KR20220064104A - Device for analyzing explosives media propagation characteristic image, drone therefor and method thereof - Google Patents

Abstract

A device for analyzing an explosive media propagation characteristic image includes: an aerial image acquisition part for acquiring an aerial image obtained by photographing an explosive in the sky; a side image acquisition part for acquiring a side image obtained by photographing the side of the explosive; a medium detection part for distinguishing an explosive medium from background noise; an image fusion part for fusing the upper image and the side image of the explosive medium in three dimensions; and a physical quantity calculation part for calculating a physical quantity for the three-dimensionally fused medium.

Description

DEVICE FOR ANALYZING EXPLOSIVES MEDIA PROPAGATION CHARACTERISTIC IMAGE, DRONE THEREFOR AND METHOD THEREOF

The present invention relates to an apparatus for analyzing the propagation characteristics of an explosive medium, a drone therefor, and a method therefor.

Explosives using fuel disperse the fuel inside the explosive isotropically with the primary detonation, and the fuel mixes with air to form a fuel medium. The fuel medium propagates in all directions, and the secondary detonation is performed when the fuel and air are optimally mixed. Due to the explosive energy generated during the second detonation, detonation occurs in the fuel medium, resulting in high-pressure shock waves and high-temperature heat. Explosives show high effectiveness in the area where the fuel medium is diffused, and facilities and equipment are destroyed due to the high pressure of the long duration characteristic. A number of pressure sensors and high-speed cameras are used around explosives to quantitatively analyze characteristics such as dispersion of fuel medium and shock wave propagation of explosives having these characteristics, but there are the following problems.

In the past, the fuel medium dispersion characteristics in the side high-speed image were analyzed using a commercial image analysis tool. However, the algorithms such as contour and contrast detection and correlation techniques provided by the image analysis tool recognize the background noise around the explosive as fuel and shock wave medium, so the medium area was manually designated and deleted for analysis. The total operation time of the explosive is several hundred ms, and the high-speed camera is set at thousands of frames per second or more to acquire high-speed images of the explosive characteristics. Therefore, it takes a considerable amount of time to manually designate and analyze the areas of fuel and shock wave medium propagating in all directions, and human error may be included.

In addition, it is important to measure the propagation speed of the shock wave generated during detonation among the test evaluation items for explosives. Since the pressure of the shock wave generated during the first detonation is low, it is difficult to identify the shock wave medium in high-speed images. And since the explosion area is wide during the second detonation, the installed pressure sensor can only measure the propagation speed of the shock wave around the explosive, and cannot analyze the propagation speed after the pressure sensor.

And, in order to disperse the fuel inside the explosive, the body of the explosive is notched at uniform intervals, so the dispersion characteristic of the actual fuel medium is sunflower-shaped when viewed from above. However, since the dispersion characteristic of the fuel medium is analyzed in two dimensions only with the side image obtained from the conventional high-speed camera, the dispersion characteristic of the fuel medium cannot be accurately analyzed.

In order to accurately analyze the dispersion characteristics of the fuel medium, a three-dimensional image analysis method using high-speed images acquired from the side and from above is required. In order to acquire high-speed images in the sky, a high-speed camera must be mounted on the drone, and a remote control such as supplying a trigger signal to the high-speed camera mounted on the drone and a method of synchronizing GPS time information and the number of frames per second are additionally required.

The technical problem to be solved by the present invention is to distinguish a medium propagating over time from background noise, to create and learn a deep learning-based medium detection model, and to use high-speed images fused in three dimensions to combine fuel and explosives An object of the present invention is to provide an image analysis apparatus capable of quantitatively analyzing the propagation characteristics of a shock wave medium, a drone therefor, and a method therefor.

A medium propagation characteristic image analysis apparatus according to an embodiment of the present invention includes an aerial image acquisition unit for acquiring an aerial image photographed from above the explosive, a side image acquisition unit for acquiring a side image photographed around the side of the explosive, and the A medium detecting unit that distinguishes the medium of the explosive from background noise, an image fusion unit that fuses the aerial image and the side image of the explosive medium in three dimensions, and a physical quantity calculation for calculating a physical quantity for the three-dimensionally fused medium includes wealth.

It may further include a synchronizer for synchronizing the frames of the aerial image and the side image by supplying a twitch signal to the overhead high-speed camera for photographing the aerial image.

It may further include an image preprocessor for discriminating the medium of the explosive from a background image containing noise.

The image preprocessor may adjust the contrast for the difference between the maximum and minimum values of image brightness and set a weight to the outline of the medium to distinguish the background image containing the noise from the medium of the explosive.

The medium detection unit calculates the amount of change in the image as the image difference using a minimum value among the difference between the current frame and the past frame and the difference between the current frame and the background image, and calculates the binarized image by binarizing the image difference, and the binarized image , a method in which the medium continues to expand may be applied to detect a region of interest, and a region excluding the region of interest may be designated as background noise.

The medium detection unit may distinguish the medium of the explosive from the background noise by using a medium detection model based on a neural network.

The image fusion unit may correct the x-axis value and the y-axis value of the medium in the side image with the x-axis value and the y-axis value obtained from the aerial image.

The physical quantity calculator may calculate a propagation distance, an area, and a volume of a medium propagating over time from the side image in which the x-axis value and the y-axis value are corrected.

A drone for analyzing an image of a medium propagation characteristic of an explosive according to another embodiment of the present invention remotely controls a parameter including a high-speed camera in the sky for capturing an image of the explosive, a resolution of the high-speed camera in the sky, and the number of frames a wireless remote control unit for confirming the location of the drone, a high-speed image transmitter for wirelessly transmitting the output image of the high-speed camera in the sky, and the detonation time of the explosive in order to synchronize the frames of the aerial image and the side image of the explosive It includes a trigger signal supply unit for setting the shooting start time of the high-speed overhead camera, and a GPS time information receiving unit for synchronizing the trigger time for the side high-speed camera and the high-speed camera to take the side image.

A power supply unit for additionally supplying power to the high-speed aerial camera separately from the power supplied to the drone may be further included, and power may be dually supplied to the high-speed aerial camera.

In a method for analyzing an image of an explosive medium propagation characteristic according to another embodiment of the present invention, a side image is taken with a plurality of side high-speed cameras installed around the side of the explosive, and an overhead high-speed camera mounted on a drone positioned above the explosive. acquiring a high-speed image of the medium of the explosive by taking an overhead image with Detecting the medium of the explosive by distinguishing the medium from the background noise, fusing the aerial image and the side image of the explosive medium in three dimensions, and calculating a physical quantity for the three-dimensionally fused medium includes steps.

The method may further include the step of synchronizing the frames of the aerial image and the side image by supplying a zoom signal to the aerial high-speed camera.

The image preprocessing may be performed by adjusting the contrast for the difference between the maximum and minimum values of the image brightness and setting a weight to the outline of the medium to distinguish the background image from the medium of the explosive.

Using the minimum value among the difference between the current frame and the past frame and the difference between the current frame and the background image, the amount of change in the image is calculated as the image difference, the image difference is binarized to calculate the binarized image, and the medium continues in the binarized image. Thus, a region of interest may be detected by applying the expansion method, and a region other than the region of interest may be designated as background noise to detect the medium of the explosive.

By correcting the x-axis value and the y-axis value of the medium in the side image with the x-axis value and the y-axis value obtained from the aerial image, the aerial image and the side image may be 3D fused.

A physical quantity for the three-dimensionally fused medium may be calculated by calculating a propagation distance, an area, and a volume of a medium propagating over time from the side image in which the x-axis value and the y-axis value are corrected.

In order to quantitatively analyze the propagation characteristics of the fuel of explosives and the shock wave medium, the medium propagating over time is distinguished from the background noise, a deep learning-based medium detection model is created and learned, and a three-dimensional fused high-speed image is produced. It can be used to quantitatively analyze physical quantities for medium propagation characteristics.

1 is a block diagram illustrating an apparatus for analyzing an image of an explosive medium propagation characteristic according to an embodiment of the present invention.
2 is an exemplary diagram illustrating a method for measuring medium propagation characteristics of an explosive according to an embodiment of the present invention.
3 is a block diagram showing the configuration of a drone according to an embodiment of the present invention.
4 is a flowchart illustrating a method of analyzing medium propagation characteristics of an explosive according to an embodiment of the present invention.
5 is an exemplary view showing results of image preprocessing and medium detection according to an embodiment of the present invention.
6 is an exemplary diagram for explaining a method of calculating a scale factor for analyzing a medium propagation characteristic in three dimensions according to an embodiment of the present invention.
7 is an exemplary diagram illustrating a result of calculating a physical quantity of a fuel medium according to an embodiment of the present invention.
8 is an exemplary diagram illustrating a result of calculating a physical quantity of a shock wave medium according to an embodiment of the present invention.

Hereinafter, with reference to the accompanying drawings, embodiments of the present invention will be described in detail so that those of ordinary skill in the art to which the present invention pertains can easily implement them. The present invention may be embodied in several different forms and is not limited to the embodiments described herein.

In order to clearly explain the present invention, parts irrelevant to the description are omitted, and the same reference numerals are assigned to the same or similar components throughout the specification.

In addition, throughout the specification, when a part "includes" a certain component, this means that other components may be further included, rather than excluding other components, unless otherwise stated.

Hereinafter, an apparatus for analyzing medium propagation characteristics of explosives according to an embodiment of the present invention will be described with reference to FIGS. 1 to 3 .

1 is a block diagram illustrating an apparatus for analyzing an image of an explosive medium propagation characteristic according to an embodiment of the present invention. 2 is an exemplary diagram illustrating a method for measuring medium propagation characteristics of an explosive according to an embodiment of the present invention. 3 is a block diagram showing the configuration of a drone according to an embodiment of the present invention.

1 to 3 , the image analysis apparatus 10 includes a high-speed image acquisition unit 100 and an image analysis unit 200 . The high-speed image acquisition unit 100 includes an overhead image acquisition unit 110 , a side image acquisition unit 120 , and a synchronization unit 130 . The image analysis unit 200 includes an image preprocessor 210 , a medium detection unit 220 , an image fusion unit 230 , and a physical quantity calculator 240 . The medium detection unit 220 may include a detection model unit 221 .

The high-speed image acquisition unit 100 acquires a side image and an aerial image of the explosive medium in conjunction with a plurality of high-speed cameras.

The aerial image acquisition unit 110 acquires an aerial image captured by the high-speed aerial camera 320 mounted on the drone 300 capable of flying still in the sky of the explosive 400 . The aerial image acquisition unit 110 may be connected to the drone 300 through wireless communication.

The side image acquisition unit 120 acquires a side image captured by a plurality of side high-speed cameras 121 and 122 installed around the side of the explosive 400 . The side image acquisition unit 120 may be connected to the plurality of side high-speed cameras 121 and 122 through wireless communication or wired communication. In FIG. 2 , it is illustrated that two side high-speed cameras 121 and 122 are installed in the x-axis direction and the y-axis direction on the ground. Can be installed in more numbers.

The synchronization unit 130 synchronizes remote control such as supplying a trigger signal to the high-speed aerial camera 320 and the GPS time information and the number of frames per second of the high-speed aerial camera 320 with the side high-speed cameras 121 and 122 . That is, the synchronizer 130 may synchronize the frames of the upper image and the side image.

The image preprocessor 210 separates the background image and the medium containing noise in order to detect the dispersion of the fuel medium, the high-speed image and the shock waves generated during the first and second detonation, which are difficult to identify with a pressure sensor.

The medium detection unit 220 distinguishes the medium that diffuses over time from the background noise in order to analyze the fuel medium and the shock wave propagation characteristics. The detection model unit 221 generates and learns a medium detection model based on deep learning. The medium detection unit 220 may distinguish the medium from the background noise based on the value learned by the detection model unit 221 .

The image fusion unit 230 fuses the side image and the aerial image of the medium that spreads in all directions at the time of detonation in three dimensions. The image fusion unit 230 may apply a chain-code to the fuel medium and the shock wave medium separated by the medium detection unit 220 and calculate in pixel units to fuse the side image and the aerial image in three dimensions.

The physical quantity calculator 240 calculates physical quantities for the fuel medium and the shock wave medium fused in three dimensions. A reference pole 125 is installed around the explosive 400 in order to quantitatively analyze the physical quantity for the propagation characteristics of the fuel and the shock wave medium. The propagation distance, area, and volume of a medium propagating over time are calculated from the lateral image with the x and y axes corrected, and the medium is isotropic around the z axis, so the volume can be calculated by integrating the lateral image. there is. The physical quantity calculator 240 may generate the calculated physical quantity as a graph of the physical quantity with respect to the time axis.

A plurality of side high-speed cameras 121 and 122 are installed on the side of the explosive 400 to analyze the propagation characteristics of the fuel and the shock wave medium in three dimensions when the explosive 400 is detonated, and the explosive 400 flies stationary in the sky. A high-speed aerial camera 320 is mounted on the drone 300 . When the explosive 400 is detonated, the side high-speed cameras 121 and 122 and the high-speed aerial camera 320 acquire a side image and an aerial image. A plurality of side high-speed cameras 121 and 122 installed on the side of the explosive 400 are installed at an interval of 90 degrees in the x-axis direction and the y-axis direction to obtain high-speed images of fuel and shock wave medium propagating isotropically. can be obtained In order to protect the drone 300 from debris and shock waves generated when the explosive 400 is detonated, the operating altitude of the drone 300 must be at least 150 m or higher.

The drone 300 may include a power supply unit 310, an aerial high-speed camera 320, a wireless remote control unit 330, a high-speed image transmission unit 340, a trigger signal supply unit 350, and a GPS time information receiving unit 360. there is.

The power supply unit 310 may be additionally mounted to additionally supply power to the high-speed camera 320 in the sky separately from power supplied to drive the drone 300 . The power supply unit 310 together with the power of the drone 300 itself prevents the high-speed images of several gigabytes or more stored in the memory of the high-speed aerial camera 320 from being deleted from the memory due to a power cut. Power can be supplied dually. In addition, the power supply unit 310 supplies DC power for operating the wireless remote control unit 330 , the high-speed image transmission unit 340 , the trigger signal supply unit 350 , and the GPS time information reception unit 360 .

The wireless remote control unit 330 provides a function that allows the operator to remotely control parameter settings such as the resolution of the high-speed camera 320 in the sky, the number of frames, and the exposure value of the lens. To this end, the wireless remote control unit 330 may be configured as a switching hub and a transceiver of an Industrial Scientific Medical (ISM) band.

The high-speed image transmitter 340 checks the x-axis and y-axis positions of the drone 300 that is still flying in the sky of the explosive 400 , and outputs a High Definition Serial Digital Interface (HD-SDI) of the high-speed camera 320 . It provides the function of wirelessly transmitting the image to the operator.

The trigger signal supply unit 350 remotely supplies a TTL level external trigger signal to the high-speed aerial camera 320 in order to set the detonation time of the explosive 400 as the zero time of the high-speed camera 320, that is, the shooting start time. provides the function to Since the high-speed aerial camera 320 acquires high-speed images at thousands of frames per second, it is necessary to set the zero time of the high-speed aerial camera 320 because the storage time of the aerial images is limited due to the capacity limitation of the memory.

The GPS time information receiving unit 360 includes a plurality of side high-speed cameras 121 and 122 so that the number of frames per second of the plurality of side high-speed cameras 121 and 122 and the high-speed aerial camera 320 can be set to be the same or an integer multiple. The trigger point for the high-speed camera 320 is synchronized. In order to three-dimensionally fuse a plurality of two-dimensional images obtained from a plurality of side high-speed cameras 121 and 122 and an overhead high-speed camera 320, a plurality of side high-speed cameras 121 and 122 and an overhead high-speed camera 320 Frames per second must be the same or an integer multiple.

The drone 300 is equipped with various devices including a high-speed camera 320, a power supply unit 310, etc. in the sky to move up to the sky of the explosive 400 spaced apart from a distance and to fly still, so a payload of 20 kg or more ), it must be operated as a medium-sized class or higher that can fly for more than 20 minutes.

Algorithms such as contour and contrast detection and correlation techniques provided by existing commercial image analysis tools are used for background noise around the test site, i.e., test structures, trees, bushes, shadows on the ground when fuel is dispersed, and weather-dependent backgrounds. Background noise was recognized as fuel and shock wave media due to changes in the color of the fuel medium according to the brightness and fuel composition, and shaking of surrounding objects due to the shock wave generated during detonation. and human error was included.

In addition, since the pressure of the shock wave generated during the first detonation of the explosive was low, it was difficult to identify the shock wave medium in high-speed images, and during the second detonation, the propagation speed after the installed pressure sensor could not be analyzed because the explosion area was wide.

Hereinafter, in order to detect the dispersion of the fuel medium, high-speed images and shock waves generated during the first and second detonation, which are difficult to identify with a pressure sensor, a deep learning-based An optimized image analysis method will be described. An image analysis method for analyzing medium propagation characteristics of an explosive using the image analysis apparatus 10 will be described with reference to FIGS. 4 to 8 .

4 is a flowchart illustrating a method of analyzing medium propagation characteristics of an explosive according to an embodiment of the present invention. 5 is an exemplary view showing results of image preprocessing and medium detection according to an embodiment of the present invention. 6 is an exemplary diagram for explaining a method of calculating a scale factor for analyzing a medium propagation characteristic in three dimensions according to an embodiment of the present invention. 7 is an exemplary diagram illustrating a result of calculating a physical quantity of a fuel medium according to an embodiment of the present invention. 8 is an exemplary diagram illustrating a result of calculating a physical quantity of a shock wave medium according to an embodiment of the present invention.

4 to 8, a plurality of side high-speed cameras 121 and 122 and a reference pole 125 are installed around the side of the explosive 400, and the drone 300 equipped with the high-speed aerial camera 320 is installed. After preparing to analyze the medium propagation characteristics of the explosive 400 by positioning it in the air above the explosive 400, the explosive 400 is detonated (S110).

The side high-speed cameras 121 and 122 take side images, and the high-speed aerial camera 320 shoots an aerial image, thereby high-speed for the medium of the explosive 400 in the first to third directions (x, y, z). An image is acquired (S120). The plurality of side high-speed cameras 121 and 122 and the high-speed aerial camera 320 may synchronize the shooting time in synchronization with the trigger signal. The number of frames per second of the plurality of side high-speed cameras 121 and 122 and the high-speed aerial camera 320 may be set to be the same or an integer multiple.

The image preprocessor 210 of the image analysis apparatus 10 performs image preprocessing on the obtained high-speed image (S130). To make the boundary between the fuel and shock wave medium and the background image containing noise clear, the contrast is adjusted for the difference between the maximum and minimum values of the image brightness, and the image is improved by performing gamma correction using a nonlinear transfer function. do. In addition, by using a medium filter, a low-pass filter, and a high-pass filter, a weight is set on the outline of the medium to preserve important image information such as the outline of the medium, and a value greater than the weight is determined and corrected as an error. Then, after removing the irregular noise, the contour features are weighted. Additionally, the brightness change of the image including the medium and background noise is uniformly improved by improving the contrast of the image after applying histogram equalization.

The medium detection unit 220 of the image analysis apparatus 10 generates and learns a medium detection model ( S140 ), and detects a medium using the medium detection model ( S150 ).

The medium detection unit 220 calculates the amount of change in the image in order to distinguish the fuel and the shock wave medium propagating with time from the background noise. Since it is difficult to distinguish between the medium and the background image in the existing contour operators such as Sobel and Gaussian, in the embodiment of the present invention, the amount of change in the image is calculated using the image difference function as in Equation 1 by using the characteristic that the propagation characteristic of an explosive is time-continuous. .

을 적용한다. 즉, 영상차를 현재 프레임과 과거 프레임과의 차, 현재 프레임과 배경 영상과의 차 중에서 최솟값을 사용하여 영상의 변화량으로 계산한다. Here, D denotes an image difference, and I denotes an input image. B is a background image. Initially, the first image is used as a background, and each frame is added to the background image.

apply That is, the image difference is calculated as the amount of change in the image by using the minimum value among the difference between the current frame and the past frame and the difference between the current frame and the background image.

The medium detection unit 220 performs binarization to distinguish the fuel medium and the background image from the calculated image difference. In the partial average model, the fuel medium disappears depending on the topography or the shape of the facilities included in the background. Accordingly, the operator applies the method of determining the explosion area by setting an appropriate threshold value according to the test environment. In addition, the following process is applied to calculate the binarized image and remove the noise region.

The noise of the explosion region is removed, the erosion operation is applied after the dilation operation to the binarized image to combine the same region, and the outer region is smoothed by combining the separated regions.

In addition, using the characteristic that the medium continuously diffuses, if a pixel classified as a fuel medium in the binarized image is near the fuel medium of the previous frame, it is determined as the fuel medium, and continues to expand from the calculation result of the previous frame. to detect the region of interest. In addition, when the fuel and shock wave medium are detected, the region excluding the region of interest is designated as background noise. Accordingly, image preprocessing and detection model generation and learning speed can be improved. Additionally, the outline of the area is extracted by applying a chain-code, and errors such as irregularly increasing areas or shadows are removed using the characteristics of the outline. Finally, the outline determined by the chain code is defined as the explosion area, all areas within the chain code are determined as the fuel medium, and then the space is filled to determine the final fuel medium. In order to use these results as training data for deep learning, the original image and the resulting image are stored in the deep learning learning folder to learn the medium detection model.

The medium detection model includes a neural network-based medium detection algorithm for distinguishing the propagating fuel and shock wave medium from background noise over time. In the existing generative adversarial network (GAN), spatial data loss occurs during downsampling, but the proposed medium detection algorithm applies a U-Net structure to the generator and then makes a skip connection to the result of upsampling having the same size. In addition, the loss of spatial data is compensated for. In addition, the medium detection algorithm divides each part of the image into a medium and a background image by applying the discriminator of the PatchGAN structure. The model used for generating and learning the medium detection model for the propagation characteristics of explosive medium is the Pix2Pix structure used for style transfer. Since the proposed medium detection model is to distinguish the regions of the fuel and shock wave medium and the background image, the output is binarized to a size of (256, 256, 1), and the generator includes 7 downsampling layers and 7 upsampling layers. It consists of a total of 14 layers, and includes 6 skip connection layers to minimize spatial data loss during downsampling and upsampling. The discriminator receives an input image and an output image, and outputs a result of concatenating the two data as (30, 30, 1) through four downsampling layers, and divides the output image by 30×30 to each part shows the results of detection and discrimination of the medium. The objective function of the discriminator is left as it is, and the objective function of the generator uses the method of adding the L1 norm by affecting the L1 norm by an arbitrary λ. print out The proposed medium detection model is not a method of learning the generator once after learning the discriminator once, but learning the discriminator n times randomly before learning the generator, like the method used in the WGAN model, and then forming a strong discriminator through the formation of a strong discriminator. It leads to improved constructor results. Accordingly, the medium can be optimally detected from the background image, and the effort and time for generating and learning the medium detection model can be reduced.

To learn the medium detection model, an image is input as RGB of size (256, 256, 3), and an image of size (256, 256, 1) that is binarized by the generator is output. The parameters of the optimization function of the generator and discriminator apply a learning rate of 4e ^-5 , Beta_1 0.5 , and Epsilon 1e ^-6 , and λ, a parameter that supports the generator to output a result close to the ground truth in the objective function, is 1000. is set The total learning of the medium detection model uses about 2000 image pairs (original, mask) generated through image preprocessing, and performs 250 iterations for each epoch, resulting in a loss after 100 epochs. Save the model for this lowest case. In addition, even if the outline of the medium is set incorrectly after learning, the medium area can be manually specified and deleted, so that analysis can be performed more accurately and the medium detection accuracy, precision, and reproducibility are improved. As illustrated in FIG. 5 , the medium may be detected separately from the background noise over time.

의 면적을 이용한 보정식을 사용한다. 또한, 매질은 등방성으로 분산하므로 측면 고속 영상의 매질의 x축, y축에 대한 스케일 인자(scale factor) κ를

으로 계산하여 측면 영상에서 각 z축에 해당하는 x축과 y축의 반지름에 대해 보정한다. The image fusion unit 230 fuses the side image and the upper image of the medium in three dimensions (S160). In order to isotropically disperse the fuel inside the explosive 400, the body of the cylindrical explosive 400 is notched at uniform intervals. As illustrated in FIG. 6 , since the medium diffuses in all directions during detonation, the propagation characteristics of the medium are quantitatively analyzed by fusing the high-speed images acquired from the side and the sky in three dimensions. Since the dispersion characteristic of the actual fuel medium is sunflower-shaped, the x-axis and y-axis values of the medium in the side image are corrected with the x-axis and y-axis values obtained from the aerial image. In order to calculate the average radius of the high-speed image medium in the _sky , r

Use the correction equation using the area of . In addition, since the medium is isotropically dispersed, the scale factor κ for the x-axis and y-axis of the medium of the side high-speed image is calculated.

, and correct the radii of the x-axis and y-axis corresponding to each z-axis in the side image.

The physical quantity calculating unit 240 calculates a physical quantity for the propagation characteristics of the fuel and the shock wave medium (S170). A reference pole 125 is installed around the explosive 400 for quantitative analysis of the physical quantity for the propagation characteristics of the fuel and the shock wave medium. As illustrated in FIGS. 7 and 8 , the propagation distance, area, and volume of a medium that propagates over time is calculated from the side image in which the x-axis and y-axis values are corrected. Since the volume of the medium is isotropic around the z-axis, it is calculated by integrating the lateral image. And the calculated physical quantity can be output in the form of graphs and documents on the time axis.

As described above, it is possible to create an algorithm for performing pre-processing on a high-speed image of an explosive medium, and generating and learning a medium detection model. In addition, data containing various background noises are collected from high-speed images including propagation characteristics of fuel and shock wave medium, image processing is performed to determine the medium area, and learning data for 25 or more cases with different environments is produced. After that, the medium detection model can be trained. After a total of 100 epoch learning, it is possible to select three images among various images used for deep learning learning and perform verification, and the medium detection performance can be continuously improved through re-learning.

In order to quantitatively analyze the propagation characteristics of the fuel and the shock wave medium of the explosive 400, the medium propagating over time is distinguished from the background noise, a deep learning-based medium detection model is created and learned, and three-dimensional fusion is performed. A physical quantity for the propagation characteristics of a medium can be quantitatively analyzed using a high-speed image.

The drawings and the detailed description of the described invention referenced so far are merely exemplary of the present invention, which are only used for the purpose of explaining the present invention, and are used to limit the meaning or the scope of the present invention described in the claims. it is not Therefore, those skilled in the art will understand that various modifications and equivalent other embodiments are possible therefrom. Accordingly, the true technical protection scope of the present invention should be defined by the technical spirit of the appended claims.

10: image analysis device 100: high-speed image acquisition unit
110: aerial image acquisition unit 120: side image acquisition unit
121, 122: side high-speed camera 125: reference pole
130: synchronization unit 200: image analysis unit
210: image preprocessor 220: medium detection unit
221: detection model unit 230: image fusion unit
240: physical quantity calculator 300: drone
310: power supply 320: high-speed camera in the sky
330: wireless remote control unit 340: high-speed video transmitter
350: trigger signal supply unit 360: GPS time information receiving unit
400: explosives

Claims (16)

An aerial image acquisition unit for acquiring an aerial image taken from the sky of the explosive;
a side image acquisition unit for acquiring a side image taken around the side of the explosive;
a medium detection unit for discriminating the medium of the explosive from background noise;
an image fusion unit that fuses the aerial image and the side image of the explosive medium in three dimensions; and
A medium propagation characteristic image analysis apparatus comprising a physical quantity calculator for calculating a physical quantity for a three-dimensionally fused medium. According to claim 1,
The medium propagation characteristic image analysis apparatus further comprising a synchronizer for synchronizing the frames of the aerial image and the side image by supplying a zoom signal to the overhead high-speed camera that captures the aerial image. According to claim 1,
The medium propagation characteristic image analysis apparatus further comprising an image pre-processing unit for discriminating the medium of the explosive from a background image containing noise. 4. The method of claim 3,
The image preprocessor adjusts the contrast for the difference between the maximum and minimum values of the image brightness, and sets a weight to the outline of the medium to distinguish the background image containing the noise from the medium of the explosive medium. According to claim 1,
The medium detection unit calculates the amount of change in the image as the image difference using a minimum value among the difference between the current frame and the past frame and the difference between the current frame and the background image, and calculates the binarized image by binarizing the image difference, and the binarized image A medium propagation characteristic image analysis apparatus for detecting a region of interest by applying a method in which the medium continues to expand in the , and designating a region excluding the region of interest as background noise. According to claim 1,
The medium detector distinguishes the medium of the explosive from the background noise by using a medium detection model based on a neural network. According to claim 1,
The image fusion unit corrects the x-axis value and the y-axis value of the medium in the side image with the x-axis value and the y-axis value obtained from the overhead image. 8. The method of claim 7,
The physical quantity calculator calculates a propagation distance, an area, and a volume of a medium propagating over time from the side image in which the x-axis value and the y-axis value are corrected. A drone for image analysis of medium propagation characteristics of explosives,
an overhead high-speed camera that takes an aerial image of the explosive;
a wireless remote control unit for remotely controlling parameters including resolution and number of frames of the high-speed camera;
a high-speed image transmission unit that confirms the position of the drone and wirelessly transmits the output image of the high-speed camera in the sky;
a trigger signal supply unit configured to set the detonation time of the explosive to the start time of shooting of the high-speed camera in order to synchronize the frames of the aerial image and the side image of the explosive; and
A drone for analyzing a medium propagation characteristic image of an explosive, comprising a side high-speed camera for photographing the side image and a GPS time information receiver for synchronizing a trigger time for the high-speed camera in the sky. 10. The method of claim 9,
A drone for analyzing a medium propagation characteristic image of an explosive, further comprising a power supply unit for additionally supplying power to the high-speed camera in the sky separately from the power supplied to the drone, wherein power is dually supplied to the high-speed camera. Obtaining a high-speed image of the medium of the explosive by photographing a side image with a plurality of side high-speed cameras installed around the side of the explosive, and capturing an aerial image with an overhead high-speed camera mounted on a drone positioned above the explosive ;
performing image preprocessing to distinguish a boundary between the medium of the explosive and a background image including noise in the high-speed image;
detecting the medium of the explosive by distinguishing the medium of the explosive from background noise;
fusing the aerial image and the side image of the explosive medium in three dimensions; and
An image analysis method of medium propagation characteristics of explosives, comprising the step of calculating a physical quantity for a three-dimensionally fused medium. 12. The method of claim 11,
The medium propagation characteristic image analysis method of explosives further comprising the step of synchronizing the frames of the aerial image and the side image by supplying a trigger signal to the overhead high-speed camera. 12. The method of claim 11,
A method of analyzing the medium propagation characteristics of explosives by adjusting the contrast for the difference between the maximum and minimum values of the image brightness and setting a weight on the outline of the medium to distinguish the background image from the medium of the explosive and perform the image preprocessing. 12. The method of claim 11,
Using the minimum value among the difference between the current frame and the past frame and the difference between the current frame and the background image, the amount of change in the image is calculated as the image difference, the image difference is binarized to calculate the binarized image, and the medium continues in the binarized image. An image analysis method for medium propagation characteristics of explosives to detect a region of interest by applying the expansion method, and to detect the medium of the explosive by designating a region other than the region of interest as background noise. 12. The method of claim 11,
The medium propagation characteristic image of an explosive that three-dimensionally fuses the aerial image and the side image by correcting the x-axis value and the y-axis value of the medium in the side image with the x-axis value and the y-axis value obtained from the aerial image analysis method. 16. The method of claim 15,
Medium propagation of an explosive that calculates a physical quantity for the three-dimensionally fused medium by calculating the propagation distance, area, and volume of the medium propagating over time from the side image in which the x-axis value and the y-axis value are corrected Characteristic image analysis method.

Images