KR20230019744A - Video Finger Printing Learning Method Using Virtual world and system using same - Google Patents

Abstract

An objective of the present invention is to build a video using a virtual camera (like moving a viewpoint during a 3D game) within a 3D virtual space (like using a real environment as a game background). The present invention relates to a video fingerprinting learning method using virtual world, comprising: a virtual space (500); a virtual camera (20) which acquires an image including a plurality of frames while virtually moving in the virtual space (500); a first feature point extraction unit (110A) which generates a grouped fingerprint for each frame of the image acquired by the virtual camera (20); an image index DB (130) which indexes a shooting angle (AA) and shooting path (FA) of the image acquired by the virtual camera (20) into the image; an image storage system (160) in which fingerprints, location information, the shooting angle (AA), and the shooting path (FA) grouped for each frame of the video are indexed and stored; a moving means which captures a video including a plurality of frames; a second feature point extraction unit (110B) which generates fingerprints grouped for each frame of the video captured by the moving means; and a video analysis unit (120) which compares the video, from which the feature points are extracted, of the moving means with the video of the image storage system.

Description

Video Finger Printing Learning Method Using Virtual World and system using same}

The present invention relates to a video fingerprinting learning method using a virtual world and a system using the same.

In particular, the present invention obtains a virtual video image having a plurality of frames by locating and moving a virtual drone in a virtual world, obtains a fingerprint from each of the plurality of video images, or divides each image to obtain a video image in a divided area. A method of obtaining image data by storing a fingerprint or a system using the same.

In order to operate a means of transportation such as a car or a drone, the location of the means of transportation is identified, and the driving direction and speed of the means of transportation are controlled based on the identified location.

Although GPS is used to determine the location of such means of transportation, GPS cannot be used in places such as inside buildings, under bridges, and tunnels.

Therefore, the current position of the means of transportation is identified by comparing the image captured by the photographing means attached to the means of transportation (the black box of the vehicle, the camera of the drone) with the previously captured image, and An invention for controlling the moving means is disclosed in Registered Patent No. 10-2096784-0000.

In addition, sub-video fingerprints for each frame image of a motion picture taken in advance of the surveillance target area or fingerprints (singular points) in the divided areas of the frame image are extracted, and after extraction, the divided areas are compared to determine the surveillance target area or the surveillance target area. Inventions aimed at enabling easy detection of changes in some areas are disclosed in Registered Patent Nos. 10-1734029-0000 and 10-2090739-0000.

However, in the above patents, in order to determine the similarity of the flight video of the designated route or the similarity of the designated route, videos of the same route are required in advance, and a lot of human, material, and time costs are required for actual shooting.

Publication Patent No. 10-2015-0033043 (2015.04.01) Registered Patent No. 10-1240924-0000 (2013.03.04) Registered Patent No. 10-2090739-0000 (2020.03.12) Registered Patent No. 10-2096784-0000 (2020.03.30)

In the present invention, it is intended to build a video using a virtual camera (like moving a viewpoint in a 3D game) in a 3D virtual space (like using a real environment as a game background).

In addition, in the present invention, in consideration of path flight error, shadow error over time, error due to camera shooting angle, etc., it is intended to generate cheaper video path data by changing the setting of a virtual environment for various videos.

In addition, it is intended to construct videos of various paths rather than a specified path error, and to understand the path and angle of the video taken by the moving camera.

The present invention includes a virtual space 500; a virtual camera 20 that obtains an image including a plurality of frames while moving virtually in the virtual space 500; a first feature point extraction unit 110A generating grouped fingerprints for each frame of the image acquired by the virtual camera 20; an image index DB (130) for indexing the photographing angle (AA) and photographing path (FA) of the image acquired by the virtual camera 20 into the image; an image storage system 160 indexing and storing grouped fingerprints, location information, capturing angles (A _A ) and capturing paths (F _A ) of each frame of the video; a moving means for capturing a moving image including a plurality of frames; a second feature point extraction unit 110B for generating grouped fingerprints for each frame of the video captured by the moving means; and an image analyzer 120 that compares the video of the means of transportation from which the characteristic points are extracted with the video of the video storage system.

In addition, the image analysis unit compares feature points extracted as fingerprints grouped for each frame image of the image of the moving means and indexed feature points of images previously stored in the image storage system to correspond to a plurality of frames in a predetermined section A pre-stored image having a similarity of feature points equal to or greater than a threshold value is found, and the location information, the shooting angle (A _A ) and the shooting path (F _A ) of the pre-stored image are converted into the location information of the moving means and the shooting angle (A _A ). And it can be determined by the photographing path (F _A ).

In addition, in the present invention, the first and second feature point extraction units 100A and 100B decode a video stream of a moving image into an RGB format or a black and white video format, and then divide each decoded frame image into a plurality of blocks, After calculating the average value of the image brightness data of each pixel for each divided block unit, a feature point may be extracted based on a difference in the average value of the image brightness data between adjacent blocks to generate a grouped fingerprint for each frame.

In addition, the storage system 160 may include a plurality of pre-stored images 200, 210, 220, and 230 including fingerprints and location information for each frame, and the moving unit stores the captured image 300. The second feature point extraction unit 110B generates grouped fingerprints for each frame, and the second feature point extraction unit 110B transmits the image 300 in which the fingerprint is generated to the image analysis unit 120 in real time. , The image analyzer 120 prints fingerprints (P ₃₀₁ , P 302 , P 303 ) by frames (301, 302, 303, 304, 305, 306, ₃₀₇ , ₃₀₈ ) of the image 300 captured by the moving unit. , P ₃₀₄ , P ₃₀₅ , P ₃₀₆ , P ₃₀₇ , P ₃₀₈ ) and the images 200 and 210 previously stored in the storage system 160 for each frame (211, 212, 213, 214, 215, 216, 217, 218 ) by comparing fingerprints (P ₂₁₁ , P ₂₁₂ , P ₂₁₃ , P ₂₁₄ , P ₂₁₅ , P ₂₁₆ , P ₂₁₇ , P ₂₁₈ ) to find a frame with similarity equal to or higher than the threshold, and find the location information of the previously stored image, the shooting angle (A _A ) and the photographing path F _A may be determined based on the location information of the moving means, the photographing angle _A , and the photographing path F _A .

In addition, the present invention includes the steps of acquiring an image including a plurality of frames while the virtual camera 20 moves virtually in the virtual space 500; generating grouped fingerprints for each frame of the image acquired by the virtual camera 20 in a first feature point extraction unit 110A; indexing, by the image index DB 130, information on the photographing angle (AA) and the photographing path (FA) of the image obtained by the virtual camera 20 and the location of each frame; storing, by an image storage system 160, the moving picture indexed with the fingerprint, location information, photographing angle (A _A ), and photographing path (F _A ) grouped for each frame; photographing a moving image including a plurality of frames by a moving unit; The moving means 400 transmits the video taken by the moving means to the second feature point extraction unit 110B, and the second feature point extraction unit 110B provides a grouped fingerprint for each frame of the video taken by the moving means. generating; and comparing the video of the moving means from which the feature points are extracted by the second feature point extractor (110B) with the video of the image storage system in the image analyzer 120. It is about a printing learning method.

In the present invention, compared to real-time filming by constructing a video using a virtual camera (like moving a viewpoint in a 3D game) in a 3D virtual space (like using a real environment as a game background), time and cost can be reduced. made it possible

In addition, in the present invention, in consideration of path flight error, error due to shadow over time, error due to camera shooting angle, etc., various videos are generated through changing the setting of the virtual environment to obtain inexpensive image path data and compare with the actual video. This was done to ensure the accuracy of the data.

In addition, it is possible to obtain path data by combining the fingerprint and index information of the path of the camera moving in the virtual space.

1 is an example of image capturing in the prior art.
2 is a configuration diagram of the present invention.
3 is a configuration diagram of photographing through the virtual moving unit 20 in a virtual space.
4a to 4d are route photos for camera shooting in various routes.
5 is an example of dividing an image frame to extract a fingerprint.
6A to 6D show a state in which images taken in the paths of FIGS. 4A to 4D are fingerprinted and stored in the image storage system 160 .
7 is an embodiment for tracking location information.
8 is an example of fingerprinting a video taken in a real space.
FIG. 9 is fingerprinted image data in a state previously stored in the image storage system 160 corresponding to the image of FIG. 8 .
10 is an example in which a moving photographing image frame 400 according to the present invention is divided into sub-frames.
11 is an example of dividing a subframe [S1(1,1)) into a plurality of blocks to extract a singular point.
12 is an example of extracting a singularity from each subframe of the frame 400 .
13a and 13b are comparative examples of singularities of the present invention.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. First, in adding reference numerals to components of each drawing, it should be noted that the same components have the same numerals as much as possible even if they are displayed on different drawings. In describing the present invention, if it is determined that a detailed description of a related known configuration or function may obscure the gist of the present invention, the detailed description will be omitted.

FIG. 1 is an exemplary view of image capture in the prior art. Since data of various paths is not sufficient, stored data moves to a path F, but when comparing actual data, an error may occur when moving to a path F'.

Therefore, in the present invention, various virtual images are acquired while moving the virtual photographing unit 20 in the virtual space as shown in FIG. 3 .

In the virtual space, images are obtained from various paths as shown in FIGS. 4A to 4D .

2 is a block configuration diagram showing a location tracking system using similarity analysis of mobile captured images according to the present invention. It includes a second feature point extraction unit 110B, an image analysis unit 120 and an image index DB 130.

The first image capturing unit 100A obtains an image including a plurality of frames in various paths while moving the virtual capturing unit 20 in the virtual space 500, and the first feature point extracting unit 110A A fingerprint, which is a feature point of an image including a plurality of frames acquired in a virtual space, is extracted.

In the second video capture unit 200B, a moving means (e.g., a drone, a car) actually moves and captures a video including a plurality of various frames through a camera mounted thereon, and the actually captured video has second feature points. It is transmitted to the extractor 110B, and the second feature point extractor 110B, like the first feature point extractor 110A, analyzes the video data included in the video file and extracts a video fingerprint, which is a video feature point.

Here, the video fingerprint refers to a kind of video unique ID element capable of uniquely identifying video data, and the first feature point extractor 110A obtains the video fingerprint from the virtual space 500 prior to extracting it. A process of selectively separating only video data from one image file may be preceded.

The first feature point extraction unit 110A selectively separates a video stream from a video file acquired in a virtual space to extract a video fingerprint, and decodes the separated video stream into an RGB format.

Subsequently, sub video feature points are extracted for each frame image constituting the decoded video data.

The process of extracting sub video feature points for each frame image by the first feature point extractor 110A is as follows. This refers to the paper by Job Oostveen, Ton Kalker, and Jaap Haitsma, “Feature Extraction and a Database Strategy for Video Fingerprinting” (Lecture Notes in Computer Science 2314 Springer 2002).

First, the first feature point extraction unit 110A divides the frame image into blocks as shown in FIG. 5 .

For example, the frame image is divided into n×m blocks by dividing the vertical axis of the image into n equal parts and dividing the horizontal axis into m equal parts.

Then, the first feature point extraction unit 110A calculates the average of Y values, which is each image brightness data, for each divided block unit, and then generates a sub-video fingerprint of a predetermined number of bits based on the difference in average image brightness values between adjacent blocks. .

[Equation 1] illustrates a bit allocation function 'B(r, c, p)' that can be used to generate a sub video fingerprint for one frame image.

Here, B(r, c, p) is a function for allocating bits to blocks located at the rth and cth positions on the basis of the horizontal axis and the vertical axis of the pth frame image, respectively. Further, F(r, c, p) denotes average image brightness values of blocks located at the rth and cth positions, respectively, on the basis of the horizontal axis and the vertical axis of the pth frame image.

In F(r, c, p) of [Equation 1], r has an integer value from 1 to n, and c has an integer value from 1 to m. Here, a is a constant that can be selected appropriately as a number less than 1.

Meanwhile, the first feature point extraction unit 110A may generate a sub video fingerprint by utilizing Cb or Cr values, which are color difference data of each pixel, in addition to Y values, which are brightness data of the image.

After the first feature point extraction unit 110A extracts the sub-video fingerprints of each frame image constituting the video, it completes the generation of video fingerprints for the video file by grouping the extracted sub-video fingerprints.

In the present invention, any algorithm capable of extracting a video fingerprint other than the method described in the paper by Jop Ustbin and one other person can be applied.

In this way, the video finger frit for the video file is generated in the first feature point extractor 110A, and the shooting angle _A and the shooting path F _A of the video file are formed in the image index DB 130 After that, the video file is stored in the video storage system 160. Accordingly, the image information obtained in the virtual space stores information about the fingerprint, the shooting angle, the shooting path, and the location information for each frame.

When a new image is captured by the second image capturing unit 100B including a capturing means mounted on a vehicle or a drone, the image of the captured image is transmitted to the second feature point extracting unit 110B for fingerprint printing is generated and the generated fingerprint is transmitted to the image analysis unit 120, and compared with the fingerprint of the video file of the storage system.

The video fingerprint generated in the newly captured image is compared with the fingerprint stored in the storage system, and it is determined whether the similarity between the fingerprints is greater than or equal to a threshold value. 160) is determined to match the location where the image was acquired.

In this way, by comparing the image captured by the photographing means of the moving means with the previous image of the storage system in which the location information is recorded, the position of the image photographing means can be grasped in real time to determine the location of the moving means. It can control vehicles or drones.

5 is an exemplary embodiment of segmenting blocks during the process of extracting feature points of an intelligent mobile monitoring system using similarity analysis of mobile captured images according to the present invention. In Fig. 5, it is divided into 9 blocks.

The first and second feature point extraction units 110A and 110B may use a color histogram graph to extract color-based feature points. This can use the contents of Wei-Lun Chao's thesis "The Core of Video Fingerprinting Example of Feature Extraction".

In addition, the first and second feature point extractors 110A and 110B may use an object recognition based feature point extraction algorithm to extract feature points based on object movement. It is preferable to use Huitao Luo's thesis (Algorithms for Video Object Detection and Segmentation with Application to Content-Based Multimedia Systems).

Meanwhile, the second feature point extractor 110B transmits the unique values of each frame from which the sub-video feature points are extracted to the image analyzer 120. The image analyzer 120 may search from a specific point in time according to the transmitted parameters or factors.

The image index DB 130 includes a photographing angle (Aa, Ab, Ac, Ad) and a photographing path obtained by obtaining an image of an image having a fingerprint from the first image capturing unit 100A and the first feature point extracting unit 110A. It is good to index (Fa, Fb, Fc, Fd) and location information.

Preferably, the image analysis unit 120 performs a function of searching the image storage system 160 by limiting a set path. It is possible to increase the search speed by limiting the search section to a search section set by the manager without searching all the data pre-stored in the image storage system 160 .

As shown in FIG. 7, the image captured on September 10 through the mobile means 400 such as a drone and the finger of the pre-stored image obtained by the virtual photographing means 20 in the virtual space 500 An image with a print value of more than a threshold value is found, and the location of the means of transportation to be photographed on September 10 can be identified through the location information of the pre-stored image.

The target image (September 10th image) currently being captured in real time is not the image of the base image itself, but the feature point (finger print) extracted by the first feature point extraction unit 110A for each frame of the image is mutually transmitted through bit operation. Comparison is fast because of comparison. Therefore, real-time comparison is also possible, and this comparison is performed by the image analysis unit 120.

In FIG. 7 , as a result of mutual comparison of feature points from point A to point D of the image, the similarity is greater than or equal to a critical value, and the location of the photographing device taken on September 10 can be identified through the location information of the previously stored image.

As shown in Figures 6a to 6d, the image storage system includes a plurality of frames (201, 202, ...; 211, 212 ...; 221, 222, ...; 231, 232,,,) The images (200, 210, 220, 230) are stored, and the fingerprint, shooting angle, and shooting path (P _211, A _A, F _A ;P ₂₁₂ , A _A, F _A ; _P213 , A _A, F _A ; P ₂₁₄ , A _{A ,} F _A ; P ₂₁₅ , A _{A ,} F _A ; P ₂₁₆ , A _{A ,} F _A ; P ₂₁₇ , A _{A ,} F _A ; P _218, A _A, F _A ;) are created.

As shown in FIG. 8 , the moving unit captures an image 300 in real time, and the captured image 300 is captured by the second feature point extractor 110B for each frame (301, 302, 303, 304, 305, 306, 307, 308) to generate fingerprints (P ₃₀₁ , P ₃₀₂ , P ₃₀₃ , P ₃₀₄ , P ₃₀₅ , P ₃₀₆ , P ₃₀₇ , P ₃₀₈ ), and the image analysis unit 120 prints the fingers of the captured image 300 Search for matching frames among pre-stored images using print.

For example, the fingerprints (P ₂₁₃ , P 214, P ₂₁₅ , P ₂₁₆ , P 217) of the frames 213, 214, 215, ₂₁₆ , and ₂₁₇ of the pre-stored image 210 of FIG. 9 are the image 300. When matching the fingerprints (P ₃₀₂ , P ₃₀₃ , P ₃₀₄ , P 305, P ₃₀₆ , P ₃₀₇ ) of the frames 302, 303, 304, _305, and 306 by more than a threshold value, capturing an image 300 The location of the moving means is determined to be the same as the location of the frames 213 , 214 , 215 , 216 , and 217 of the image 210 , the photographing angle A _B , and the path information F _B .

In addition, when the similarity between the fingerprint values (P ₃₀₇ , P 218 ) of the frame 307 of the image 300 and the frame ₂₁₈ of the image 210 is less than or equal to a threshold value (for example, 90% or less), the image The analysis unit finds again an image whose similarity to the frame-by-frame fingerprint of the image 300 captured by the moving means is greater than or equal to a threshold value.

As described above, in the present invention, an image is acquired in the virtual space 500, feature points are extracted from the acquired image by frame-by-frame fingerprinting, and after storing the image including the shooting angle and the shooting path of the virtual space image in a storage system, movement Compared with feature points of images taken by means, the location and movement path of the current moving means can be grasped through location information of matching frames of videos in the storage system.

In addition, in the present invention, each frame 400 of an image acquired in a virtual space is divided into a plurality of sub-regions [SI(1,1), SI(1,2), SI(1,3) as shown in FIG. ...], and each of the divided sub-regions can be divided into a plurality of block units (Fig. 11 is divided into 9 blocks) as shown in FIG.

For example, each subregion [SI(1,1), S(1,2), S(1,3) ...] subregion image of a frame by dividing the vertical axis of the image into n equal parts and dividing the horizontal axis into m equal parts. can be divided into n × m blocks, and in FIG. 11, the subregion [SI(1,1)] is divided into 3x3 blocks.

The first feature point extraction unit 110A calculates the average of Y values, which is each image brightness data, for each divided block unit, and then generates a sub-video fingerprint of a predetermined number of bits based on the difference in average image brightness values between adjacent blocks.

As shown in FIG. 12, fingerprints such as P(1,1), P(1,2), P(1,3), P(1,4)... are generated for each sub-region, For example, P(1, 1) can be set by digitizing RGB values in units of a plurality of blocks shown in FIG. 11 .

As shown in FIG. 12, the first feature point extraction unit 110A prints fingerprints [P(1,1), P(1,2), P(1,3), P(1,4)...] for each sub-region. And the image index DB 130 indexes the photographing angle (A _A ) and the photographing path (F _A) of the image captured in the virtual space together with the fingerprint information and stores it in the image storage system 160.

13A and 13B are examples of comparing an image acquired in a virtual space and stored in an image storage system with a photographed image.

The five frames 400ba, 400bb, 400bc, 400bd, and 400be of FIG. 13A are images being captured in real time, and are a set of frames captured during a specific section, T1. Assuming that the starting point of photography is the time Ti, each of the sub-regions of the five frames is fingerprinted by the second feature point extraction unit 110B.

In addition, the 8 frames of FIG. 13B are pre-stored images, and the 5 frames (400ba, 400bb, 400bc, 400bd, 400be) having a value equal to or higher than a threshold value and a partial fingerprinted value of the sub-region image ( 400aa, 400ab, 400ac, 400ad, 400ae) and a specific section, _TS, are synchronized.

That is, the fingerprint values P(1,1), P(1, 2), P(1,3), P(1, 4) of each sub-region of 5 frames (400ba, 400bb, 400bc, 400bd, 400be) ) ... If some of them are found to correspond to each other above the threshold with some of the fingerprints of 5 frames (400aa, 400ab, 400ac, 400ad, 400ae) of the previously stored DB, the starting points of similarity comparison are Ti and Ts As a result, feature points of sub-region images synchronized and included in each subsequent frame are mutually compared.

In order to perform synchronization, synchronization is performed when a certain number of sub-regions of each frame and a certain number of fingerprint values of each frame are identical, rather than matching all sub-region fingerprint values of a frame.

[For example, 5 fingerprint values of sub-regions SI(1,1) and SI(2,1) of the first to fifth frames (400ba, 400bb, 400bc, 400bd, 400be) of the captured image are stored. When the difference in the fingerprint values of the sub-regions SI(1,1) and SI(2,1) of the two frames (400aa, 400ab, 400ac, 400ad, 400ae) is less than 10% (when the threshold is 90%) 5 captured frames (400ba, 400bb, 400bc, 400bd, 400be) and 5 previously stored frames (400aa, 400ab, 400ac, 400ad, 400ae) may be synchronized.

After synchronization, the fingerprint values of sub-regions of each frame are compared, and if an error of 10% or more occurs in some parts, it can be seen that an unusual event has occurred.

The present invention described above is not limited by the foregoing embodiments and the accompanying drawings, and it is common in the technical field to which the present invention belongs that various substitutions, additions, and changes are possible without departing from the technical spirit of the present invention. It will be clear to those who have knowledge of

400a, 400b: first and second imaging unit
110A, 110B: first and second feature point extraction units
120: image analysis unit 130: image index DB
160: video storage system

Claims (5)

virtual space 500;
a virtual camera 20 that obtains an image including a plurality of frames while moving virtually in the virtual space 500;
a first feature point extraction unit 110A generating grouped fingerprints for each frame of the image acquired by the virtual camera 20;
an image index DB (130) for indexing the photographing angle (AA) and photographing path (FA) of the image acquired by the virtual camera 20 into the image;
an image storage system 160 indexing and storing grouped fingerprints, location information, capturing angles (A _A ) and capturing paths (F _A ) of each frame of the video;
a moving means for capturing a moving image including a plurality of frames;
a second feature point extraction unit 110B for generating grouped fingerprints for each frame of the video captured by the moving means; and
An image analyzer 120 comparing the video of the moving means from which the feature points are extracted with the video of the video storage system;
The image analysis unit compares feature points extracted as fingerprints grouped for each frame image of the image of the moving means with indexed feature points of images pre-stored in the image storage system to obtain feature points corresponding to a plurality of frames in a predetermined section. Finds a pre-stored image whose similarity is greater than or equal to a threshold value, and converts the location information of the pre-stored image, the shooting angle (A _A ), and the shooting path (F _A ) to the location information of the moving means, the shooting angle (A A ), and the shooting path (F _A ). A system using video fingerprinting using a virtual world, characterized in that the path (F _A ) is determined. According to claim 1,
The first and second feature point extraction units 100A and 100B decode the video stream of the moving image into an RGB format or a black and white video format, divide each decoded frame image into a plurality of block units, and divide the divided block units into blocks. After calculating the average value of the image brightness data of each pixel for each pixel, a feature point is extracted based on the difference in the average value of the image brightness data between adjacent blocks to generate a fingerprint grouped for each frame. Video using virtual world, characterized in that A system using fingerprinting. According to claim 2,
The storage system 160 includes a plurality of pre-stored images 200, 210, 220, 230 including fingerprints and location information for each frame,
The moving means generates grouped fingerprints for each frame in the second feature point extraction unit 110B from the captured image 300, and the second feature point extraction unit 110B generates the image 300 from which the fingerprint is generated. Transmit to the image analysis unit 120 in real time,
The image analysis unit 120 prints fingerprints (P ₃₀₁ , P ₃₀₂ , P ₃₀₃ , P ₃₀₄ , P ₃₀₅ , P ₃₀₆ , P ₃₀₇ , P ₃₀₈ ) and the images 200 and 210 previously stored in the storage system 160 for each frame (211, 212, 213, 214, 215, 216, 217, 218) by comparing the fingerprints (P ₂₁₁ , P ₂₁₂ , P ₂₁₃ , P ₂₁₄ , P ₂₁₅ , P ₂₁₆ , P ₂₁₇ , P ₂₁₈ ) to find a frame with a similarity of more than the threshold, and find the location information of the previously stored image, the shooting angle (A _A ) And the photographing path (F _A ) is determined by the location information of the moving means, the photographing angle (A _A ), and the photographing path (F _A ). A system using video fingerprinting using a virtual world. obtaining an image including a plurality of frames while the virtual camera 20 moves virtually in the virtual space 500;
generating grouped fingerprints for each frame of the image acquired by the virtual camera 20 in a first feature point extraction unit 110A;
indexing, by the image index DB 130, information on the photographing angle (AA) and the photographing path (FA) of the image obtained by the virtual camera 20 and the location of each frame;
storing, by an image storage system 160, the moving picture indexed with the fingerprint, location information, photographing angle (A _A ), and photographing path (F _A ) grouped for each frame;
photographing a moving image including a plurality of frames by a moving unit;
The moving means 400 transmits the video taken by the moving means to the second feature point extraction unit 110B, and the second feature point extraction unit 110B provides a grouped fingerprint for each frame of the video taken by the moving means. generating; and
Comparing the video of the moving means from which the feature points are extracted by the second feature point extractor (110B) with the video of the image storage system in the image analyzer 120;
The image analysis unit compares feature points extracted as fingerprints grouped for each frame image of the image of the moving means with indexed feature points of images pre-stored in the image storage system to obtain feature points corresponding to a plurality of frames in a predetermined section. Finds a pre-stored image whose similarity is greater than or equal to a threshold value, finds a pre-stored image that is greater than or equal to the threshold value, and converts the location information of the pre-stored image, the shooting angle (A _A ) and the shooting path (F _A ) to the location information of the moving means, Video fingerprinting learning method using a virtual world, characterized in that the determination is made by the shooting angle (A _A ) and the shooting path (F _A ). According to claim 4,
The first and second feature point extraction units 100A and 100B decode the video stream of the moving image into an RGB format or a black and white video format, divide each decoded frame image into a plurality of block units, and divide the divided block units into blocks. After calculating the average value of the image brightness data of each pixel for each pixel, a feature point is extracted based on the difference in the average value of the image brightness data between adjacent blocks to create a grouped fingerprint for each frame,
The storage system 160 includes a plurality of pre-stored images 200, 210, 220, 230 including fingerprints and location information for each frame,
The moving means generates grouped fingerprints for each frame in the second feature point extraction unit 110B from the captured image 300, and the second feature point extraction unit 110B generates the image 300 from which the fingerprint is generated. Transmit to the image analysis unit 120 in real time,
The image analysis unit 120 prints fingerprints (P ₃₀₁ , P ₃₀₂ , P ₃₀₃ , P ₃₀₄ , P ₃₀₅ , P ₃₀₆ , P ₃₀₇ , P ₃₀₈ ) and the images 200 and 210 previously stored in the storage system 160 for each frame (211, 212, 213, 214, 215, 216, 217, 218) by comparing the fingerprints (P ₂₁₁ , P ₂₁₂ , P ₂₁₃ , P ₂₁₄ , P ₂₁₅ , P ₂₁₆ , P ₂₁₇ , P ₂₁₈ ) to find a frame with a similarity of more than the threshold, and find the location information of the previously stored image, the shooting angle (A _A ) and the photographing path (F _A ) is determined by the location information of the moving means, the photographing angle (A _A ), and the photographing path (F _A ).

Images