JP2003156387A - Device and method for detecting abnormal sound - Google Patents

Abstract

PROBLEM TO BE SOLVED: To precisely detect abnormal sounds by an accident, etc., by collecting a sound signal emitted from a running car, and processing it. SOLUTION: The sound of a running car, generated from a sound source (a car being the object), is received by a sound input portion 102, is converted into a digital value by an analog-to-digital converter 103, and is stored in a memory portion 104 after that. The stored data is inputted into an FET processing portion 105 as single-frame data, after the passage of a predetermined time, and hear time frequency axis conversion processing is performed. After that, levels by bands are computed in a data-by-bands computing portion 106. Next, the level for all the bands is computed in an all-band level computing portion 107. After that, variations among a plurality of frames are computed in a plurality of frame data processing portion 108. On the basis of these data, abnormal sound determination is performed in a detection portion 109. Abnormal sound detection information 110 is outputted, when a sound is determined as an abnormal sound, here.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an abnormal sound detection apparatus and method, and more particularly to an abnormal sound detection apparatus and method suitable for use in an apparatus for detecting and identifying an accident such as a traffic accident.

[0002]

2. Description of the Related Art Conventionally, as this kind of technology, Japanese Patent Laid-Open No.
There is an accident sound detection circuit disclosed in No. 42042. As shown in FIG. 9, this accident sound detection circuit has a waveform shaping circuit, an A / D converter, a difference calculation means, and a comparison circuit, and outputs an accident sound from an acoustic signal 91 measured by a microphone. Accident sound detector that detects and outputs the accident sound detection signal 92
93, the accident sound detection signal 92 as a trigger signal, a power spectrum calculation means 94 for obtaining a power spectrum of an acoustic signal, a spectrum pattern calculation means 95 for forming a power spectrum pattern, and the formed spectrum pattern And a pattern comparison circuit 96 that outputs a final accident sound detection signal 98 when both patterns match by comparing the setting pattern set in advance with the pattern generation circuit 97, and is generated from the vehicle. To detect an accident by measuring sound data with a microphone and using the threshold value judgment based on the difference information as the primary judgment, and then comparing and collating the spectrum pattern, autocorrelation pattern, and signal duration with the preregistered pattern. Is.

[0003]

However, among the sounds actually emitted around the road, there are various sounds with a high level, such as construction sounds, airplane sounds, and sounds of birds, beasts, and insects, in addition to accident sounds. Not only is it difficult to distinguish these from accident sounds and abnormal sounds by discrimination based on the difference information based on one input signal, but the accident sounds generated when an actual accident occurs, such as when a person is involved in an accident, The level is not always high compared with sound, and cannot be detected by the above-mentioned conventional technique.

The present invention has been made in view of the above problems, and provides an abnormal sound detection device and an abnormal sound detection method capable of accurately detecting an abnormal sound caused by a traffic accident or the like. .

[0005]

SUMMARY OF THE INVENTION An abnormal sound detecting apparatus of the present invention comprises a band level calculating unit for periodically cutting out an output signal of a sound receiving unit at every specific time and calculating a level for each band, and the band level. An all-band level calculation unit that calculates the level of the entire audible band from the output of the calculation unit, a total-band level calculation unit and a temporal change amount calculation unit that calculates the temporal change amount of the output, and the temporal change amount calculation And an abnormal sound detection unit that detects an abnormal sound from the output of the unit. With this configuration, it is possible to detect a sudden sound generation or a level change that does not occur during normal traveling, and it is possible to detect an abnormal sound.

Further, the abnormal sound detecting device of the present invention is provided with an abnormal sound discriminating section for comparing the abnormal sound data previously stored in the database with the output of the abnormal sound detecting section to discriminate the abnormal sound. Characterize. With this configuration, after the abnormal sound is detected, the abnormal sound is identified by comparing with a database of the abnormal sound prepared in advance.

Further, the abnormal sound detecting apparatus of the present invention is characterized by including an inverse filter for suppressing noise and normal running sound. With this configuration, the influence of ambient noise and normal vehicle running noise can be removed, and the accuracy of abnormal sound detection can be improved.

In the abnormal sound detecting device of the present invention, the sound receiving unit is a microphone array, and the sound receiving unit has a directional sound collecting unit for collecting directional sound in the detection target area using the output of the microphone array. Characterize. With this configuration, the sound in the detection target area is efficiently collected, and the influence of external noise is reduced, so that the abnormal sound can be detected with high accuracy.

In the abnormal sound detection method of the present invention, the output signal of the sound receiving unit is periodically cut out at every specific time to calculate the level of each band, and the level of the entire audible band is calculated from the level of each band. , Calculating the amount of temporal change in the level of the entire audible band, and detecting an abnormal sound from the amount of temporal change. With this configuration, it is possible to detect a sudden sound generation or a level change that does not occur during normal traveling, and it is possible to detect an abnormal sound.

[0010]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the drawings.

(First Embodiment) FIG. 1 shows a first embodiment of the present invention.
6 is a flowchart of abnormal sound detection processing in the embodiment of FIG. As shown in FIG. 1, when the abnormal sound detection process is started, the output of the sound receiving unit (microphone) 101 is accumulated in the memory unit 104 via the sound input unit 102 and the A / D conversion unit 103.
After the time set as the time window in advance has elapsed, the data accumulated at that time is taken as one frame of data,
FFT (Fast Fourier Transform) processing section 105 FFT
Processing is performed, and time axis data is converted to frequency axis data. The data converted here is subjected to level calculation for each band by the band-specific data calculation unit 106. Here, for example, band pass filter processing or band level calculation for each 1/3 octave band may be performed. Further, the full band level calculation unit 107 calculates the level in all bands. Next, in the multi-frame data processing unit 108,
The amount of change in bandwidth level between multiple frames and the amount of change in all bandwidth level between multiple frames are calculated. After that, in the detection unit 109, the threshold processing of the band level and the threshold processing of the entire band level are performed, and if any of the levels exceeds the threshold, the flag is set to 1. Then, after repeating the processing up to this point a certain number of times, if the number of continuations of the flag 1 exceeds the threshold value, it is determined as an abnormal sound,
The abnormal sound detection signal 110 is output.

At this time, the sound collection unit 101 or the sound input unit 102
To the memory unit 104 and the FFT processing unit 105,
The processing of the abnormal sound detection unit 111 may be sequentially performed, and after the processing of abnormal sound detection is performed, the routine from the memory unit 104 to the input of the A / D conversion unit 103 may be performed to reduce the processing load. Alternatively, processing from the sound input unit 102 to the memory unit 104,
The processing of the FFT processing unit 105 and the abnormal sound detection unit 111 may be performed in parallel to improve the processing speed.

As described above, according to the first embodiment of the present invention, the output signal of the microphone is periodically cut out by using the time windows, and the sound pressure in the band level of each time window data and the entire audible band. Calculate the level, and then calculate the amount of change over time in multiple frames of these data,
Next, by finding the amount of change over time in these parameters,
It becomes possible to detect sudden sound generation and level change that do not occur during normal running, and abnormal noise can be detected.

(Second Embodiment) In the second embodiment, after the abnormal sound is detected, it is possible to identify the abnormal sound by comparing with a database of the abnormal sound prepared in advance. Is.

FIG. 2 is a flow chart of abnormal sound detection / identification processing according to the second embodiment of the present invention. In this figure, the sound receiving unit 201 to the FFT processing unit 205 and the abnormal sound detecting unit 213 are equivalent to the processing units of the same names in the first embodiment, respectively.

In the present embodiment, after the abnormal sound is detected by the detection unit 209, the abnormal sound identification unit 210 performs comparison with the band level accumulated in the accident type sound database 212 in advance and pattern matching. Thus, the abnormal sound is identified and the abnormal sound detection identification information 211 is output.

Here, in the database 212, in addition to vehicle running sounds, data relating to sounds other than normal running, such as accident sounds, emergency vehicle sounds, birds and animals crying, construction sounds, airplane sounds, etc., are stored. . Further, as a method of identification, a method of performing all-band level, a method of discriminating and splitting in which a plurality of frame processing parameters are input, a method based on linear prediction analysis, a spectrum shape comparison by LPC cepstrum, and the like may be used. An abnormal sound can be identified based on this identification result.

As described above, according to the second embodiment of the present invention, after the abnormal sound is detected, it is compared with the centroid value or the spectrum pattern for each type existing in the database of the abnormal sound prepared in advance. By doing so, the abnormal sound can be identified.

(Third Embodiment) The third embodiment is provided with an inverse filter for noise suppression for eliminating the influence of ambient noise and normal vehicle running noise, thereby making it possible to detect abnormal sounds accurately. It is a heightened one.

FIG. 3 is a flow chart of abnormal sound detection / identification processing according to the third embodiment of the present invention. In this figure, from the sound receiving unit 301 to the FFT processing unit 305, an abnormal sound detecting unit
The process 314 and the abnormal sound identification unit 311 are respectively the same as the processes of the same name in the second embodiment.

In the present embodiment, the filter unit 306 is provided before the abnormal sound detector 314. In this filter unit 306,
Performs active noise suppression processing by inverse filter processing.

As a concrete inverse filter processing, as shown in FIG. 4, the output from the FFT processing section is used as an input signal x (n),
By performing filtering processing on the input signal x (n) using the approximate inverse filter 401, noise and normal running noise are suppressed, and the noise suppression signal y (n) represented by the following formula [1] is obtained. Output.

Y (n) = H (z) · x (n) Equation [1]

Here, the transfer function H of the recent inverse filter 401
(z) is dynamically updated by the noise data defined by the delay amount z (-m) in the delay circuit 403 and the output n (k) of the vehicle traffic noise detection circuit 402. Therefore, by performing feedback with the spectrum average data of the acoustic signals of the past several frames received by the microphone, dynamically generating an inverse filter, and performing convolution with the received signal, background noise and vehicle noise from the received signal can be obtained. Since it is possible to reduce the normal traveling sound of the vehicle and to make the abnormal sound and the accident sound that suddenly occur visible, it is possible to detect the abnormal sound with higher accuracy in the abnormal sound detection unit 314 in the subsequent stage. Is possible.

As described above, according to the third embodiment of the present invention, by providing the noise suppression inverse filter for eliminating the influence of the ambient noise and the normal vehicle running noise,
It is possible to improve the S / N as an abnormal sound and improve the abnormal sound detection accuracy.

(Fourth Embodiment) In the fourth embodiment, the sound in the target area is efficiently collected by carrying out the directional sound collection, and the influence of the external noise is reduced, so that This enables accurate abnormal sound detection.

FIG. 5 is a flow chart of abnormal sound detection / identification processing according to the fourth embodiment of the present invention.

In the present embodiment, the sound receiving unit 501 is a microphone array in which a plurality of microphones are arranged in a one-dimensional line shape or a two-dimensional surface shape. Further, it is appropriate that the number of microphones constituting the sound receiving unit 501 is about 3 to 16, and the frequency band of the abnormal sound is taken into consideration, and the interval between the microphones is set in consideration of the target wavelength λ. Here, about 3 to 10 cm is suitable.

FIG. 6 is a diagram showing an installation example of the microphone array. The microphone array 601 is arranged on the road 602 side so that the center of the constituent surface of the microphone array 601 faces the center of the measurement target area.
It is installed as shown in the elevation view of FIG. 6B and the plan view of FIG. Here, an angle formed by a straight line connecting the center of the constituent surface of the microphone array 601 and the center of the road with a vertical line passing through the center of the constituent surface of the microphone array 601, is defined by the vertical line and the constituent surface of the microphone array 601. The angle is θ,
The angle deviation from a straight line passing through the center of the constituent surface of the microphone array 601 and extending along the road surface of the road 617 is β.

The output from the sound receiving unit 501 configured as described above is input to the sound input unit 502, sampled for each microphone by the AD conversion unit 503, and further stored individually in the memory unit 504. It These data are stored in the FFT processing unit 505 after the preset frame time has elapsed.
Is subjected to FFT processing, and then the directivity control unit 51
Coefficient convolution unit 506 in 5 performs coefficient convolution and synchronous addition unit
Synchronous addition is performed in 507 to collect directional sound in the target area.

Here, as a means for realizing directional sound collection, a method of utilizing a time difference between microphones such as a delay sum array may be used, but if it is necessary to provide more directivity, a filter such as an LMS array is used. You may use the method of comprising.

FIG. 7 shows a configuration example of the LMS array. LM
The S-array assumes a plurality of sound source directions in space, and calculates a filter coefficient capable of collecting only a desired direction among them. The transfer function between N sound sources and M microphones is represented as Cn, m. Focusing only on the delay time of the direct sound, Cn, m can be expressed as the following formula [2].

Cn, m = е ^{-jω {DO- (m-1) τn}} Equation [2]

Here,   Dm = D0- (m-1) τn (m = 1,2, ..., M: number of microphones) Equation [3]   τn = (dsinθn) / с Equation [4] Is.

D0 is a preset fixed delay amount, с
Is the speed of sound. At this time, Dm represents the delay amount in which the phases match when the sound arrives from the desired direction θn (the sound source direction for which the sensitivity is desired to be improved is θn).

When a matrix having Cn, m as an element is C and a row vector having a filter coefficient Xm as an element is X, a row vector Y having a response Yn of each sound source at the output as an element is given by the following equation [5]. It is represented by.

CX = Y Equation [5]

The equation [5] is a simultaneous equation. For example, Y is set to 0 in directions other than the desired direction, and the least squares solution of X satisfying this is obtained by the normal equation of the following equation [6]. It is possible to calculate the filter coefficient that directs the direction.

X = (C ^H C + βI) C ^H Y Equation (6)

In this equation, H represents a transposed matrix, and I
Represents an identity matrix. Here, by setting Y that directivity is directed to the target area and calculating the filter coefficient X,
Since directional sound collection is possible, it is possible to improve the S / N ratio of abnormal sound generated in the target area and reduce the influence of noise generated in the surroundings. The operation of the abnormal sound detecting unit 516 and the abnormal sound identifying unit 512 in the latter stage will be described in the second section.
It is equivalent to the part of the same name in the embodiment of.

As described above, according to the fourth embodiment of the present invention, the sound receiving unit has a microphone array structure including a plurality of microphones, and by performing directional sound collection, Highly accurate abnormal sound detection is possible by efficiently collecting sounds and reducing the influence of external noise.

(Fifth Embodiment) The fifth embodiment is an integration of the first to fourth embodiments.
FIG. 8 shows the abnormal sound detection and the abnormal sound detection in the fifth embodiment of the present invention.
It is a flowchart of an identification process.

In the present embodiment, the filter unit 806 can suppress noise and normal running sound, as in the filter unit 306 of the third embodiment, and the directional sound collecting unit 81 can be used.
As in the case of the directional sound collecting unit 515 of the fourth embodiment, the sound collection targeting the target area can be performed by the 6, and the abnormal sound detecting unit 817 allows the abnormal sound detecting unit 111 of the first to fourth embodiments, 213,
Similar to 314 and 516, abnormal sound can be detected by processing band level data, all band level data, and multiple frame data, and the abnormal sound identifying unit 813 can detect abnormal sounds in the second to fourth embodiments. Similar to the sound identifying sections 210, 311, 512, it is possible to identify abnormal sounds with high accuracy.

In the present embodiment, the sound receiving section (microphone array) 80 receives the audible band sound generated from the sound source (target vehicle).
The sound input unit 802 receives the analog electric signal output from the sound receiving unit 801 by the sound input unit 802, and after AD conversion by the AD conversion unit 803,
The time window cutout unit sequentially cuts out every specific time window. The cut-out data is Fourier-transformed by the FFT processing unit 805, the ambient noise and the normal running sound are dynamically suppressed by the filter unit 806, and then the sound generated in the target area is extracted by the directional sound collecting unit 816. Then, each parameter is sent to the abnormal sound detector 817. Then, the abnormal sound detection unit 817 makes an abnormal sound detection determination from these parameters, and if the abnormal sound is determined, the abnormal sound identification unit 813 identifies the abnormal sound. By integrating and outputting these pieces of information, it is possible to detect an actual abnormal sound generation event and specify an abnormal sound type.

[0045]

As described above, according to the present invention, the processing by the temporal change amount of the acoustic signal level in the entire audible band, the dynamic suppression of the ambient noise and the normal running sound by the inverse filter processing, and the target area by the directional sound collection are performed. It is possible to provide an abnormal sound detection apparatus and method having an excellent effect that it is possible to accurately detect abnormal sounds and accidents by performing limited collection of sounds.

[Brief description of drawings]

FIG. 1 is a flow chart of abnormal sound detection processing according to the first embodiment of the present invention,

FIG. 2 is a flowchart of abnormal sound detection / identification processing according to the second embodiment of the present invention;

FIG. 3 is a flowchart of abnormal sound detection / identification processing according to the third embodiment of the present invention;

FIG. 4 is a diagram showing a configuration example of an inverse filter according to a third embodiment of the present invention,

FIG. 5 is a flow chart of abnormal sound detection / identification processing according to the fourth embodiment of the present invention;

FIG. 6 is an installation diagram of an abnormal sound detection / identification method according to a fourth embodiment of the present invention;

FIG. 7 is a diagram showing a configuration of a directivity control unit according to a fourth embodiment of the present invention,

FIG. 8 is a flowchart of abnormal sound detection / identification processing according to the fifth embodiment of the present invention;

FIG. 9 is a block diagram of a conventional accident sound detection circuit.

[Explanation of symbols]

101, 201, 301, 501, 801 Sound receiving part 106, 206, 307, 508, 809 Band-specific data calculator 107, 207, 308, 509, 810 Total bandwidth level calculation unit 108, 208, 309, 510, 811 Multiple frame data processing unit 109, 209, 310, 511, 812 Detector 210, 311, 512, 813 Abnormal sound identification section 212, 313, 514, 815 Abnormal sound database 306, 806 Filter section 515, 816 Directional sound collector

   ─────────────────────────────────────────────────── ─── Continued front page    F-term (reference) 2G064 AB16 AB21 BA05 BD02 CC06                       CC13 CC42 CC43 CC46 CC52                       CC61                 5D015 DD02

Claims (5)

[Claims] 1. A band level calculation unit that periodically cuts out an output signal of a sound collection unit for each specific time and calculates a level for each band, and a level of the entire audible band is calculated from an output of the band level calculation unit. Full band level calculation unit, temporal change amount calculation unit that calculates the temporal change amount of the output of the full band level calculation unit, and abnormal sound detection unit that detects an abnormal sound from the output of the temporal change amount calculation unit An abnormal sound detection device comprising: 2. The abnormal sound identifying section for comparing the abnormal sound data stored in advance in a database with the output of the abnormal sound detecting section to identify the abnormal sound. Sound detector. 3. The abnormal sound detection device according to claim 1, further comprising an inverse filter for suppressing noise and normal running sound. 4. The sound receiving unit is a microphone array, and has a directional sound collecting unit that collects directional sound in a detection target area using an output of the microphone array. Abnormal sound detection device. 5. The level of the entire audible band is calculated by periodically cutting out the output signal of the sound receiving unit at specific time intervals to calculate the level of each band, and calculating the level of the entire audible band from the level of each band. Is calculated, and an abnormal sound is detected from the temporal change amount.