연산 각각은 하기의 수학식 2와 같이 나타낼 수 있다.Next, the input gate layer 253 determines whether new information is stored in the cell state. Input gate layer 253 is a sigmoid (

) And hyperbolic tangent (tanh) operation. Here, the sigmoid operation determines which value to update, and the hyperbolic tangent operation produces a vector of candidate values to be added to the cell state. In this case, each of the sigmoid operation and the hyperbolic tangent operation performed in the input gate layer 253 may be represented by Equation 2 below.

Where i _t is the output of the sigmoid operation in the input gate layer 253 and c _t is the output of the hyperbolic tangent operation in the input gate layer 253. Again, W _i is the weight of the sigmoid operation, and W _c is the weight of the hyperbolic tangent operation. In addition, b _i is a bias value of the sigmoid operation, b _c is a bias value of the hyperbolic tangent operation. W _i , W _c , b _i , and b _c are values set by learning of the RNN.

The update cell state 254 updates the values output from the forge gate layer 252 and the input gate layer 253 to the cell states C _t-1 and C _t . The update of the cell state may be expressed as in Equation 3 below.

Here, C _t represents the cell state and C _{t- 1} represents the previous cell state. f _t is the output of the forget gate layer 252, i _t is the output of the sigmoid operation in the input gate layer 253, and c _t is the output of the hyperbolic tangent operation in the input gate layer 253.

Next, the output gate layer 255 determines the output value h _t . h _t is the filtered value of the cell state (C _t ). The output gate layer 255 may include a sigmoid operation and a hyperbolic tangent operation. Here, the sigmoid operation may calculate cell state determination information o _t for determining a portion of the cell state. The hyperbolic tangent operation performs a hyperbolic tangent operation on the updated cell state C _t and multiplies the cell state determination information (o _t ) output between -1 and 1. Each of the sigmoid operation and the hyperbolic tangent operation performed at the output gate layer 255 may be represented by Equation 4 below.

O _t is the output of the sigmoid operation in the output gate layer 255, and h _t is the output of the hyperbolic tangent operation in the output gate layer 255. Here, W _o is a weight of the sigmoid operation, and b _o is a bias value of the sigmoid operation. W _o and b _o are the values set by the learning of the RNN.

6 is a diagram for describing a process of outputting a probability result for each pixel, according to an exemplary embodiment.

As shown in FIG. 6, the probability classifier 140 may classify the probabilities 280 for the foreground and the background with respect to the temporal information 260. The result of can be obtained. To this end, the probability classifier 140 may classify the probabilities using the FCNN 270.

Since the probability classifier 140 applies the same parameters used for each pixel in the RNN 250 in the FCNN 270, the use of the parameter for separating the foreground and the background may be minimized. For example, when the FCNN 270 is applied to all pixels, the parameter that is proportional to the resolution is required, but the probability classifier 140 uses the same parameters used in the FCNN 270 for each pixel in the RNN 250. The same parameters can be used regardless of the resolution.

As described above, since the probability classification unit 140 uses the RNN 250 that outputs a continuous result with respect to the continuous input, in addition to the visual information of the video, the probability classification unit 140 interprets the temporal information represented as a continuous arrangement of frames. can do.

7 is a flowchart illustrating an image processing method, according to an exemplary embodiment.

As illustrated in FIG. 7, the image processing apparatus 100 may extract visual information that may be configured as a feature map from an input frame (S310). The image processing apparatus 100 may extract visual information in the form of a feature map, and may extract a feature map by using a convolution filter on an input frame. Here, the feature map may be defined as a CNN feature by using a convolution filter implemented using the CNN.

The image processing apparatus 100 may obtain a feature patch by dividing the feature map by pixel units (S320). In this case, when the feature map includes a plurality of layers, the image processing apparatus 100 may divide the feature map into feature patches including all visual information about each of the plurality of layers.

The image processing apparatus 100 may extract temporal information for each feature patch using the RNN (S330). The image processing apparatus 100 may include an external memory (for example, h _t-1 , h _t , h _{t + 1,} etc.) and an internal memory (for example, FIG. 6). Feature vector including information stored in the external memory and the internal memory, while updating the memory in response to the temporal flow using the cell states C _t-1 , C _t , C _{t + 1,} etc.). Can be extracted. As described above, the image processing apparatus 100 may include a feature vector (for example, weights W _f , W _i , W _c , and W _o of FIG. 6) and a bias b _f , b _i , b _c , and b as temporal information. _o )) can be extracted.

The image processing apparatus 100 does not rely only on visual information by using the RNN, but may learn a neural network that may utilize temporal information. Therefore, the image processing apparatus 100 may detect motion information about the moving object. In addition, the image processing apparatus 100 may store a past value obtained from the image frame by using the RNN and predict the next value.

The image processing apparatus 100 may extract a probability result for each pixel from the temporal information (S340). The image processing apparatus 100 may extract the probability result from the temporal information using the FCNN.

The image processing apparatus 100 may detect a movement of an object in the input image by using the extracted temporal information (S350).

The image processing apparatus 100 may separate and terminate the foreground and the background through motion detection in the image (S360).

Meanwhile, in the operation of the image processing apparatus 100 illustrated in FIG. 7, images may be separated using CNN, RNN, and FCNN sequentially, and the process illustrated in FIG. 7 may be used to separate the foreground and the background. It can be used for learning operation, and can also be used for real image analysis operation for separating the real foreground and background. Here, the image processing apparatus 100 may learn the parameters in the CNN, RNN, and FCNN during the learning operation or the actual image analysis, and may update the parameters by using the learned parameters.

The image processing apparatus 100 according to the present exemplary embodiment may perform background difference (for example, motion detection), which is one of automatic analysis techniques for a video. In this way, the image processing apparatus 100 may detect the moving object in the video and reduce the area for analysis to the area in which the moving object exists, thereby improving the image analysis speed. In addition, the image processing apparatus 100 may obtain information on the presence, direction, and type of motion in the video. Therefore, the image processing apparatus 100 may be used in a surveillance field using an image such as security and security, and may be applied to various other fields in which functions such as video synthesis, object recognition, image search, etc. may be utilized.

The term '~ part' used in the present embodiment refers to software or a hardware component such as a field programmable gate array (FPGA) or an ASIC, and the '~ part' performs certain roles. However, '~' is not meant to be limited to software or hardware. '~ Portion' may be configured to be in an addressable storage medium or may be configured to play one or more processors. Thus, as an example, '~' means components such as software components, object-oriented software components, class components, and task components, and processes, functions, properties, procedures, and the like. Subroutines, segments of program patent code, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays, and variables.

The functionality provided within the components and 'parts' may be combined into a smaller number of components and 'parts' or separated from additional components and 'parts'.

In addition, the components and '~' may be implemented to play one or more CPUs in the device or secure multimedia card.

In addition, the image processing method according to an embodiment of the present invention may be implemented as a computer program (or computer program product) including instructions executable by a computer. A computer program includes programmable machine instructions processed by a processor and may be implemented in a high-level programming language, object-oriented programming language, assembly language, or machine language. In particular, it can be implemented in the R language, the Python language, the Ruby language, and the Scheme language. Computer programs may also be recorded on tangible computer readable media (eg, memory, hard disks, magnetic / optical media or solid-state drives, etc.).

Accordingly, the image processing method according to an embodiment of the present invention may be implemented by executing the computer program as described above by the computing device. The computing device may include at least a portion of a processor, a memory, a storage device, a high speed interface connected to the memory and a high speed expansion port, and a low speed interface connected to the low speed bus and the storage device. Each of these components are connected to each other using a variety of buses and may be mounted on a common motherboard or otherwise mounted in a suitable manner.

Here, the processor may process instructions within the computing device, such as to display graphical information for providing a graphical user interface (GUI) on an external input, output device, such as a display connected to a high speed interface. Instructions stored in memory or storage. In other embodiments, multiple processors and / or multiple buses may be used with appropriately multiple memories and memory types. The processor may also be implemented as a chipset made up of chips comprising a plurality of independent analog and / or digital processors.

The memory also stores information within the computing device. In one example, the memory may consist of a volatile memory unit or a collection thereof. As another example, the memory may consist of a nonvolatile memory unit or a collection thereof. The memory may also be other forms of computer readable media, such as, for example, magnetic or optical disks.

The storage device can provide a large storage space to the computing device. The storage device may be a computer readable medium or a configuration including such a medium, and may include, for example, devices or other configurations within a storage area network (SAN), and may include a floppy disk device, a hard disk device, an optical disk device, Or a tape device, flash memory, or similar other semiconductor memory device or device array.

The above description of the present invention is intended for illustration, and it will be understood by those skilled in the art that the present invention may be easily modified in other specific forms without changing the technical spirit or essential features of the present invention. will be. Therefore, it should be understood that the embodiments described above are exemplary in all respects and not restrictive. For example, each component described as a single type may be implemented in a distributed manner, and similarly, components described as distributed may be implemented in a combined form.

The scope of the present invention is shown by the following claims rather than the above description, and all changes or modifications derived from the meaning and scope of the claims and their equivalents should be construed as being included in the scope of the present invention. do.

100: image processing apparatus 110: feature extraction unit
120: patch division unit 130: temporal information processing unit
140: probability classification unit

Claims (14)

A feature extractor which extracts visual information by applying a convolution filter to the input frame;
A patch dividing unit dividing the feature map including the visual information by pixel unit to obtain a feature patch;
A temporal information processor extracting temporal information about the feature patch;
A probability classification unit for extracting one probability result for each pixel from the temporal information; And
A background separator for separating the foreground and the background of the input frame based on the extracted probability result;
The temporal information processing unit extracts the temporal information using a cyclic neural network, and the probability classification unit extracts the probability result from the temporal information using a whole neural network, wherein the parameters used for each pixel in the cyclic neural network are all connected neural networks. The same applies to the image processing apparatus. The method of claim 1,
The patch divider,
And when the feature map includes a plurality of layers, the feature patch includes all pixel information of each of the plurality of layers. delete The method of claim 1,
The temporal information processing unit,
And an information including information on the flow of visual information over time, and extracting the temporal information which is a prediction value for separating the foreground and the background. The method of claim 1,
The convolution filter,
An image processing apparatus, which is a filter implemented using a convolutional neural network. delete The method of claim 1,
The background separator,
The image processing device detects the movement of the object within the input frame and separates the foreground and the background through the motion detection. In the image processing method performed by the image processing apparatus,
Extracting visual information by applying a convolution filter to the input frame;
Dividing the feature map including the visual information by pixel unit to obtain a feature patch;
Extracting temporal information about the feature patch;
Extracting one probability result for each pixel from the temporal information; And
Separating the foreground and the background of the input frame based on the extracted probability result;
Extracting the temporal information,
Extracting the temporal information using a circulatory neural network,
Extracting the probability result,
And extracting the probability result from the temporal information using the whole neural network, wherein the parameters used for each pixel in the cyclic neural network are equally applied to the whole neural network. The method of claim 8,
Acquiring the feature patch,
And when the feature map includes a plurality of layers, splitting the feature patch to include all pixel information of each of the plurality of layers. delete The method of claim 8,
The temporal information is,
An image processing method comprising information on the flow of visual information according to the flow of time, and is a prediction value for separating the foreground and the background. The method of claim 8,
Extracting the feature,
And extracting a feature using the convolution filter implemented using a convolutional neural network. delete The method of claim 8,
Separating the foreground and the background,
Detecting a movement of an object within the input frame,
And separating the foreground and the background through the motion detection.

Images