KR20230094015A - Control system of digital continuous zoom image and control method thereof - Google Patents

Abstract

A digital continuous zoom image control system according to an embodiment of the present invention includes a first image capturing unit for capturing a wide angle image, a second image capturing unit for capturing a standard angle image, and a third image capturing unit for capturing a telephoto image. Interface units connected to each other; a control unit connected to the interface unit and having a deep learning neural network therein; a memory unit connected to the control unit; and an output unit connected to the control unit, wherein the control unit overlaps the wide angle video image, the standard angle video image, and the telescopic angle video image, and the standard angle formed along the edge of the standard angle video image. The video image detected while moving the plurality of detection ends (A1, A2, B1, B2) along the boundary line and the boundary line of the telescopic angle formed along the edge of the video image of the telescopic angle is input to the deep learning neural network to perform deep learning. It is characterized by matching while learning.

Description

Digital continuous zoom image control system and digital continuous zoom image control method

The present invention relates to a digital continuous zoom image control system and a method for controlling a digital continuous zoom image, and in particular, digital continuous zoom images are generated by matching digital images of various angles of view, including a telephoto image, a wide angle image, and a standard angle image. It relates to a zoom image control system and a digital continuous zoom image control method.

A super resolution (hereinafter referred to as SR) technique is a technique of generating a low-resolution image or continuous image into a high-resolution image or continuous image. A high-resolution image can provide a high pixel density and thus provide more detailed information about the original scene. In general, super-resolution (SR) techniques are techniques that use a series of related images to enhance content within a given image. Briefly, super-resolution (SR) techniques are techniques for enhancing the representation of an image by using subsequent and/or previous images related to the one image.

In many cases, super-resolution (SR) techniques can provide significant improvements in image quality, where the image sensor is relatively fixed to the object, and the object is substantially stationary, i.e. there is no motion in the landscape of the image being captured. It can produce remarkable image quality at times.

These super-resolution (SR) techniques include a single image-based method and a super-resolution reconstruction method using multiple images.

First, in the case of a method using a single image, there is a self-example-based method using similar data repeatedly appearing inside the image and a learning-based resolution restoration method using external data. In particular, the self-example-based method has a disadvantage in that restoration performance is limited because the image is reconstructed using only data inside the image, whereas the learning-based method using external data has performance that depends on the amount and characteristics of external data and the learning process. There are disadvantages that greatly depend on it.

In the case of the method using multiple images, it is difficult to acquire multiple images at the same viewpoint, and image restoration performance depends on matching performance between images.

Patent Document: Publication No. 10-2016-0142760

The present invention has been made to solve the above problems, and an object of the present invention is to control a digital continuous zoom image to generate a digital continuous zoom image by matching digital images of various angles of view, including a telephoto image, a wide angle image, and a standard angle image. to provide the system.

Another object of the present invention is to provide a digital continuous-zoom image control method for generating a digital continuous-zoom image by matching digital images of various angles of view, including a telephoto image, a wide-angle image, and a standard-angle image.

A digital continuous zoom image control system according to an embodiment of the present invention includes a first image capturing unit for capturing a wide angle image, a second image capturing unit for capturing a standard angle image, and a third image capturing unit for capturing a telephoto image. Interface units connected to each other; a control unit connected to the interface unit and having a deep learning neural network therein; a memory unit connected to the control unit; and an output unit connected to the control unit, wherein the control unit overlaps the wide angle video image, the standard angle video image, and the telescopic angle video image, and the standard angle formed along the edge of the standard angle video image. The video image detected while moving the plurality of detection ends (A1, A2, B1, B2) along the boundary line and the boundary line of the telescopic angle formed along the edge of the video image of the telescopic angle is input to the deep learning neural network to perform deep learning. It is characterized by matching while learning.

In the digital continuous zoom image control system according to an embodiment of the present invention, the interface unit transmits video images acquired by the first imaging unit, the second imaging unit, and the third imaging unit, respectively, to a plurality of pixels. It is characterized in that it is converted into a digital image and transmitted to the control unit.

In the digital continuous zoom image control system according to an embodiment of the present invention, the deep learning neural network of the control unit is a convolutional neural network (CNN) having a plurality of hidden layers between an input layer and an output layer. ) characterized in that it comprises a neural network.

Alternatively, in a method for controlling a digital continuous zoom image according to another embodiment of the present invention, (a) the control unit captures a wide-angle video image and a standard image through the first, second, and third imaging units according to a user's command. Obtaining a video image of each angle and a video image of a telescopic angle; (b) performing, by the controller, correction on each of the wide-angle video image, the standard-angle video image, and the telescopic-angle video image; and (c) overlapping, by the controller, the corrected wide-angle video image, the standard-angle video image, and the telescopic-angle video image, and matching them while learning through deep learning using a CNN neural network.

In the digital continuous zoom image control method according to another embodiment of the present invention, in the step (a), the interface unit converts the video images obtained from each of the first imaging unit, the second imaging unit, and the third imaging unit to a plurality of pixels. It is characterized in that it is converted into a digital image made of (pixels) and transmitted to the control unit.

In the digital continuous zoom image control method according to another embodiment of the present invention, the step (c) is performed by the controller in (c-1) each of the wide-angle video image, the standard-angle video image, and the telescopic-angle video image. overlapping the standard angle image on the wide angle image and overlapping the telescopic angle image on the standard angle image; (c-2) The control unit has a plurality of detection stages A1, A2, B1, detecting a video image while moving B2); and (c-3) inputting video images detected while moving the plurality of detection stages (A1, A2, B1, B2) by the control unit to the CNN neural network and matching them while learning through deep learning. characterized by

In the digital continuous zoom image control method according to another embodiment of the present invention, the plurality of detection stages (A1, A2, B1, B2) are first moving along the inner side of the boundary line of the telescopic angle in the video image of the telescopic angle. A detection end (A1), a second detection end (A2) moving along the outer side of the telescopic angle boundary line in the video image of the standard angle, moving along the inner side of the standard angle boundary line in the video image of the standard angle It is characterized in that it includes a third detection end (B1) and a fourth detection end (B2) moving along the outside of the standard angle boundary line in the wide-angle video image.

Features and advantages of the present invention will become more apparent from the following detailed description taken in conjunction with the accompanying drawings.

Prior to this, the terms or words used in this specification and claims should not be interpreted in a conventional, dictionary sense, and the inventor properly defines the concept of the term in order to explain his/her invention in the best way. Based on the principle that it can be done, it should be interpreted as meaning and concept consistent with the technical spirit of the present invention.

The digital continuous zoom image control system according to an embodiment of the present invention learns three video images obtained from each of the first imaging unit, the second imaging unit, and the third imaging unit using a deep learning neural network, and obtains an image of a standard angle. There is an effect of generating a digital continuous zoom image by matching based on .

A method for controlling a digital continuous zoom image according to another embodiment of the present invention learns a wide-angle video image, a standard-angle video image, and a telephoto-angle video image using a deep learning neural network, and uses a wide-angle video image as a standard. There is an effect of generating a digital continuous zoom image by matching the video image and the video image of the telescopic angle.

1 is a block diagram of a digital continuous zoom image control system according to an embodiment of the present invention.
2 is a flowchart illustrating a digital continuous zoom image control method according to another embodiment of the present invention.
3 is a structural diagram of a CNN neural network applied to a digital continuous zoom image control method according to another embodiment of the present invention.
4 is an exemplary diagram for explaining a process of convolution processing of video images in a digital continuous zoom image control method according to another embodiment of the present invention;
5 is an exemplary diagram for explaining a video image pooling process in a digital continuous zoom image control method according to another embodiment of the present invention;
6 is an exemplary view of a wide-angle image obtained in a digital continuous zoom image control method according to another embodiment of the present invention.
7 is an exemplary view of a standard angle image obtained in a digital continuous zoom image control method according to another embodiment of the present invention.
8 is an exemplary view of a telescopic angle image obtained in a digital continuous zoom image control method according to another embodiment of the present invention.
9 is an exemplary diagram for explaining a process of matching three video images in a digital continuous zoom image control method according to another embodiment of the present invention.

Objects, specific advantages and novel features of the present invention will become more apparent from the following detailed description and preferred embodiments taken in conjunction with the accompanying drawings. In adding reference numerals to components of each drawing in this specification, it should be noted that the same components have the same numbers as much as possible, even if they are displayed on different drawings. Also, terms such as first and second may be used to describe various components, but the components should not be limited by the terms. These terms are only used for the purpose of distinguishing one component from another. In addition, in describing the present invention, if it is determined that a detailed description of related known technologies may unnecessarily obscure the subject matter of the present invention, the detailed description will be omitted.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. 1 is a block diagram of a digital continuous zoom image control system according to an embodiment of the present invention.

A digital continuous zoom image control system according to an embodiment of the present invention includes an interface unit 140 connected to a first imaging unit 110, a second imaging unit 120, and a third imaging unit 130, and an interface unit 140. ) and includes a control unit 150 having a deep learning neural network therein, a memory unit 160 and an output unit 170 respectively connected to the control unit 150.

The first imaging unit 110, the second imaging unit 120, and the third imaging unit 130 each use a camera including a lens and an imaging sensor and having different focal lengths, The first image capturing unit 110 may capture a wide angle image, the second image capturing unit 120 may capture a standard angle image, and the third image capturing unit 130 may capture a telephoto image.

The angles of view and focal lengths of the lenses constituting each of the first imaging unit 110, the second imaging unit 120, and the third imaging unit 130 are as described in [Table 1] below.

angle of view focal length Lens of the first imaging unit 110 84° to 63° 15mm - 35mm Lens of the second imaging unit 120 47° 50mm Lens of the third imaging unit 130 28° to 8° 85mm to 300mm

Of course, the angle of view and focal length of the lenses constituting each of the first imaging unit 110, the second imaging unit 120, and the third imaging unit 130 may be provided in a configuration other than [Table 1].

Through the first imaging unit 110, it is possible to obtain a wide-angle image, that is, an image capturing a wide range although the size of the subject is reduced and the perspective is exaggerated, and a standard angle image can be obtained through the second imaging unit 120. It is possible to obtain an image, that is, an image of a subject having a size most similar to that seen by the human eye, and through the third image pickup unit 130, an image of a telescopic angle, that is, a very narrow range image obtained by enlarging a close-up of a distant subject can be obtained

The interface unit 140 digitizes video images obtained from each of the first imaging unit 110, the second imaging unit 120, and the third imaging unit 130, and transmits them to the controller 150. Accordingly, video images acquired by each of the first imaging unit 110, the second imaging unit 120, and the third imaging unit 130 are converted into digital images composed of a plurality of pixels and sent to the control unit 150. can be conveyed

The controller 150 is a component that controls the overall operation of the digital continuous zoom image control system, and has a deep learning neural network therein to learn and match three image images input through the interface 140. . Here, the deep learning neural network may include, for example, a Convolutional Neural Network (CNN) neural network, and of course, other deep learning neural networks may be applied in addition to the CNN neural network.

At this time, the CNN neural network has a structure including a plurality of hidden layers (302-1,302-2,302-3) between an input layer (301) and an output layer (303), as shown in FIG. may be provided.

The memory unit 160 is connected to the controller 150, and includes three video images input through the interface 140, a program for image correction, and information necessary for the controller 150 to learn and match the three video images. Save the

The output unit 170 includes a monitor, a speaker, etc. connected to the control unit 150, and the control unit 150 learns and matches the three video images to display the resulting image or processes the three video images with sound. can be output as

The digital continuous zoom image control system according to an embodiment of the present invention configured as described above includes three video images obtained from each of the first imaging unit 110, the second imaging unit 120, and the third imaging unit 130. A digital continuous zoom image can be generated by learning using a deep learning neural network and matching images of standard angles as a criterion.

Hereinafter, a digital continuous zoom image control method according to another embodiment of the present invention will be described with reference to FIGS. 2 to 9 . 2 is a flowchart illustrating a digital continuous zoom image control method according to another embodiment of the present invention, and FIG. 3 is a structure diagram of a CNN neural network applied to a digital continuous zoom image control method according to another embodiment of the present invention. 4 is an exemplary diagram for explaining a process of convolution processing of video images in a digital continuous zoom image control method according to another embodiment of the present invention, and FIG. 5 is a digital continuous zoom image control method according to another embodiment of the present invention. 6 is an exemplary view for explaining a video image pooling process, and FIG. 6 is an exemplary view of a wide-angle image obtained in a method for controlling a digital continuous zoom image according to another embodiment of the present invention, and FIG. 8 is an exemplary view of a telescopic angle image obtained in a digital continuous zoom image control method according to another embodiment of the present invention, and FIG. It is an exemplary diagram for explaining a process of matching three video images in a digital continuous zoom image control method according to another embodiment of the present invention.

In a digital continuous zoom image control method according to another embodiment of the present invention, the control unit 150 controls each of the first imaging unit 110, the second imaging unit 120, and the third imaging unit 130 according to a user's command. Through this, a video image of a wide angle, a video image of a standard angle, and a video image of a telescopic angle are acquired (S210).

That is, as shown in FIG. 6 through the first imaging unit 110, the size of the subject is reduced and the perspective is exaggerated, but a wide-angle video image capturing a wide range is acquired, and the second imaging unit 120 As shown in FIG. 7, a video image of a standard angle is obtained by capturing a subject having a size most similar to that seen by the human eye, and as shown in FIG. 8 through the third imaging unit 130, A close-up of the subject is obtained and an enlarged telescopic image is acquired.

Each of these three video images is converted into a digital video image composed of a plurality of pixels in the interface unit 140 and transmitted to the controller 150.

At this time, the controller 150 performs correction on each of the three received digital video images (S220).

That is, as shown in FIG. 6, the wide-angle video image acquired through the first imaging unit 110 causes image distortion such as exaggeration of perspective and barrel distortion, and as shown in FIG. In the telescopic image obtained through the third imaging unit 130, the sense of perspective is reduced and image distortion such as pincushion distortion may occur.

In order to correct such image distortion, the controller 150 uses various correction filters such as a Gaussian filter stored in the memory unit 160 to correct image distortion occurring in a wide-angle image and a telephoto image. can do.

After the correction process, the control unit 150 overlaps the three digital video images and performs matching while learning through deep learning using a CNN neural network (S230).

Specifically, as shown in FIG. 9 , the controller 150 overlaps the standard-angle video image on the wide-angle video image by matching the center points of each of the wide-angle video image, the standard-angle video image, and the telephoto-angle video image. And, the video image of the telescopic angle is overlaid on the video image of the standard angle.

Accordingly, a video image of a telephoto angle having a relatively high resolution magnified by an optical zoom like a magnification is placed in the center area of a video image of a standard angle, and likewise, a video image of a telephoto angle having a relatively high resolution is placed in the central area of a video image of a wide angle by an optical zoom like a magnification. A video image of a standard angle having a relatively high resolution is arranged.

Accordingly, the three digital video images are overlapped based on the same center, and a plurality of boundary lines 501, 502, and 503 are formed along the edges of each of the three digital video images. That is, a wide angle boundary line 501 is formed along the edge of the wide-angle video image, a standard angle boundary line 502 is formed along the edge of the video image of the standard angle, and a telescopic angle boundary line is formed along the edge of the video image of the telephoto angle. (503).

With respect to these boundary lines 501, 502, and 503, the controller 150 moves the detection stages A1, A2, B1, and B2 having predetermined area sizes along the standard angle boundary line 502 and the telescopic angle boundary line 503. while detecting video images. Here, each of the detection stages A1, A2, B1, and B2 has the same shape, for example, a rectangle or a square, and moves along the normal angle boundary line 502 and the telescopic angle boundary line 503, respectively.

That is, as shown in FIG. 9 , the controller 150 moves the first detection end A1 along the inside of the telescopic angle boundary line 503 in the telescopic angle video image, and the telescopic angle in the standard angle video image. The second detection end A2 is moved along the outside of the boundary line 503.

At the same time, the controller 150 moves the third detection end B1 along the inside of the standard angle boundary line 502 in the standard angle video image, and moves the outside of the standard angle boundary line 502 in the wide angle video image. The fourth detection stage (B2) is moved accordingly.

As video images are detected while moving the detection stages A1, A2, B1, and B2, the controller 150 inputs a plurality of detected video images to the CNN neural network and extracts a feature map.

Specifically, the feature map extraction process uses a deep neural network such as the CNN neural network shown in FIG. It may include inputting an image and executing a convolution operation using the filter shown in FIG. 4 while passing through the first hidden layer 302-1.

At this time, the convolution operation using the filter can be executed using [Equation 1] below.

Here, l means a layer, size ^l means the size of a layer, In means the number of data input to the input layer, and Ia means the number of labels. where O denotes an output convolution layer, w denotes a weight, and b denotes a bias of the feature map.

After extracting the feature maps processed through the convolution operation, the controller 150 sums the extracted feature maps and performs a convolution operation on the summed feature maps to regenerate the feature maps.

At this time, the process of regenerating the feature map may be executed by concatenating the summed feature maps using [Equation 2] below and performing a convolution operation using [Equation 3].

Here, In denotes the number of input data, Ia denotes the number of labels, O denotes an output convolution layer, and w denotes a weight.

In this case, f _c indicates a result value obtained by performing a convolution operation using the concatenated feature map, and may mean a regenerated feature map.

After regenerating the feature map, the controller 150 may calculate the feature map regenerated in the second hidden layer 302-2 by substituting it into an activation function. Here, the activation function may use a sigmoid function or a ReLu function.

Regarding the output value output using the activation function, the controller 150 performs a pooling operation using the obtained output value.

Specifically, the pooling operation is for reducing the dimension size of data, and is an operation that reduces the size of vertical and horizontal spaces in data. Such a pooling operation may use various parameters, for example, an average, a median value, a maximum value, a minimum value, and the like, and here, as shown in FIG. 5 , a maximum value pooling operation using the maximum value is applied. It is possible to extract the maximum value from the limited region of the image using max polling, remove noise from the data, and prevent overfitting in the process of reducing the data.

This maximum value pooling operation can be executed using [Equation 4] below.

Here, x means the matrix input for the pooling operation, l means the corresponding layer of the pooling operation, i means the row of the input matrix, j means the column of the input matrix, and size ^l denotes the size of a corresponding layer of the pooling operation, Im denotes the number of data input to the corresponding layer of the pooling operation, and Ia denotes the number of labels.

After performing the pooling operation, the control unit 150 calculates a loss value using the pooling operation value and a preset target output value in the third hidden layer 302-3.

Specifically, the loss value calculation can be performed using MSLE of [Equation 5], RMSLE of [Equation 6], or sMAPE of [Equation 7], and the preset target output value is GT (Ground Truth ) can be.

In this case, GT may be, for example, a value obtained by performing a maximum pooling operation based on a convolution operation value obtained by performing a convolution operation on image data in the first hidden layer 302-1.

는 풀링 연산 단계의 풀링 연산값을 의미하고,

는 미리 설정한 목표 출력값을 의미한다. here,

Means the pooling operation value of the pooling operation step,

denotes a preset target output value.

After calculating the loss value, the control unit 150 obtains a correction value for the parameter using the calculated loss value.

In this case, the parameter may mean a weight of w, and the control unit 150 may update the parameter using the acquired correction value for the parameter.

Using these updated parameters, the control unit 150 may re-perform the steps from regenerating the feature map to obtaining modified values of the parameters a set number of times.

If this re-performation process is completed once, epoch 1 learning is completed, and if the re-performation process is repeated 100 times, epoch 100 learning is completed.

According to the process of re-executing the set number of times, the controller 150 acquires a plurality of clean lines (601, 602, 701, 702) in a clear black-and-white binary image.

That is, the controller 150 obtains the first claw line 601 from the video image of the telescopic angle in FIG. 9 and correspondingly obtains the second claw line 602 from the standard angle video image. The first clone line 601 and the second clone line 602 are obtained by using characteristics having the same gradient and the same brightness contrast, and a plurality of them are obtained based on the telescopic angle boundary line 503, but the telescopic angle boundary line Obtain at least two pairs on each side of 503.

Similarly, the controller 150 obtains the third claw line 701 from the standard-angle video image, and correspondingly obtains the fourth claw line 702 from the wide-angle video image. The third and fourth clew lines 701 and 702 are obtained using characteristics having the same gradient and the same brightness contrast, and a plurality of them are obtained based on the standard angle boundary line 503, but the standard angle boundary line Obtain at least two pairs on each side of 503.

Accordingly, the first claw line 601 and the second claw line 602 are interdigitated and matched, and the third claw line 701 and the fourth claw line 702 are interdigitated and matched. At this time, the video image of the standard angle The telescopic image or the wide-angle image is reduced or enlarged based on , so that a plurality of clew lines correspond to each other and are accurately matched.

To this end, the controller 150 performs an erosion operation or a dilation operation on the telephoto or wide-angle video image to reduce or enlarge it.

로 표기한다. 즉, B에 의한 A의 침식은 Z에 의해 천이된 B가 A 내에 포함되는 모든 점 Z의 집합이다. Specifically, the erosion operation is a morphologically-based operation and is performed according to [Equation 8] below. where A and B are the sets within that image pixel (Z ² ), and the erosion of A by B is

marked with That is, the erosion of A by B is the set of all points Z where B, transitioned by Z, is included in A.

When such an erosion operation is performed, the size of the video image may be reduced while the subject becomes smaller.

로 표기한다. 즉, B에 의한 A의 팽창은 B와 A가 적어도 한 개 요소에 의해 겹쳐지도록 모두 Z만큼 전치된 집합이다. On the other hand, the expansion operation is a morphologically-based operation and is performed according to [Equation 9] below. where A and B are the sets within that image pixel (Z ² ), and the dilatation of A by B is

marked with That is, the dilation of A by B is the set in which both B and A are transposed by Z such that they overlap by at least one element.

If this expansion operation is repeatedly performed a number of times, the size of the video image may be enlarged.

By performing an erosion operation or a dilation operation, a telescopic angle image or a wide angle image image that has been reduced or enlarged is interlocked with a standard angle image image, so that a plurality of clone lines can be accurately matched in correspondence with each other.

Accordingly, a method for controlling a digital continuous zoom image according to another embodiment of the present invention learns a wide-angle video image, a standard-angle video image, and a telescopic-angle video image using a deep learning neural network, and based on the standard-angle video image A digital continuous zoom image may be generated by matching a wide-angle video image and a telephoto video image.

Although the technical idea of the present invention has been specifically described according to the above preferred embodiments, it should be noted that the above-described embodiments are for explanation and not for limitation.

In addition, those skilled in the art will understand that various implementations are possible within the scope of the technical spirit of the present invention.

110: first imaging unit 120: second imaging unit
130: third imaging unit 140: interface unit
150: control unit 160: memory unit
170: output unit 501: wide-angle boundary line
502 standard angle boundary line 503 telescopic angle boundary line
601,602,603,604: Clue line A1,A2,B1,B2: Detection stage

Claims (7)

an interface unit connected to a first image capturing unit for capturing a wide angle image, a second image capturing unit for capturing a standard angle image, and a third image capturing unit for capturing a telescopic image;
a control unit connected to the interface unit and having a deep learning neural network therein;
a memory unit connected to the control unit; and
an output unit connected to the control unit;
including,
The controller overlaps the wide-angle video image, the standard-angle video image, and the telephoto-angle video image, and along the standard angle boundary line formed along the standard angle video image edge and the telephoto angle video image edge. Digital continuous zoom image, characterized in that the video image detected while moving the plurality of detection stages (A1, A2, B1, B2) along the formed telescopic angle boundary line is input to the deep learning neural network and matched while learning through deep learning control system.
According to claim 1,
The interface unit converts video images obtained from each of the first imaging unit, the second imaging unit, and the third imaging unit into a digital image composed of a plurality of pixels and transmits the digital image to the control unit. Continuous zoom video control system.
According to claim 1,
The deep learning neural network of the control unit comprises a convolutional neural network (CNN) neural network having a plurality of hidden layers between an input layer and an output layer. Digital continuous zoom image control system .
(a) acquiring, by a controller, a wide-angle image, a standard-angle image, and a telescopic-angle image through the first, second, and third image pickup units, respectively, according to a user's command;
(b) performing, by the controller, correction on each of the wide-angle video image, the standard-angle video image, and the telescopic-angle video image; and
(c) overlapping, by the control unit, the corrected wide-angle video image, the standard-angle video image, and the telescopic-angle video image, and matching them while learning through deep learning using a CNN neural network;
Digital continuous zoom image control method comprising a.
According to claim 4,
In the step (a), the interface unit converts the video image obtained from each of the first imaging unit, the second imaging unit, and the third imaging unit into a digital image composed of a plurality of pixels and transmits it to the control unit. Digital continuous zoom image control method, characterized in that.
According to claim 4,
The step (c) is
(c-1) the control unit aligns the center of each of the wide-angle video image, the standard-angle video image, and the telescopic-angle video image, and overlaps the standard-angle video image on the wide-angle video image; superimposing the video image of the telescopic angle on the video image of the standard angle;
(c-2) The control unit has a plurality of detection stages A1, A2, B1, detecting a video image while moving B2); and
(c-3) inputting video images detected while moving the plurality of detection terminals (A1, A2, B1, B2) to the CNN neural network and matching them while learning through deep learning;
Digital continuous zoom image control method characterized in that it further comprises.
According to claim 6,
The plurality of detection stages A1, A2, B1, and B2 include a first detection stage A1 moving along the inner side of the boundary line of the telescopic angle in the video image of the telephoto angle, and the telephoto angle in the video image of the standard angle. A second detection end (A2) moving along the outside of each boundary line, a third detection end (B1) moving along the inside of the standard angle boundary line in the standard angle video image, and the wide angle image A digital continuous zoom image control method comprising a fourth detection end (B2) moving along the outside of the standard angle boundary line.

Images