close

Вход

Забыли?

вход по аккаунту

?

JP2010008474

код для вставкиСкачать
Patent Translate
Powered by EPO and Google
Notice
This translation is machine-generated. It cannot be guaranteed that it is intelligible, accurate,
complete, reliable or fit for specific purposes. Critical decisions, such as commercially relevant or
financial decisions, should not be based on machine-translation output.
DESCRIPTION JP2010008474
PROBLEM TO BE SOLVED: To consider that a non-daily sound is a dangerous signal that does not
occur regularly, detect that sound, and further detect the direction of the non-daily sound source,
that is, the direction of the sound source generated from a place where it does not normally
occur. By simultaneously detecting information, a system capable of watching over public space
can be provided from these two pieces of information. The system comprises a sound
observation means for observing a sound, an extraordinary sound detection means for detecting
whether the observation sound is an extraordinary sound different from the ordinary sound, and
a sound source direction of the observation sound is an extraordinary sound source direction
And an unusual sound source detection means for detecting whether or not it is detected.
[Selected figure] Figure 1
Unusual sound detection system
[0001]
The present invention relates to a system capable of detecting an unusual sound in a space to be
detected and also its sound source direction.
[0002]
Many of conventional systems for detecting abnormal sound or dangerous sound in a sound
environment are (1) those in which an abnormal sound to be extracted is defined in advance and
the defined abnormal sound is detected (for example, Patent Document 1), (2) A normal sound is
defined in advance, and it can be roughly divided into a case where it is determined that an
abnormal event has occurred when a sound other than the sound is generated.
04-05-2019
1
[0003]
However, the problems of (1) and (2) lie in the fact that data of sounds and normal sounds that
are desired to be detected in advance must be collected.
Furthermore, using the collected data, it is also necessary to do the troublesome work of
modeling the sound and the normal sound that you want to detect.
Also, as a conventional method that does not perform modeling, there is a system that
determines a sudden sound as an abnormal sound and detects only that sound, but since it reacts
each time a sudden sound is generated, an erroneous detection occurs. The possibilities are quite
large. JP 2007-104546 A
[0004]
The present invention has been made in view of the above circumstances, and considers that an
extraordinary sound is a dangerous signal that does not occur regularly, detects that sound, and
further determines the direction of an unusual sound source, that is, It aims at providing the
system which can watch public space from these two information by detecting simultaneously
the information on the direction of the sound source generated from the place which does not
usually occur.
[0005]
In order to solve the above problems, the present invention first provides: sound observation
means for observing a sound; and unusual sound detection means for detecting whether the
observation sound is an unusual sound different from the ordinary sound An unusual sound
detection system comprising: an extraordinary sound source direction detecting means for
detecting whether or not the sound source direction is an extraordinary sound source direction
when the observation sound is an extraordinary sound.
[0006]
Second, the non-daily sound detection means is a means for classifying the observation sound
according to the size of its power spectrum, and a means for calculating the similarity to all the
past sounds classified into the same class with respect to the classified sound And means for
determining whether or not the observation sound is an extraordinary sound based on the
04-05-2019
2
calculation similarity.
[0007]
Third, the above-mentioned unusual sound detection system is provided, further comprising a
database for accumulating observation sounds.
[0008]
Fourth, when the observation sound is determined to be similar to a certain extraordinary sound
in the database, the characteristic sound representing the extraordinary sound is updated using
the feature amount of the observation sound, the said extraordinary sound detection system I will
provide a.
[0009]
Fifthly, the non-daily sound source direction detecting means comprises means for calculating a
value used to estimate the direction of the sound source from the observation sound, means for
calculating the similarity between the calculated value and the previous sound source direction
estimation data, Means for determining whether the sound source direction of the observation
sound is an extraordinary sound source direction or not based on the degree of similarity.
[0010]
Sixth, the above-mentioned unusual sound detection system is provided, further comprising a
database for accumulating sound source direction data of observation sound.
[0011]
Seventh, when it is determined that the sound source direction of the observation sound is
similar to the non-daily sound source direction in the database, the feature amount related to the
non-daily sound source direction is updated using the feature amount of the sound source
direction of the observation sound The non-ordinary sound detection system is provided.
[0012]
Eighth, the non-everyday sound detection system is provided, further comprising a warning unit
that issues a warning when one or both of the non-daily sound and the non-daily sound source
direction are determined.
04-05-2019
3
[0013]
The present invention further solves the above-mentioned problems. Ninth, the sound is observed
to detect whether the observation sound is an extraordinary sound different from the ordinary
sound, and the observation sound is an unusual sound. A non-daily sound detection method is
provided that detects whether or not the sound source direction is a non-daily sound source
direction in some cases.
[0014]
Tenth, in the detection of extraordinary sounds, classify the observation sound according to the
power spectrum size, calculate the similarity with all sounds in the past classified into the same
class with respect to the classified sound, and calculate the calculation similarity It is determined
whether the observation sound is an extraordinary sound based on the above.
[0015]
Eleventhly, in the detection of the non-daily sound source direction, calculate the value used to
estimate the direction of the sound source from the observation sound, calculate the similarity
with the past sound source direction estimation data for the calculated value, and based on the
calculated similarity The non-daily sound detection method is provided to determine whether or
not the sound source direction of the observation sound is the non-daily sound source direction.
[0016]
Twelfth, the non-daily sound detection method is provided, wherein a warning is issued when
one or both of the non-daily sound and the non-daily sound source direction are determined.
[0017]
Thirteenth, the present invention provides an unusual sound detection program for causing a
computer to execute the unusual sound detection method.
[0018]
Fourteenthly, the present invention provides a computer readable recording medium having the
unusual sound detection program recorded thereon.
[0019]
According to the present invention as described above, it is possible to detect whether it is an
everyday sound or an extraordinary sound and whether it is an everyday sound source direction
04-05-2019
4
or an unusual sound source direction through daily sound environment measurement, and (2)
Since a database of daily sounds and daily sound source directions is constructed through
environmental sound measurement, definitions of daily sounds, unusual sounds, daily sound
source directions, and unusual sound source directions become unnecessary in advance, and (3)
stored in the database Daily sound, daily sound source direction feature amount is updated with
newly measured daily sound, daily sound source direction feature amount, and as a result, using
the updated feature amount, the daily sound and daily sound source direction Detection accuracy
can be increased.
[0020]
Hereinafter, embodiments of the present invention will be described with reference to the
drawings.
First Embodiment FIG. 1 is a flowchart of each process executed by the system of the present
invention.
[0021]
First, the sound in the space is observed by a sensor installed in the monitoring target space, and
the observation sound is subjected to sound source localization processing after passing through
an A / D converter or the like as necessary.
Sound source localization is performed using the equation in FIG.
[0022]
More specifically, the microphone array illustrated in FIGS. 2A and 2B can be used as the sound
observation sensor, for example.
It goes without saying that other sensors may be used as long as the unusual sound and unusual
sound source direction detection according to the present invention can be realized.
04-05-2019
5
The illustrated microphone array is a cross-shaped microphone array as shown in FIG. 2 (a), in
which one microphone is installed on each of o, x, y and z axes in a cube-shaped rigid housing as
shown in FIG. 2 (b). The four rigid frames are attached to one another, which are installed near
the ceiling of the target space, for example.
[0023]
The signals observed by the o, x, y and z microphones at each microphone are
[0024]
[0025]
The observation signal data can be obtained from a total of 16 microphones.
[0026]
The signal observed by the microphone array is
[0027]
[0028]
By taking the difference between o and x, y, z.
However, the same value is used for o.
[0029]
The illustrated microphone array is characterized by source localization when the signal from the
signal source is a low frequency signal.
[0030]
04-05-2019
6
In the above equation, the input / output relationship of the observation signal y (t) (all
observation signals are handled in the frequency domain) is as follows.
[0031]
[0032]
Transforming the above equation into the relationship between the signal just before reaching
the microphone and the observed signal,
[0033]
[0034]
となる。
[0035]
Based on the observation sound data from the microphone array as described above,
[0036]
[0037]
If the calculated Lxl (θij) is larger than the predetermined threshold a, sound source localization
is performed, and it is regarded that there is a sound source, and detection of extraordinary
sound and unusual sound source is performed for sound and sound source information from that
sound source. Do.
[0038]
In the above equation,
[0039]
[0040]
04-05-2019
7
である。
However, since it can not be determined directly, the following equation is used.
[0041]
[0042]
The non-daily sound detection will be described in more detail below with reference to FIG.
[0043]
First, the detected sound is classified according to the size of the histogram of the power
spectrum as the pre-processing for detecting the extraordinary sound (step S31).
Although it is possible to detect unusual sounds without classification, since it takes a
considerable amount of time to calculate the degree of similarity, which will be described later,
classification is performed here in order to significantly reduce the execution time of the degree
of similarity calculation. Do.
[0044]
FIG. 4 is an example of a histogram of the power spectrum of the detected sound.
For example, three levels are set according to the size of dm.
a.dm>I1 b.
I1>=dm>I2 c.
04-05-2019
8
I2> = dm> I3 Classification is made into three classes according to the number of dm in each of a,
b and c, and the non-daily sound is detected in each class.
Detection sounds are accumulated for each of these classes in a sound type DB to be described
later.
[0045]
Next, for each class in the sound type DB, the feature amounts of the past detected sound
(everyday sound and non-everyday sound) and the current detected sound are calculated (step
S32).
As feature quantities, for example, MFCC (Mel-Frequency Cepstral Coefficients) used for speech
recognition or the like, Fluctuation Pattern (FP) for analyzing the loudness of each frequency
band, etc. can be used. The type of feature amount is not particularly limited as long as the
difference from the ordinary sound appears in the calculated feature amount and both can be
clearly distinguished.
[0046]
Subsequently, based on the feature amount of the past detected sound and the calculated feature
amount of the present detected sound in the sound type DB, the similarity αi of the present
detected sound and the present detected sound is calculated (step S33).
The similarity is calculated for feature amounts of all sounds detected in the past.
[0047]
Then, if the degree of similarity is greater than the predetermined threshold b with respect to the
past detected sound showing the closest degree of similarity, it is determined that the detected
04-05-2019
9
sound has a sound similar to the past, that is, it is an everyday sound ( Step S34-Yes).
Conversely, if the similarity is equal to or less than the threshold b, it is determined that the
sound is an extraordinary sound (step S34-No).
[0048]
The daily sounds and the non-daily sounds determined as described above are stored for each
class in the sound type DB together with the calculated feature amounts.
[0049]
The non-daily sound source detection will be described in more detail below with reference to
FIG.
[0050]
First, the feature quantity of the detected sound source direction and the feature quantity for the
data of the past sound source direction in the sound source direction DB are calculated (step
S51).
The feature quantities to be calculated are as follows.
An eigenvector for an absolute maximum eigenvalue of a matrix having Lxl (θij) as an element.
The average value of Lxl (θij) calculated for i, j.
[0051]
Next, based on the feature quantity in the sound source direction DB and the calculated feature
quantity, the similarity between the past sound source direction data and the detected sound
source direction is calculated (step S52).
04-05-2019
10
The similarity βi relates to the eigenvectors, and the similarity γi relates to the Lxl (θij)
average value.
[0052]
To further describe the eigenvectors and Lxl (θij) average values as feature quantities, the values
of each square in the sound source localization processing data (see, eg, FIG. 6) are regarded as
elements of a matrix, and eigenvalue decomposition of the matrix is performed.
At this time, among the obtained eigenvalues, the eigenvector for the eigenvalue having the
largest absolute value is regarded as the feature quantity, and the similarity βi with all the
feature quantities calculated in the past, ie, the eigenvector is calculated.
FIG. 6 shows an example of the result data of the sound source localization processing, and in
this example, the yellow to brown squares indicate the direction of the sound source.
The eigenvector is such that the value of the vector changes as the direction of the sound source
changes.
That is, if the direction of the sound source changes, that is, if the position of the brown square in
FIG. 6 changes, the value of each square also changes, and the value of the matrix changes, so the
value of the eigenvector also changes.
Focusing on this change, the eigenvector for the eigenvalue with the largest absolute value is
regarded as the feature amount that is the basis of the sound source direction detection.
Further, the similarity γi is calculated for the detected sound source position and all the sound
source positions detected in the past with respect to Lxl (θij) of Expression 5.
[0053]
04-05-2019
11
Then, if the sum of the similarities βi and γi is larger than the predetermined threshold c with
respect to the past sound source direction showing the closest similarity, the detected sound
source direction has a sound source direction similar to the past, ie It is determined that it is a
daily sound source (step S53-Yes).
Conversely, if the similarity is equal to or less than the threshold c, it is determined that the
sound source is an extraordinary sound source (step S53-No).
Here, by using two feature quantities / similarities, it is possible to prevent the detection
omission of the unusual sound source direction and improve the detection accuracy.
[0054]
The daily sound source direction and the non-daily sound source direction determined as
described above are stored in the sound source direction DB together with the calculated feature
amounts.
[0055]
After that, as illustrated in FIG. 7, for example, “Warning!” Warns when either or both of the
extraordinary sound and the extraordinary sound source direction are detected.
Output warning such as signs.
[0056]
[Device Configuration] Each processing in the present invention as described above can be
executed by a device having the overall configuration illustrated in FIG.
The apparatus 1 of FIG. 8 includes a processing unit (CPU) 10, a storage unit (internal memory,
external memory) 11, an input unit (keyboard, mouse) 12, an output unit (monitor, printer) 13
and a communication control unit 14 in a bus 15. Are connected.
04-05-2019
12
The device 11 is connected to the network 16 via the communication control unit 14.
It is possible to obtain various data necessary for each process stored in another terminal or the
like from a terminal such as the device 1 connected to the network 16.
The processing unit 10 executes the various processes described above by reading the program
stored in the storage unit 11.
[0057]
The figure which showed the functional block and process flow for demonstrating this invention.
FIG. 2 is a perspective view showing an example of a microphone array.
The figure which showed an example of the unusual sound detection flow.
The figure which showed an example of the histogram of the power spectrum of a detection
sound.
The figure which showed an example of the unusual sound source detection flow.
The figure which showed an example of sound source localization processing data. The figure
which showed an example of the warning flow. The figure which showed the system block for
demonstrating this invention.
04-05-2019
13
Документ
Категория
Без категории
Просмотров
0
Размер файла
21 Кб
Теги
jp2010008474
1/--страниц
Пожаловаться на содержимое документа