KR20040025148A - Apparatus and method for security using an environmental sound - Google Patents

Abstract

PURPOSE: A security system using environmental sound and a method thereof are provided to sense correctly an abnormal state by using the sound generated according to environmental elements. CONSTITUTION: A security system using environmental sound includes a microphone(200), an analog/digital converter(210), a feature extractor(220), a memory(240), a trouble determination unit(250), and an alarm output unit(260). The microphone(200) is installed at a predetermined monitoring place to convert a sensed sound wave to an electrical signal. The analog/digital converter(210) is used for converting an output signal of the microphone to a digital signal. The feature extractor(220) is used for extracting features of abnormal states from an output signal of the analog/digital converter. The memory(240) stores an environmental reference model corresponding to an abnormal signal. The trouble determination unit(250) compares the environmental reference model of the memory to the extracted features of the feature extractor and generates an abnormal state signal according to a compared result. The alarm output unit(260) generates alarm sound according to the abnormal state signal.

Description

Security device and method using environmental sound {Apparatus and method for security using an environmental sound}

The present invention relates to a security device and a method using the environmental sound to monitor whether or not the abnormal occurrence of the place by using the environmental sound generated according to the environment of the predetermined place that requires monitoring.

In general, a security device is a device installed and used in a place such as a bank or a department store in order to prevent a crime, and photographs a place to be monitored, and judges whether a crime or the like occurs while the watcher monitors the captured image from a remote place. In the event of a crime, an observer can take immediate response to the crime.

1 is a view showing the configuration of a conventional security device. As shown in the drawing, a plurality of CCTV cameras 100 photographing a predetermined place to be monitored and a plurality of video signals captured by the plurality of CCTV cameras 100 may be sequentially selected one by one or displayed on one screen. A video controller 110 for performing image control such as editing such that the video is controlled, a plurality of monitors 120 displaying an image controlled by the image controller 110, and an image controlled by the image controller 110. A video recorder 130 is stored in a storage medium such as a tape.

Conventional security device having such a configuration is installed first to shoot a predetermined place to monitor a plurality of CCTV cameras 100 first. Then, the plurality of CCTV cameras 100 respectively photographs a predetermined place to be monitored, and outputs the captured video signal to the image controller 110.

The video controller 110 sequentially selects or edits video signals photographed by the plurality of CCTV cameras 100 according to a monitor operation, and outputs a plurality of video signals output by the video controller 110. The monitor 120 is displayed on the screen so that the monitor can visually monitor a predetermined place, and the video recorder 130 stores the video in a predetermined storage medium such as a video tape.

Such a conventional security device can visually and effectively monitor a predetermined place where a CCTV camera is photographed to take an immediate countermeasure against a criminal act when a criminal act occurs. However, since only the video signal photographed at a certain place is monitored visually, the monitor must keep watching the monitor, which makes it very inconvenient and easy to get tired, while the monitor cannot monitor the monitor for other reasons. When a criminal act occurs, the watcher does not recognize the occurrence of the criminal act and cannot take action on the occurrence of the criminal act promptly. There was a problem in that if the criminal act was avoided, the criminal act could not be monitored.

In addition, the above-described conventional security device does not monitor using the environmental sound generated at a predetermined place to be monitored. That is, the plurality of CCTV cameras 100 are typically provided with a microphone to convert the sound at the corresponding place into an electrical sound signal, and the electrical sound signal and the video signal photographed by the plurality of CCTV cameras 100 Although they are transmitted together, the plurality of sound signals must be listened to by the monitor in order to monitor the sound signals, and the number of sound signals to be listened to is too high to listen to all of them. I can't do it.

In addition, a plurality of sound signals may be monitored by a plurality of monitors while listening to each other. However, when a lot of human resources are wasted and the monitoring place is a place where there is a lot of noise, the sound signal accurately monitors the occurrence of criminal activity. There was a problem such as not being able to.

Accordingly, it is an object of the present invention to provide a security apparatus and method using environmental acoustics for monitoring whether an abnormality occurs in a monitoring location using environmental acoustics generated according to the environment of a predetermined location to be monitored.

It is another object of the present invention to generate an environmental acoustic reference model in advance by using the environmental sound generated according to the environment of a predetermined place to be monitored, and accurately detect the abnormality of the monitoring location by using the generated environmental acoustic reference model. The present invention provides a security device and method using environmental sound.

Security apparatus and method using the environmental sound of the present invention having such an object is to install a microphone in the place to be monitored to detect the environmental sound, and to extract a predetermined feature from the detected environmental sound.

In general, the characteristics of environmental acoustics can be represented by spectral characteristics due to differences in environmental conditions and spectral variation characteristics due to periodic noise conditions in the environment. As parameters effective for the detection of environmental sound, LPC (Linear Prediction Coding) coefficients, autocorrelation coefficients and cesptrum are known. Among the above parameters, the septum is very effective for the detection of environmental sound. In the present invention, the septum is extracted from the environmental sound detected by the microphone, for example, at unit time intervals. Learn to create an environmental acoustic reference model.

The learning of the septum is, for example, by constructing a recognition model of each environmental sound by using acoustic characteristic parameters according to the environment of the place to be monitored, HMM (Hidden Markov Model), pattern time matching (Dynamic Time Warping) , An expert system (Spectrogram Reading System) and neural network (Neural Network) to learn by generating an environmental acoustic reference model, preferably by learning with an HMM to generate an environmental acoustic reference model.

When the environmental acoustic reference model is generated, it detects the sound of the place to be monitored in real time, extracts the septum, and compares the extracted septum with the environmental acoustic reference model to determine whether an abnormality occurs.

Therefore, according to the security device using the environmental sound of the present invention, by installing a microphone roll in a predetermined place to be monitored, the sound generated from the surrounding objects are detected and converted into an electrical signal, and the output signal of the microphone is converted into an analog / digital converter Converts it into a digital signal, extracts a feature from the output signal of the analog / digital converter to detect whether an abnormal signal is generated, and stores an environmental acoustic reference model to detect whether an abnormal signal is generated in advance in the memory; The feature extraction unit compares the extracted feature with the environmental acoustic reference model stored in the memory to determine whether an abnormal signal is generated.When the abnormality determination unit determines that an abnormal signal is generated, an alarm output unit generates an alarm to alert the monitor. It is characterized by the alarm.

A feature extracted by the feature extractor is any one of a linear prediction coding (LPC) coefficient, an autocorrelation coefficient, and a septum, and the FFT (Fast Fourier Transform) FFT output signal of the analog-to-digital converter. The power spectrum operation unit obtains a power spectrum of a frequency domain corresponding to a voice signal from the output signal of the FFT unit, the natural log operation unit takes a natural log value from the power spectrum obtained by the power spectrum operation unit, and the natural log. The IFFT (Inverse Fast Fourier Transform) IFFTs the natural logarithm taken by the operation unit to obtain a time-domain septum.

The abnormality determination unit detects a difference value by comparing the feature extracted by the feature extractor with the environmental acoustic reference model, and compares the absolute value of the detected difference value with a threshold value preset by the second comparator. To determine whether the abnormality occurs.

The present invention further includes a feature learning unit for learning a feature extracted by the feature extracting unit and storing a learning result as an environmental acoustic reference model in the memory. The learning of the feature learning unit includes a HMM (Hidden Markov Model) algorithm and a pattern definition. Learning by one of legal algorithms, expert system algorithms, and neural network algorithms.

In addition, the security method using the environmental sound of the present invention, by extracting a predetermined feature from the digital sound signal received from the monitoring place, and learning a plurality of extracted features to create and store the environmental acoustic reference model from the monitoring place Extracts a predetermined feature from the digital sound signal received in real time at unit time intervals, detects a difference value by comparing the extracted predetermined feature with the environmental acoustic reference model stored in the first step, and detects the absolute value of the detected difference value. The alarm control signal is generated when the value is equal to or greater than the preset threshold value.

1 is a block diagram showing the configuration of a conventional security device,

2 is a block diagram showing the configuration of the security device of the present invention,

3 is a signal flow diagram showing a security method of the present invention,

4 is a signal flow diagram illustrating an operation of generating and storing an environmental acoustic reference model of FIG. 3.

* Description of the symbols for the main parts of the drawings *

200: microphone 210: analog to digital converter

220: feature extraction unit 221: FFT unit

223: power spectrum calculator 225: natural logarithm calculator

227: IFFT 230: feature learning unit

240: memory 250: abnormal occurrence determination unit

251: first comparator 253: second comparator

260: alarm output unit

Hereinafter, with reference to the accompanying drawings of Figures 2 to 4 will be described in detail a security device and method using the environmental sound of the present invention.

2 is a block diagram showing the configuration of the security device of the present invention. As shown in the drawing, a plurality of microphones 200 are respectively installed in a plurality of places to be monitored to convert sound according to the environment of the place into an electrical signal, and the plurality of microphones 200 are converted into electrical signals. An analog / digital converter 210 for converting a sound signal into a digital signal, a feature extractor 220 for extracting a predetermined feature of environmental sound from an output signal of the analog / digital converter 210, and the feature extractor A feature learning unit 230 for generating an environmental acoustic reference model by learning the extracted feature 220, a memory 240 for storing the environmental acoustic reference model generated by the feature learning unit 230, and the feature extraction The abnormality occurrence determination unit 250 and the abnormality occurrence determination unit 250 which determine whether an abnormality occurs in a predetermined place by comparing the feature extracted by the unit 220 in real time with the environmental acoustic reference model stored in the memory 240. More than The alarm output unit 260 is configured to generate an alarm when the occurrence is determined.

The feature extractor 220 extracts a septum from, for example, environmental sound, and performs an FFT (Fast Fourier Transform) on the output signal of the analog-to-digital converter 220 to convert data into frequency data. A power spectrum calculator 223 for obtaining a power spectrum of a frequency domain corresponding to sound from the output signal of the FFT unit 221, and a natural log value of the power spectrum calculated by the power spectrum calculator 223; A natural logarithm calculation unit 225 to take up; An IFFT unit 227 is obtained by IFFTing the natural log value taken by the natural log operation unit 225 to obtain a time-domain septum.

The abnormality occurrence determining unit 250 compares a predetermined feature extracted by the feature extractor 220 at unit time intervals with an environmental acoustic reference model stored in the memory 250 to detect a difference value. The second comparator comparing the comparator 251 and the output signal of the first comparator 215 with a threshold value experimentally set in advance to determine whether an abnormality has occurred, and generating an alarm control signal when the abnormality is determined. It consisted of (253).

In the security device of the present invention configured as described above, the microphone 200 installed at a predetermined place to be monitored receives electric sound, that is, a predetermined sound including various noises or vibration sounds generated according to the environment of the installed place as an electrical signal. After the conversion, the converted environmental sound signal is converted into a digital signal by the analog-to-digital converter 210 and input to the feature extraction unit 220.

The feature extractor 220 converts the digital signal input from the analog-to-digital converter 210 into the data of the frequency domain by FFTing the FFT unit 221, and the power spectrum calculator 223 performs the data of the frequency domain. The power spectrum of the frequency domain corresponding to the sound is calculated. The natural log operator 225 extracts the natural log value from the calculated power spectrum, and the IFFT unit 227 IFFTs the natural log value to obtain a septum in the time domain.

As described above, before the feature extractor 220 determines whether there is an abnormality in a predetermined place by using the septums extracted at a set unit time interval, the feature extractor 220 needs to learn a plurality of the septums to generate an environmental acoustic reference model. The unit 230 learns a plurality of concepts and stores the learned results in the memory 240 as an environmental acoustic reference model. Learning of the feature learning unit 230, for example, learn the septum with the HMM.

As described above, in the state in which the environmental acoustic reference model for detecting an abnormality is stored in the memory 240, the analog / digital converter 210 performs a digital signal in real time on the analog sound signal output from the microphone 200 in real time. The feature extractor 220 extracts and outputs the septum from the digital signal.

The septum extracted by the feature extraction unit 220 is input to the first comparator 251 of the abnormality occurrence determination unit 250, and the first comparator 251 is an environmental acoustic reference model previously stored in the memory 240. And the difference extracted by the feature extractor 220 to detect the difference value, and compares the absolute value of the detected difference value with a threshold value experimentally preset by the second comparator 253.

As a result of the comparison, when the absolute value of the difference value is smaller than the threshold value, the second comparator 253 determines that no abnormality has occurred at a predetermined place to be monitored. When the absolute value of the difference value is larger than the threshold value, the second comparator 253 determines that a predetermined abnormality such as an intruder is generated at a predetermined place to be monitored, and the determination signal of the second comparator 253 is determined. In accordance with the alarm output unit 260 informs that the abnormality occurred to the monitor by the alarm sound and light, such that the action can be taken for the occurrence.

Meanwhile, the feature extractor 220, the feature learner 230, the memory 240, and the abnormality occurrence determiner 250 are not configured with hardware, but are configured with a microcomputer, a microprocessor, and the like. As can be configured to perform according to Figure 3 is a signal flow diagram showing the security method of the present invention when performed in software. As shown in FIG. 300, the controller first generates and stores an environmental acoustic reference model.

Generating and storing the environmental acoustic reference model, as shown in FIG. 4, converts the digital signal output from the analog-to-digital converter 210 to the data in the frequency domain by FFTing at a unit time interval in step 400, and In 402, the power spectrum of the frequency domain corresponding to the voice signal is obtained from the data of the frequency domain. In step 404, the natural logarithm of the power spectrum is taken, and then IFFT is performed in step 406, the time-domain Find the strum. When the septum is obtained, the septum is learned in step 408 to generate an environmental acoustic reference model as a learning result, and the generated environmental acoustic reference model is stored in step 410 and then returned.

When the environmental acoustic reference model is stored, the digital signal output from the analog-to-digital converter 210 is converted into data in the frequency domain by FFT in unit time intervals in step 302, and the data in the frequency domain in step 304. In step 306, the power spectrum of the frequency domain corresponding to the voice signal is obtained. In step 306, the natural logarithm of the power spectrum is taken, and then IFFT is obtained in step 308 to obtain the receptive spectrum in the time domain.

In the next step 310, the obtained difference is compared with the stored environmental acoustic reference model to detect the difference value, and the absolute value of the detected difference value is the threshold value experimentally preset in step 312. Determine if it is abnormal. If the absolute value of the difference value is not equal to or greater than the threshold value, the determination returns to step 302 where no abnormality occurs at the place to be monitored, and the monitoring operation is continued. When the absolute value of the difference is greater than or equal to the threshold, the determination result indicates that an intruder or the like has occurred in the place to be monitored. In step 314, an alarm is outputted and the alarm is returned. Continue.

On the other hand, while the present invention has been shown and described with respect to specific preferred embodiments, various modifications and changes of the present invention without departing from the spirit or field of the invention provided by the claims below It can be easily understood by those skilled in the art. For example, as described above, the sound is detected and monitored at any one place to be monitored. In the practice of the present invention, the sound is detected at a plurality of places to be monitored, and each of the predetermined sounds is predetermined. By extracting and learning features, each environmental acoustic reference model can be generated, and each environment acoustic reference model can be configured to monitor a plurality of places.

As described in detail above, the present invention can accurately monitor whether or not an abnormality occurs by using sound generated according to the environment at a predetermined place to be monitored, and also when an error occurs when installed with an apparatus for monitoring using an image. You can fully monitor whether or not.

Claims (7)

A microphone installed at a predetermined place to be monitored to detect sound generated from surrounding objects and convert the sound into an electrical signal; An analog / digital converter for converting an output signal of the microphone into a digital signal; A feature extraction unit for extracting a feature for detecting whether an abnormal signal is generated from an output signal of the analog / digital converter; A memory for storing in advance an environmental acoustic reference model for detecting whether an abnormal signal is generated; An abnormality occurrence determining unit which determines whether an abnormal signal is generated by comparing the extracted feature by the feature extracting unit with an environmental acoustic reference model stored in the memory; And Security device using an environmental sound consisting of an alarm output unit for generating an alarm when the error occurrence determination unit determines the occurrence of an abnormal signal. The method of claim 1, wherein the feature extractor extracts the extracted feature; A security device using environmental sound, characterized in that any one of LPC (Linear Prediction Coding) coefficient, autocorrelation coefficient and the septum. The method of claim 1, wherein the feature extraction unit; An FFT unit converting the output signal of the analog-to-digital converter into fast fourier transforms (FFTs) to data in a frequency domain; A power spectrum calculation unit for obtaining a power spectrum of a frequency domain corresponding to a voice signal from the output signal of the FFT unit; A natural logarithm calculation unit which takes a natural logarithm of the power spectrum obtained by the power spectrum operation unit; And And a IFFT unit for obtaining the natural domain values obtained by the natural log calculation unit by inverse fast fourier transform (IFFT). According to claim 1, wherein the abnormality determination unit; A first comparator configured to detect a difference value by comparing the feature extracted by the feature extractor with the environmental acoustic reference model; And And a second comparator configured to determine whether the abnormality occurs by comparing an absolute value of the difference value detected by the first comparator with a preset threshold value. The method of claim 1, And a feature learning unit for learning the features extracted by the feature extracting unit and storing a learning result as an environmental acoustic reference model in the memory. The method of claim 5, wherein the learning of the feature learning unit; Security device using environmental sound, characterized in that learning by any one of HMM (Hidden Markov Model) algorithm, pattern matching algorithm, expert system algorithm and neural network algorithm. A first step of extracting a predetermined feature from a digital sound signal received from a monitoring place; A second step of generating and storing an environmental acoustic reference model by learning a plurality of features extracted in the first step; A third process of extracting a predetermined feature from the digital sound signal received in real time from the monitoring place at unit time intervals; A fourth step of detecting a difference value by comparing the predetermined feature extracted in the third step with an environmental acoustic reference model stored in the first step; And And a fifth step of generating an alarm control signal when the absolute value of the difference value detected in the fourth step is equal to or greater than a preset threshold value.