JP6056527B2 - Intruder detection device - Google Patents

Description

  The present invention relates to a technique for detecting an intruding object using a vibration sensor.

  2. Description of the Related Art Conventionally, a technique is known in which vibration propagated from an intruding object is detected using a vibration sensor such as a pressure sensor or an acceleration sensor, and the intruding object is detected by analyzing the detection result. This type of prior art is disclosed in, for example, Japanese Patent Application Laid-Open No. 4-104018 (Patent Document 1) and Japanese Patent Application Laid-Open No. 2012-58340 (Patent Document 2).

  The detection device disclosed in Patent Document 1 detects a vibration wave from an intruding object that enters the monitoring area using a vibration sensor, analyzes the sensor output by a linear prediction analysis method, and determines the characteristics of the vibration wave. The type of the intruding object (for example, a vehicle or a person) is identified by extraction. Further, in Patent Document 2, even when there is fluctuation in the walking cycle of a person who is an intruding object, the sound period of the intruding object is calculated from the sounds detected by the microphone using a trigonometric function such as a cosine function. A resulting sound detection device is disclosed.

  Japanese Patent Laid-Open No. 2004-125641 (Patent Document 3) proposes a technique for identifying an abnormal sound generated by a target device based on an average kurtosis value of a waveform distribution of sound detected by a microphone. Japanese Patent Laid-Open No. 2002-23776 (Patent Document 4) compares the kurtosis of probability distributions for speaker speech signals and non-speech noise signals separated based on the blind separation method, and based on the results. Thus, a method capable of identifying speaker voice even in a noisy environment is disclosed.

JP-A-4-104018 JP 2012-58340 A JP 2004-125641 A JP 2002-23776 A

  The sound detection device disclosed in Patent Document 2 detects continuous sound such as footsteps from acoustic data using a threshold value, and classifies the detected sound according to its frequency structure. The sound detection device calculates the period of the classified sound using a trigonometric function such as a cosine function. However, since the threshold value that is referred to in the sound detection is a preset value and not a value optimized according to the installation environment of the sound detection device, there is a problem that the type of the intruding object is erroneously detected. is there.

  For example, in an environment where vehicles, aircraft, and personnel move, vibration waves generated by movement of the vehicles and aircraft are vibration waves similar to Gaussian noise. Since the average level of such vibration waves arriving at the sound detection device can fluctuate unpredictably with the movement of the vehicle and aircraft, under these circumstances, it is possible to accurately detect footsteps emitted by personnel using the above threshold. It is difficult to detect.

  In view of the above, an object of the present invention is to provide an intruding object detection device that can detect an intruding object at a distance and can identify the intruding object with high accuracy as it approaches.

  An intruding object detection device according to an aspect of the present invention orthogonally transforms a time-series sensor signal representing a vibration wave including background noise into a frequency domain signal, and separates a band signal of a predetermined frequency band from the frequency domain signal. A frequency analysis unit, a background noise estimation unit that calculates a determination threshold based on a statistic of temporal change of the band signal when the background noise follows a Gaussian distribution, and the determination threshold from the sequence of the band signal A signal extraction unit that extracts a signal having a signal level exceeding, a significant vibration extraction unit that extracts a significant signal sequence that constitutes a periodic vibration from an output sequence of the signal extraction unit, and the significant signal sequence is specified When it is determined that a frequency determination unit that determines whether or not the vibration generation frequency has an output sequence of the signal extraction unit has the specific vibration generation frequency, a band of the sensor signal is filtered. A bandpass filter that outputs a filter signal having the predetermined frequency band and a signal identification unit that calculates a kurtosis of a waveform distribution of the filter signal and detects a type of an intruding object based on the kurtosis. It is characterized by providing.

  According to the present invention, an intruding object can be detected and identified with high accuracy.

It is a block diagram which shows schematic structure of the intruding object detection apparatus which is embodiment which concerns on this invention. It is a block diagram which shows schematic structure of a frequency analysis part. (A)-(D) are the graphs which illustrate the relationship between the various waveforms of a band signal, and a determination threshold value. (A) is a figure which shows schematically the waveform example of a filter signal, (B) is a figure which shows kurtosis. (A)-(C) are the figures for demonstrating kurtosis. (A)-(C) are figures which show the signal waveform in a periodic vibration generation period.

  Embodiments according to the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing a schematic configuration of an intruding object detection device 1 according to an embodiment of the present invention. As shown in FIG. 1, the intruding object detection device 1 includes a vibration sensor 10, a preamplifier 11, and an analysis circuit 20, and the analysis circuit 20 includes an A / D converter (ADC) 21 and a frequency analysis unit 22. , Detection processing units 23 ₁ to 23 _M (M is a positive integer), band-pass filters 24 _{1 to} 24 _M , a signal identification unit 40, and an intruding object detection unit 43. The analysis circuit 20 of the present embodiment has _M detection processing units 23 ₁ to 23 _M as a preferable configuration, but is not limited to this, and has only one detection processing unit. The configuration of the analysis circuit 20 may be changed.

  The vibration sensor 10 has a function of detecting a vibration wave including background noise and outputting an analog electric signal indicating the detection result as a sensor signal. As the vibration sensor 10, for example, a pressure sensor that detects a vibration wave that has propagated through a ground from a vibration source may be used, and the vibration sensor 10 may be configured using a piezoelectric vibration element. The preamplifier 11 amplifies the sensor signal input from the vibration sensor 10 with an appropriate gain and outputs the amplified signal to the analysis circuit 20.

In the analysis circuit 20, the A / D converter 21 converts the sensor signal input from the preamplifier 11 into a digital sensor signal TS. The frequency analysis unit 22 performs frequency analysis on the digital sensor signal TS input from the A / D converter 21 and outputs _M band signals SF _{1 to} SF _M.

FIG. 2 is a block diagram illustrating a schematic configuration of the frequency analysis unit 22. As shown in FIG. 2, the frequency analysis unit 22 includes an orthogonal transform unit 221, a signal separation unit 222, and a signal addition unit 223. The orthogonal transform unit 221 performs orthogonal transform such as fast Fourier transform on the digital sensor signal TS to generate a frequency domain signal FS. For example, the orthogonal transform unit 221 samples thousands of discrete signals having a predetermined time length from the sequence of the digital sensor signal TS, and performs fast Fourier transform on the sampled discrete signals to generate thousands of frequency domain signals FS. Can be generated. Signal separating unit 222, the signal _F 1 of the first through the frequency band of the M from the frequency domain signal FS, ..., to separate the _{F M.} These first to Mth frequency bands may overlap each other. The signal adder 223 adds the signals to each other for each frequency band to generate _M band signals SF ₁ ,..., SF _M. Specifically, the signal adder 223 includes integrators 223 ₁ ,..., 223 _M to which _M band signals SF ₁ ,..., SF _M are respectively input, and these integrators 223 ₁ ,. , 223 _M, the bandwidth signal _SF 1, ..., band signal _SF 1 of M lines and respectively integrating separately SF _M, ..., it is possible to generate an SF _M.

As the frequency band, it is desirable to select a plurality of characteristic frequency bands of vibration waves generated by intrusion of an intruding object such as a person into the monitoring area. For example, the following first to third band signals SF ₁ , SF ₂ , SF ₃ can be generated based on the frequency domain signal FS in accordance with the type of ground (clay, sand, etc.) through which the vibration wave propagates. .
Ground A: signal SF ₁ in a low frequency band of 50 Hz or less,
Ground B: signal SF ₂ in an intermediate frequency band of 50 Hz to 100 Hz,
Ground C: signal SF _{3 in} a high frequency band of 100 Hz to 150 Hz.

Detection processing unit ₂₃ 1, ..., 23 _M, the bandwidth signal _SF 1 from the frequency analysis section 22 are supplied respectively, ..., and inputs the SF _M, these band signals _SF 1, ..., each significant vibrations from SF _M Can be detected. Since the configurations and operations of the detection processing units 23 ₁ ,..., 23 _M are all the same, the configuration of the detection processing unit 23 ₁ will be described below.

Detection processing unit 23 _1, as shown in FIG. 1, is configured to include a background noise estimation unit 31, the signal extraction unit 32, significant vibration extraction unit 34 and the frequency determination unit 35. Background noise estimation unit 31, the determination threshold value TH based on the statistics of the time variation of the band signal SF ₁ (such as arithmetic mean and variance value) when the background noise detected by the vibration sensor 10 is assumed to follow a Gaussian distribution _It has a function of calculating ₁ . Specifically, if the band signal SF ₁ is a complex signal representing the background noise, each of the real and imaginary parts follow a Gaussian distribution. It is known that the square sum (= Re ² + Im ² ) of the real part Re and the imaginary part Im of the complex signal represents spectral power and follows a chi-square distribution with two degrees of freedom. Desired false alarm probability on the basis of the chi-square distribution can be determined determination threshold value TH ₁ to be constant. Similarly, in the background noise estimation unit 31 of the detection processing unit ₂₃ 2 ~ 23 _M of other series, the determination threshold value _TH 2 to TH _M respectively corresponding to the band signal _SF 2 - SF _M is determined.

Next, the signal extraction unit 32 extracts a signal having a signal level exceeding the determination threshold TH ₁ from the series of the band signal SF ₁ . In other words, the signal extraction unit 32 cuts out and outputs only a signal exceeding the determination threshold TH ₁ from the series of the band signal SF ₁ .

3A to 3D are graphs illustrating the relationship between the various waveforms 51 to 54 of the band signal SF ₁ and the determination threshold value TH ₁ . In FIG. 3 (A) ~ (D) , the horizontal axis represents time and the vertical axis represents the signal level of the band signal SF _1. FIG. 3A shows a band signal waveform 51 obtained from a walking person, FIG. 3B shows a band signal waveform 52 obtained from a traveling vehicle, and FIG. The obtained band signal waveform 53 is shown, and FIG. 3D shows the band signal waveform 54 obtained from rainfall.

The significant vibration extraction unit 34 pays attention to the temporal continuity of the signal extracted by the signal extraction unit 32, and can extract a significant signal sequence constituting the periodic vibration from the signal sequence. For example, a signal appearing at a substantially constant time interval may be extracted as a significant signal sequence. Each of the band signal waveforms 51 and 54 shown in FIGS. 3A and 3D includes a significant vibration component in which a signal waveform exceeding the determination threshold TH ₁ appears at a substantially constant time interval. On the other hand, such a significant vibration component is not found in the band signal waveforms 52 and 53 shown in FIGS.

Frequency determination unit 35, significant signal sequence extracted in significant oscillation extracting unit 34 determines whether a predetermined vibration frequency, and outputs the determination result to the band-pass filter 24 _1. Specifically, the frequency determination unit 35 can determine whether or not the significant signal series has a vibration occurrence frequency due to the intrusion of personnel.

Bandpass filter 24 _1, when the significant signal sequence is determined to have a predetermined vibration frequency, filter the band of the digital sensor signal TS and outputs a filter signal having the first frequency band. At the same time, the other band filters 24 _{2 to} 24 _M filter the band of the digital sensor signal TS and output the filter signals having the second to Mth frequency bands, respectively. The band-pass filters 24 _{1 to} 24 _M may be configured with, for example, a digital filter of FIR (Finite Impulse Response) type or IIR (Infinite Impulse Response) type.

In the signal identification unit 40, the kurtosis calculation units 42 _{1 to} 42 _M calculate the kurtosis of the waveform distribution of the filter signals respectively input from the bandpass filters 24 _{1 to} 24 _M, and generate periodic vibrations based on the kurtosis. Specify the period. For example, a period in which the kurtosis continuously exceeds a predetermined threshold can be set as the periodic vibration generation period. The kurtosis is calculated for each waveform cut out from the filter signal series at regular intervals. For example, it is possible to calculate the kurtosis k _N according to the following equation (1).

Here, x _i is a discrete value of the extracted signal waveform, and N is the number of discrete values of the extracted signal waveform. <X> is an arithmetic average value calculated according to the following equation.

The value of “3” in the second term on the right side of the above equation (1) is the value of the absolute kurtosis of the normal distribution (Gaussian distribution). Therefore, kurtosis k _N denotes the relative kurtosis against a normal distribution. FIG. 4A is a diagram illustrating a waveform example of the filter signal, and FIG. 4B is a graph illustrating the kurtosis of the signal waveform cut out at intervals of 0.5 seconds.

Since personnel walking is a continuous motion, the kurtosis increases continuously. Therefore, the calculated kurtosis k _N is continuously whether exceeds a predetermined magnitude, whether the vibration generated by personnel walking, whether or not the vibration is a vibration caused by rain, or the vehicle or It is possible to determine whether the vibration is caused by the movement of the aircraft.

5A to 5C are diagrams for explaining the kurtosis. The kurtosis is a value for the distribution of discrete values (digital values) obtained by cutting out the waveform itself at regular intervals. Compared with the Gaussian distribution caused by background noise, whether or not the average value is sharply sharp, It evaluates whether it is thick. For example, vibration generated by walking or raining personnel becomes an impulsive signal as illustrated in FIG. 5A, and as shown in FIG. 5C, the distribution Di is around the average value. Becomes sharper and thicker near the hem, and the kurtosis increases. On the other hand, the waveform of the vibration generated by the movement of the vehicle or aircraft is similar to the Gaussian background noise as shown in FIG. 5B, and the amplitude itself becomes large. As shown in FIG. 5C, since the distribution Dg itself is still a Gaussian distribution, the kurtosis has a value of about zero which is the same as the Gaussian background noise. Further, the kurtosis calculated using the waveform processed by the bandpass filters 24 _{1 to} 24 _M is a characteristic frequency of vibration generated by the intrusion of the intruding object as compared with the case of using the waveform not filtered. Since the value is calculated using the extracted waveform of the band, the sensitivity of kurtosis is improved.

  The duration detection unit 41 detects a duration length ΔT of each signal waveform (particularly an impulsive signal) constituting the periodic vibration from the series of digital sensor signals TS in the designated periodic vibration generation period. For example, it is possible to detect the time average value of the amplitude value of the digital sensor signal TS immediately before the period of occurrence of the periodic vibration as a threshold Th, and to detect the duration of a continuous discrete signal exceeding the threshold Th as the duration time ΔT. The intruding object detection unit 43 compares the duration time ΔT with a reference value, and determines whether the intruding object is a person based on the comparison result, or can determine whether there is rainfall. FIG. 6A is a graph showing a waveform example of the digital sensor signal TS in the period of occurrence of periodic vibration. In the graph of FIG. 6A, the horizontal axis represents time, and the vertical axis represents the signal amplitude Y. 6B is a graph showing the square of the amplitude Y of the signal waveform in FIG. 6A, and FIG. 6C is a logarithmic display of the waveform in FIG. 6B.

  The vibration continuation time due to rainfall is characterized by being shorter than the vibration continuation time due to personnel walking (vibration continuation time generated by one step of walking of personnel). For this reason, for example, a threshold value of 50 milliseconds is used, and if the detected duration length ΔT is 50 milliseconds or more, it is determined that the vibration is caused by walking, and if the detected duration length ΔT is less than 50 milliseconds, it is raining. Therefore, it can be determined that the vibration is generated. The intruding object detection unit 43 only needs to output to the outside only the result determined as vibration due to walking.

As described above, in the present embodiment, the background noise estimation unit 31 dynamically calculates the determination thresholds TH _{1 to} TH _M , and the significant vibration extraction unit 34 uses the determination thresholds TH _{1 to} TH _M. A significant signal sequence can be extracted from the extracted signal sequence. Since the determination thresholds TH _{1 to} TH _M may vary according to the average level of background noise, the determination thresholds adapted to a Gaussian band signal waveform (FIGS. 3B and 3C) caused by a vehicle or an aircraft. May not be calculated. In this case, the frequency determination unit 35 may erroneously determine that the significant signal sequence extracted by the significant vibration extraction unit 34 has the vibration occurrence frequency of the impulsive signal. Even in such a case, since the signal identification unit 40 detects the type of the intruding object based on the kurtosis of the filter signal waveform distribution, it is possible to reliably avoid erroneous detection based on the erroneous determination by the frequency determination unit 35. it can.

  Furthermore, since the duration detection unit 41 determines whether the waveform of the sensor signal TS including a continuous impulse signal is due to personnel walking or due to rainfall based on the duration length ΔT, the signal due to rainfall is excluded. Can be narrowed down to only signals due to walking vibration. As a result, it is possible to detect only the vibration due to the intrusion of personnel and output the result to the outside, thereby reducing false detection. Therefore, the type of the intruding object can be detected with very high accuracy.

  As mentioned above, although embodiment concerning this invention was described with reference to drawings, this is an illustration of this invention and various forms other than the above are also employable. For example, a part of the function of the analysis circuit 20 may be realized by a hardware configuration, or may be realized by a computer program executed by a microprocessor including a CPU. When a part of the function is realized by a computer program, the microprocessor loads and executes the computer program from a computer-readable recording medium (for example, an optical disk, a magnetic recording medium, or a flash memory). Part of the function can be realized.

  Also, a part of the configuration of the analysis circuit 20 may be realized by an LSI (Large Scale Integrated Circuit) such as an FPGA (Field-Programmable Gate Array) or an ASIC (Application Specific Integrated Circuit).

DESCRIPTION OF SYMBOLS 1 Intrusion object detection apparatus, 10 Vibration sensor, 11 Preamplifier, 20 Analysis circuit, 21 A / D converter (ADC), 22 Frequency analysis part, 221 Orthogonal transformation part, 222 Signal separation part, 223 Signal addition part, 23 ₁ ˜23 _M detection processing unit, 24 ₁ ˜24 _M band filter, 31 background noise estimation unit, 32 signal extraction unit, 34 significant vibration extraction unit, 35 frequency determination unit, 40 signal identification unit, 41 duration detection unit, 42 ₁ -42 _M kurtosis calculation part, 43 Intrusion object detection part.

Claims (7)

A frequency analysis unit that orthogonally transforms a time-series sensor signal representing a vibration wave including background noise into a frequency domain signal, and separates a band signal of a predetermined frequency band from the frequency domain signal;
A background noise estimation unit that calculates a determination threshold based on a statistic of temporal change of the band signal when it is assumed that the background noise follows a Gaussian distribution;
A signal extraction unit that extracts a signal having a signal level exceeding the determination threshold from the band signal sequence;
A significant vibration extractor for extracting a significant signal sequence constituting periodic vibrations from the output sequence of the signal extractor;
A frequency determination unit for determining whether the significant signal series has a specific vibration occurrence frequency; and
A band filter that filters a band of the sensor signal and outputs a filter signal having the predetermined frequency band when it is determined that the output sequence of the signal extraction unit has the specific vibration occurrence frequency;
An intruding object detection apparatus comprising: a signal identification unit that calculates a kurtosis of a waveform distribution of the filter signal and detects a type of an intruding object based on the kurtosis. The intruding object detection device according to claim 1,
The signal identification unit is
A kurtosis calculation unit for calculating the kurtosis and designating a period of occurrence of periodic vibration based on the calculated kurtosis;
A duration detection unit for detecting a duration of each signal waveform constituting periodic vibrations from the series of sensor signals in the periodic vibration generation period;
An intruding object detection apparatus comprising: an intruding object detection unit that detects the type of the intruding object based on the detected duration time.   The intruding object detection device according to claim 2, wherein the kurtosis calculation unit specifies a period in which the calculated kurtosis continuously exceeds a predetermined threshold as the periodic vibration occurrence period. Intruding object detection device.   The intruding object detection device according to claim 2, wherein the intruding object detection unit compares the detected duration time with a reference value of a vibration duration time generated in one step of walking of a person, An intruding object detection device that determines whether or not the intruding object is a person based on a comparison result.   5. The intruding object detection device according to claim 2, wherein the intruding object detection unit compares the detected duration time with a reference value of a vibration duration time caused by rainfall. And an intruding object detection device that determines the presence or absence of rainfall based on the comparison result.   The intruding object detection device according to any one of claims 1 to 5, wherein the frequency analysis unit extracts a signal of the predetermined frequency band from the frequency domain signal, and the extraction is performed. An intruding object detection device, wherein the band signal is generated by adding the received signals to each other. The intruding object detection device according to any one of claims 1 to 6,
A vibration sensor that detects the vibration wave and outputs an analog sensor signal;
An intruding object detection device further comprising an A / D converter that converts the analog sensor signal into the sensor signal that is a digital signal.

Images