KR102477416B1 - Accident Site Monitoring System - Google Patents

Abstract

The embodiment of the present invention relates to a system for monitoring the scene of an accident. According to one aspect of the present invention, the system for monitoring the scene of an accident comprises: a sound collection unit which collects sound information within a tunnel; an accident analysis unit which determines whether there is an accident in the tunnel through the sound information; and a video collection unit for photographing an accident area in the tunnel based on the information detected by the accident analysis unit. The video collection unit includes a rail and a camera moving along the rail. Accordingly, a response to the accident can be made immediately.

Description

Accident site monitoring system {Accident Site Monitoring System}

This embodiment relates to an accident scene monitoring system.

According to the current road tunnel disaster prevention facility installation and management guidelines (Ministry of Land, Infrastructure and Transport Regulations, No. 308), all road tunnels with a disaster prevention level of 3 or higher are required to install CCTVs (TV equipment for monitoring) and accident detection equipment.

The purpose of installing CCTV (TV equipment for monitoring) in the tunnel is to grasp the situation in the event of a disaster in the tunnel and to monitor the occurrence of an accident.

In addition, the purpose of installing the accident detection equipment in the tunnel is to analyze the received detection data in real time using the CCTV image information or frequency installed in the tunnel and automatically inform the management system when a set accident situation occurs.

The installation interval of CCTVs (monitoring television equipment) in the tunnel is 200m to 400m as a standard, and CCTVs are also installed at points within 500m from the entrance and exit of the tunnel so that the traffic flow at the entrance and exit of the tunnel can also be checked.

However, CCTV is easy to grasp the flow of vehicles on the road, but the shooting distance of CCTV is not long enough to confirm the occurrence of an accident, so it is difficult to clearly identify the location of the accident and the situation where the accident occurred. there is.

In addition, since the CCTV is fixed at the installation point, a situation in which the view for filming may be blocked by topographical factors such as the linear and longitudinal slopes of the tunnel and factors such as obstacles or passing vehicles on the road may occur.

In this case, a situation may arise in which the tunnel manager has to be dispatched directly to the site, and the accident handling is delayed by the amount of time it takes, resulting in a risk of a secondary accident by a follow-up vehicle and a delay in the emergency rescue of the accident victim, resulting in a fatal accident. .

Furthermore, since a plurality of CCTVs (surveillance television equipment) must be installed at regular intervals in the tunnel, installation, maintenance, and repair are costly due to the problem that unnecessary equipment must be installed even in areas where actual accidents do not occur. occurs

The present invention has been proposed to improve the above problems, and relates to an accident site monitoring system capable of promptly grasping a situation and responding to an accident immediately.

The accident site monitoring system according to the present embodiment includes a sound collection unit for collecting sound information in a tunnel; an accident analysis unit that determines whether or not there is an accident in the tunnel through the sound information; And based on the information detected by the accident analysis unit, a video collection unit for photographing an accident area in the tunnel, wherein the video collection unit includes a rail and a camera moving along the rail.

Since the accident occurrence point can be rapidly photographed and monitored by the manager through this embodiment, there is an advantage in that an immediate response can be made in the event of an accident.

In addition, even if a plurality of CCTVs are not provided compared to the prior art, since the entire area in the tunnel can be closely photographed by moving the camera, there is an advantage in that the economic cost required for installation and maintenance can be reduced.

1 is a block diagram of an accident site monitoring system according to an embodiment of the present invention;
Figure 2 is a diagram showing the data flow of the accident site monitoring system according to an embodiment of the present invention.
3 is a diagram showing a process of processing data in a memory according to an embodiment of the present invention;
4 is a diagram showing a process of classifying input data according to an embodiment of the present invention.
5 is a diagram for explaining the configuration of input values according to an embodiment of the present invention;
6 is a structural diagram of a CNN algorithm according to an embodiment of the present invention.
7 is a result table of an accident detection test through a CNN algorithm according to an embodiment of the present invention.
8 is a result table of an accident false detection test through a CNN algorithm according to an embodiment of the present invention.
9 is an experimental table of a thought test through a CNN algorithm according to an embodiment of the present invention.
10 is an exemplary diagram of matrix data in the RNN algorithm according to an embodiment of the present invention.
11 is a schematic diagram of a neural network structure according to an embodiment of the present invention.
12 is a cross-sectional view showing the installation of an image collection unit in a tunnel according to an embodiment of the present invention.
13 is a side view illustrating installation of an image collection unit in a tunnel according to an embodiment of the present invention.

Hereinafter, some embodiments of the present invention will be described through exemplary drawings. In describing the reference numerals for the components of each drawing, the same numerals indicate the same components as much as possible, even if they are displayed on different drawings.

Also, terms such as first, second, A, B, (a), and (b) may be used to describe components of an embodiment of the present invention. These terms are only used to distinguish the component from other components, and the nature, order, or order of the corresponding component is not limited by the term. When an element is described as being 'connected', 'coupled' or 'connected' to another element, the element may be directly connected, coupled or connected to the other element, but not between the element and the other element. It should be understood that another component may be 'connected', 'coupled' or 'connected' between elements.

1 is a block diagram of an accident site monitoring system according to an embodiment of the present invention, and FIG. 2 is a diagram showing a data flow of the accident site monitoring system according to an embodiment of the present invention.

Referring to FIG. 1 , an accident scene monitoring system 100 according to an embodiment of the present invention includes a sound collection unit 110 that collects sound information on a road and a device that determines whether an accident on the road has occurred through the collected sound information. An accident analysis unit 120, an accident processing unit 130 that provides and performs a handling plan when an accident occurs, a receiver 140 activated based on the result detected by the accident analysis unit 120, and an image collection unit ( 150) may be included.

The sound collection unit 110 may collect sound information on the road. The sound collecting unit 110 may include a microphone, an audio codec, a microprocessor, and Ethernet.

The microphone may convert sound heard from the road into an electrical signal. The microphones may be disposed on one side and the other side of the body that forms the outer shape of the accident site monitoring system 100 and faces each other.

The audio codec can transfer the analog signal collected by the microphone to the accident analysis unit 120 and convert it into digital information so that the accident analysis unit 120 can easily analyze it.

Acoustic information collected through the microphone and the audio codec may be stored in a 16-bit, 2-channel, 48000 sample size.

The microprocessor controls peripheral chips, collects data, performs calculations, and receives and stores sound digital data received from the audio codec in a memory. In addition, the microprocessor may perform a CPU role of the sound collecting unit 110 to control Ethernet.

The Ethernet enables communication between the sound collecting unit 110 and the accident analysis unit 120, and between the sound collecting unit 110 and an external server through TCP/IP.

The sound collecting unit 110 may include a function capable of measuring sound pressure through sounds collected in real time.

The accident analysis unit 120 analyzes accidents through processes of collection, processing, analysis, and external linkage interface.

As shown in FIG. 2 , in the collection process, the sound data received from the sound collection unit 110 may be transmitted to the MsgManager 210 . Accordingly, the MsgManager 210 can transmit audio data to each process requiring audio data for data processing.

Each of the above processes may include DBSAVE 220, StreamGW 230, and ADSS 240.

In the DBSAVE 220, sound data information is stored as a database 260 and a sound file every minute.

The StreamGW 230, as its own server, can deliver collected real-time sound information to a manager (Client).

The ADSS 240 analyzes the received acoustic data to detect an accident, transmits the detected data to the GWIF 250 to be described later, and stores the detected data in the database 260.

The GWIF 250 is in charge of control information input from an external server and all interfaces of transactions, and can perform controller setting and data storage functions.

In summary, when the acoustic data received from the sound collection unit 110 is transmitted to each process, each process can perform functions of storage, real-time streaming, and accident detection through the received acoustic data. In addition, the accident-detected data is stored in the database 260 through an external interface, and the management server extracts the stored data from the database 260 and delivers it to the manager.

The deep learning algorithm of the accident analysis unit 120 will be described.

The accident analysis unit 120 may analyze the accident through an algorithm of a convolution neural network (CNN) or a recurrent neural network (RNN).

3 is a diagram showing a process of processing data in a memory according to an embodiment of the present invention, FIG. 4 is a diagram showing a process of classifying input data according to an embodiment of the present invention, and FIG. 5 is a diagram showing a process of classifying input data according to an embodiment of the present invention. 6 is a structure diagram of a CNN algorithm according to an embodiment of the present invention, and FIG. 7 is a result table of an accident detection test through a CNN algorithm according to an embodiment of the present invention. 8 is a result table of the accident detection test through the CNN algorithm according to an embodiment of the present invention, and FIG. 9 is an experimental table of the accident test through the CNN algorithm according to an embodiment of the present invention.

Referring to FIGS. 3 to 9, the accident analysis unit 120 stores data received from the sound collection unit 110 in memory. And, when the storage capacity in the memory reaches 0.05 seconds (4800 bytes), it is transferred to the analysis program. Among the data in the memory, the data of the first 0.05 seconds (2400 bytes) is removed, and the latter data can be preserved for the next procedure.

Data transmitted to the analysis program may be converted into frequency components through Fast Fourier Transform (FFT). In this case, the frequency band may be 300 Hz to 7 kHz. Due to this, about 650 to 700 arrays can be created. Also, the changed data is converted into a vector of size n*1. The formula for FET is:

, (수식 1)FFT formula:

, (Equation 1)

Here, n is determined according to the frequency band to be used through preliminary analysis of the collected sound, and in general, a frequency band of around 7 KHz can be used. In addition, when data bundled in units of 0.05 seconds is accumulated for a total of 3 seconds, 60 n*1 vectors are created, and n*60 matrix data can be generated by combining them.

Therefore, for the matrix obtained through the above process, the convolutional neural network (CNN) technique among deep learning techniques can be used to analyze the pattern of the matrix for thinking. Here, a Convolution Neural Network (CNN) is a technique used for image recognition and is a type of artificial neural network using convolution.

As shown in FIG. 4, input values (3-second data, 670*30 matrix) input in real time are put into a model created in deep learning learning to derive label values (accident status and type), data intensity, and probability. Also, by utilizing the continuity of time, the order of label values is set and determined as the final accident.

On the other hand, in the case of an input value, as shown in FIG. 5 , the first 1 second of data is deleted and the next 1 second is added to configure the input value of the next 3 seconds. That is, if the first input value consists of sound data from t seconds to t+3 seconds, the next frame may consist of sound data from t+1 seconds to t+4 seconds.

Referring to FIG. 6, Input is an input value, and each line predicts what kind of sound data it is with the value that comes out through the weight and activation function (Sigmoid, Relu, etc.) for which nodes are calculated, and the softmax() function at the end. will do

7 shows the output value predicted by the algorithm and the actual sound value when 30 seconds of accident (skid or collision) sound is input to the collector as an input value.

8 compares the output value predicted by the algorithm with the actual sound value when 24-hour sound data on a specific date (2019.8.21) is input as an input value.

On the other hand, in the case of an algorithm for detecting the location of an accident, sound collected by at least two or more sound collection units 110 for the same accident may be used.

Specifically, in the case of a sound signal, as a type of radio wave, the intensity of the signal attenuates as the distance from the initially generated sound increases, which can be expressed as an attenuation rate.

Lp(R2) = Lp(R1) - 20 Log 10 (R2/R1), (Where Lp(R1) is the sound pressure at point 1, Lp(R2) is the sound pressure at point 2, R1: occurs It is the distance from the point to point 1, and R2: is the distance from the point of occurrence to point 2.)

In the case of radio waves, since they are transmitted in both directions, the above equation can be used by using the absolute value of the distance when using sound collectors located one on each side with the sound source as the center.

Specifically, if T is the distance between the two sound collection units 110 and the sound source is located between the two distances, it can be defined as R1+R2 = T or R1 = T - R2.

Therefore, if the distance T between the two sound receivers 110 and the sound pressure (Lp(R1), Lp(R2)) in each collector can be measured, R1 and R2 are calculated using the following equation, In other words, it is possible to specify the point where the accident occurred.

Lp(R2) = Lp(R1) - 20 Log 10 (T-R1/R1)

Accordingly, the accident analysis unit 120 may transmit the contents of the accident identified through the above algorithm and the accident occurrence point to the image collection unit 150 or the accident processing unit 130 .

Hereinafter, the process of detecting accidents through the RNN algorithm will be described.

10 is an exemplary diagram of matrix data in the RNN algorithm according to an embodiment of the present invention, and FIG. 11 is a schematic diagram of a neural network structure according to an embodiment of the present invention.

Referring to FIGS. 10 and 11, when an input value has a sequential property such as time, the previous input value affects the next input value. Conventional DNNs do not reflect this temporal dependence, but recurrent neural networks (RNNs) can determine whether there is an accident by considering the state of previous input values through a recurrent layer.

First, acoustic data in the time domain is converted into values in the frequency domain through fast Fourier transform (FFT), and at this time, the frequency band may be 300 Hz to 7 kHz. Due to this, about 650 to 700 arrays can be created. Using this, sound data with a length of 3 seconds is converted.

This can be expressed as a matrix and used as an input value. That is, 28 pieces of 670x30 matrix data of 2 seconds each may be generated from one piece of sound data having a length of 30 seconds.

A Recurrent Neural Network (RNN) uses the hyperbolic tangent as an activation function to the hidden state calculated by multiplying the weights from the input values and adding the bias, and the previous hidden state by multiplying the weights and adding them. As the result becomes the next hidden state, temporal dependence can be considered.

Therefore, the expression of the hidden state at an arbitrary point in time t is as follows.

)

)

Unlike CNN, which uses ReLU as an activation function, RNN uses hyperbolic tangent function. At this time, the expression of the hyperbolic tangent function is as follows.

f(x)=tanh(x)

The neural network structure schematized for all input values is shown in FIG. 11 .

Therefore, it is possible to predict whether or not an accident occurs according to the last output value y28 of FIG. 10 .

Based on the result detected by the accident analysis unit 120, when an accident in the road (tunnel) is detected, the receiver 140 may be activated.

The receiver 140 may include a speaker, a camera, and control buttons.

When the receiver 140 is activated, the user can specifically deliver road conditions including accident notifications to the manager through the sound collection unit 110 . According to the present embodiment, a separate microphone for receiving a user's voice in a tunnel is not required, and the user's voice can be received only by the sound collecting unit 110, so the number of parts can be reduced.

Accordingly, when the receiver 140 is activated, the sound collection unit 110 can function as a microphone to deliver the voice of the person involved in the accident or the discoverer of the accident and the sound of the scene to the manager.

In addition, an accident party or an accident finder may receive a manager's voice through the speaker.

Furthermore, a camera for forward shooting may be disposed on the outer surface of the sound-based automatic road accident detection system 100, and the accident party, the accident finder, and the scene of the accident may be photographed through the camera.

In addition, the sound-based automatic road accident detection system 100 includes a control button, and the user activates the receiver 140 through manipulation of the control button, or controls the volume of the speaker and the sound collection unit 110. You can adjust the volume of the input to Rho.

Meanwhile, for smooth two-way voice communication, when the receiver 140 is activated, the reception sensitivity of the sound collection unit 110 may be higher than that in the accident detection mode.

Based on the result detected by the accident analysis unit 120, when an accident in the road (tunnel) is detected, the image collection unit 150 for monitoring the accident site may be activated.

12 is a cross-sectional view showing installation of an image collection unit in a tunnel according to an embodiment of the present invention, and FIG. 13 is a side view showing installation of an image collection unit in a tunnel according to an embodiment of the present invention.

Referring to FIGS. 12 and 13 , the image collecting unit 150 may be disposed in a tunnel. The image collection unit 150 may be installed on the wall of the tunnel.

The image collection unit 150 may include a rail 152 , an arm 154 , and a camera 156 .

The rail 152 may be installed on a wall surface of the tunnel. The rail 152 may be installed on the ceiling surface of the tunnel or the inner surface of the side wall. For example, when the tunnel has a semicircular cross-sectional shape, components in the image collecting unit 150 including the rail 152 are disposed below the light source 160 for radiating light into the tunnel, It may be disposed above the sound collecting unit 110 . Grooves or protrusions may be formed on the surface of the rail 152 so that the arm 154 and the camera 156 can slide.

The rail 152 may be formed as a single line within a section defined from the entrance to the exit of the tunnel. However, this is exemplary, and the rails 152 may be provided in plural and disposed adjacent to each other. In this case, the arm 154 and the camera 156 may be provided on each of the plurality of rails 152 .

A power supply unit (not shown) electrically connected to the camera 156 may be provided on the rail 152 . Accordingly, power may be supplied to the camera 156 through the power supply unit or the camera 156 may be charged. Charging of the camera 156 may be performed at normal times when an accident does not occur.

The arm 154 may have one end coupled to the rail 152 and the other end coupled to the camera 156 . The arm 154 may slide along the rail 152 .

The camera 156 is disposed at the other end of the arm 154 and can photograph the space within the tunnel. The camera 156 may be moved along the rail 152 together with the arm 154 . The camera 156 may include a thermal imaging camera, an infrared camera, and the like, and a smoke detector capable of detecting whether or not smoke is generated may be provided on the outer surface.

The camera 156 may rotate with respect to the arm 154 in a horizontal direction with respect to the ground.

Therefore, when it is determined that an accident is detected at a specific point in the accident analysis unit 120 based on the acoustic information collected by the sound collection unit 110, the arm 154 and the camera 156 The accident area can be photographed by moving to a specific point where an accident occurred along the rail 152 .

Meanwhile, for the convenience of a manager, the movement of the arm 154 and the camera 156 on the rail 152 may be controlled wired or wirelessly.

According to the structure as described above, since the accident occurrence point can be rapidly photographed and monitored by a manager, there is an advantage in that an immediate response is possible in the event of an accident.

In addition, even if a plurality of CCTVs are not provided compared to the prior art, since the entire area in the tunnel can be closely photographed by moving the camera, there is an advantage in that the economic cost required for installation and maintenance can be reduced.

In the above, even though all the components constituting the embodiment of the present invention have been described as being combined or operated as one, the present invention is not necessarily limited to these embodiments. That is, within the scope of the object of the present invention, all of the components may be selectively combined with one or more to operate. In addition, terms such as 'include', 'comprise' or 'having' described above mean that the corresponding component may be present unless otherwise stated, and thus exclude other components. It should be construed as being able to further include other components. All terms, including technical or scientific terms, have the same meaning as commonly understood by a person of ordinary skill in the art to which the present invention belongs, unless defined otherwise. Commonly used terms, such as terms defined in a dictionary, should be interpreted as being consistent with the contextual meaning of the related art, and unless explicitly defined in the present invention, they are not interpreted in an ideal or excessively formal meaning.

The above description is merely an example of the technical idea of the present invention, and various modifications and variations can be made to those skilled in the art without departing from the essential characteristics of the present invention. Therefore, the embodiments disclosed in the present invention are not intended to limit the technical idea of the present invention, but to explain, and the scope of the technical idea of the present invention is not limited by these embodiments. The protection scope of the present invention should be construed according to the claims below, and all technical ideas within the equivalent range should be construed as being included in the scope of the present invention.

Claims (5)

a sound collection unit that collects sound information within the tunnel;
a light source radiating light within the tunnel;
an accident analysis unit that determines whether or not there is an accident in the tunnel through the sound information;
a receiving unit activated based on the result detected by the accident analysis unit; and
Based on the information detected by the accident analysis unit, it includes an image collection unit for photographing an accident area in the tunnel,
The image collection unit includes a rail and a camera moving along the rail,
The rail is arranged to form a line on the wall surface of the tunnel within a section defined from the entrance to the exit of the tunnel,
The image collection unit is disposed below the light source and above the sound collection unit,
When an accident occurrence is detected through the accident analysis unit, the receiver unit is activated and the user's voice in the tunnel is received through the sound collection unit,
When the receiver is activated, the reception sensitivity of the sound collector is higher than when the receiver is deactivated.
delete delete delete According to claim 1,
The accident analysis department,
DBSAVE for storing the sound information in a database in units of 1 minute;
StreamGW delivering real-time sound information; and
Includes an ADSS that analyzes the acoustic information to detect an accident,
The accident analysis algorithm in the ADSS is an accident site monitoring system performed through a convolution neural network (CNN) or a recurrent neural network (RNN).

Images