JP2003214942A - Accident sound detector, method and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an accident sound detector, a method and a program capable of improving accuracy for detecting an accident sound, reducing erroneous detection, determining a type of the detected accident sound, and reducing the erroneous detection occurring depending on an arrival direction. <P>SOLUTION: This detector is composed of a means 120 for converting an analog acoustic signal detected by a microphone to a digital signal of frame configuration, a means 130 for generating information on broadband power of the analog acoustic signal for each frame based on the digital signal, a means 143 for calculating a means value in the broadband power for each frame among frames commencing from the frame as a calculation object for the accident sound back to the past frame through prescribed number of frames, and a means 144 for determining if the signal of the frame as the calculation object for the accident sound is the signal corresponding to the accident sound or not based on information on the broadband power and the mean value of the broadband power. <P>COPYRIGHT: (C)2003,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an accident sound detection device, method, and program for detecting accident sound caused by a traffic accident or the like.

[0002]

2. Description of the Related Art FIG. 11 is a block diagram showing a conventional accident sound detecting apparatus 1100 disclosed in Japanese Patent Laid-Open No. 10-142042. The accident sound detection device 1100 is
A waveform shaping circuit 1102 for shaping the waveform of the acoustic signal input to the acoustic signal input terminal 1101, and an AD converter 110 for converting the acoustic signal after the waveform shaping into a digital signal.
3. Signal smoothing means 1104 for smoothing the converted digital signal, difference calculating means 1105 for calculating the difference in the signal level in the time direction, and comparing the digital signal after the difference processing with a predetermined reference value. Comparing circuit 1 for
106, and a memory 1 for comparing the above reference values
It is composed of 107.

Next, the operation of the conventional accident sound detecting apparatus 1100 will be described. The acoustic signal recorded near the road or the like is sent to the waveform shaping circuit 1 via the acoustic signal input terminal 1101.
The signal is input to 102 and unnecessary signal components are removed. The output from the waveform shaping circuit 1102 is converted from an analog signal to a digital signal by the AD converter 1103. AD
The digital signal output from the converter 1103 is input to the signal smoothing unit 1104, and the signal smoothing unit 1104.
Is smoothed by a moving average circuit or a low-pass filter for the purpose of preventing a rapid level change caused by an unnecessary extraneous signal.

The output of the signal smoothing means 1104 is input to the difference calculating means 1105, and the difference value between the level of the present signal and the level of the signal past in the fixed time is calculated. The comparison circuit 1106 outputs an accident sound detection signal from the accident sound detection output terminal 1108 when the difference value is larger than a reference value which is a fixed value stored in the memory 1107.
With these, an accident sound accompanied by a sudden change in sound pressure can be detected.

[0005]

However, in the conventional accident sound detecting apparatus, since the time response of the accident sound greatly differs depending on the kind of accident sound, there is a problem that the detection accuracy is lowered depending on the kind of accident sound. It was Further, among the types of accident sounds, the sudden braking sound and the horn sound do not always have a sudden level change, so that there is a problem that the detection accuracy is lowered. In addition, there is a problem that erroneous detection occurs due to the siren sound of the emergency vehicle. Also,
There was a problem that the type of accident sound detected was unknown. Further, there is a problem that erroneous detection occurs due to a sound coming from a direction other than the desired direction.

The present invention has been made in order to solve such a problem and has been detected by improving the accuracy of detecting an accident sound in an accident sound detecting device and reducing the false detection due to the siren sound of an emergency vehicle. It is an object of the present invention to provide an accident sound detection device, method, and program capable of determining the type of accident sound and further reducing erroneous detection caused by sounds coming from directions other than the desired direction.

[0007]

The accident sound detecting apparatus of the present invention comprises a converting means for converting an analog acoustic signal detected by a microphone into a digital signal composed of a frame, and the frame based on the digital signal. For each of the above, the wide band power data generating means for generating information about the wide band power which is the signal level of the signal component occupying the predetermined band of the analog acoustic signal, and each frame from the frame of the accident sound calculation target to a predetermined number of past frames The average value calculating means for calculating the average value of the wideband power obtained for, the signal of the frame of the accident sound calculation target based on the information of the wideband power and the information of the average value of the wideband power to the accident sound And a determination means for determining whether or not the signal is a corresponding signal. With this configuration, instead of simply detecting the accident sound based on the power difference value between samples, the accident sound power is estimated using multiple frames and compared with the average power immediately before that It is possible to realize an accident sound detection device capable of accurately detecting accident sounds of different times.

Further, the accident sound detecting apparatus of the present invention is a maximum value calculation for calculating the maximum value of the wide band power obtained for each frame from the frame of the accident sound calculation to a frame of a predetermined number of past frames. Means, a difference value calculation means for calculating a difference value between the maximum value of the wideband power and the average value of the wideband power, the determination means, the determination of the accident sound calculation target based on the information of the difference value. It has a configuration for determining whether or not the signal of the frame corresponds to the accident sound. With this configuration, instead of simply detecting the accident sound based on the power difference value between samples, the accident sound power is estimated using multiple frames and compared with the average power immediately before that It is possible to realize an accident sound detection device capable of accurately detecting accident sounds of different times.

Further, the accident sound detecting apparatus of the present invention comprises total value calculating means for calculating a total value of the wide band power obtained for each frame from the accident sound calculation target frame to a predetermined number of past frames. , A difference value calculating means for calculating a difference value between the total value of the wideband power and the average value of the wideband power, the determining means, of the frame of the accident sound calculation target based on the information of the difference value. It has a configuration for determining whether or not the signal corresponds to the accident sound. With this configuration, instead of simply detecting the accident sound based on the power difference value between samples, the accident sound power is estimated using multiple frames and compared with the average power immediately before that It is possible to realize an accident sound detection device capable of accurately detecting accident sounds of different times.

Further, the accident sound detecting apparatus of the present invention is a narrow band for calculating a narrow band power which is a signal level of a signal component occupying any frequency band in the audible frequency band of the analog acoustic signal detected by the microphone. A power data generation unit, a narrowband property detection unit that detects narrowband property based on the wideband power information and the narrowband power information, and a signal of the frame of the accident sound calculation target by the determination unit is an accident sound. When it is determined that the signal corresponding to the, the narrowband characteristic detection means for classifying the accident sound based on the information whether or not the narrowband property is detected, There is.
With this configuration, instead of simply detecting the accident sound based on the power difference value between samples, the accident sound power is estimated using multiple frames and compared with the average power immediately before that Accident sounds with different times can also be detected accurately, and accident sounds can be detected based on wideband acoustic signals and narrowband acoustic signals.
To evaluate the duration of a given narrowband acoustic signal,
It is possible to realize an accident sound detection device that can accurately detect accident sounds with different rise times.

Further, in the accident sound detecting apparatus of the present invention, the narrow band characteristic detecting means calculates a difference value between the wide band power and the narrow band power, and a difference value between the wide band power and the narrow band power. Has a configuration for detecting the narrow band property based on. With this configuration, instead of simply detecting the accident sound based on the difference value of the power between the samples,
By estimating the accident sound power using multiple frames and comparing it with the average power immediately before that, it is possible to detect even accident sounds with different rise times with high accuracy, and a wideband acoustic signal and a narrowband acoustic signal can be detected. It is possible to realize an accident sound detection device that can accurately detect even accident sounds with different rise times because the accident sound is detected based on the above and the duration of a predetermined narrowband acoustic signal is evaluated. it can.

Further, in the accident sound detecting apparatus of the present invention, any frequency band in the audible frequency band for which the narrow band power data generating means calculates the narrow band power is a frequency of 800 Hz or more and 2 kHz or less. It has a configuration that is a frequency band of any range in the range. With this configuration, instead of simply detecting the accident sound based on the power difference value between samples, the accident sound power is estimated using multiple frames and compared with the average power immediately before that Accident sounds with different times can be detected with high accuracy.The accident sound is detected based on the wideband acoustic signal and the narrowband acoustic signal, and the rising time is evaluated to evaluate the duration of the predetermined narrowband acoustic signal. It is possible to realize an accident sound detection device capable of accurately detecting accident sounds of different times.

Further, the accident sound detecting apparatus of the present invention comprises a periodicity calculating means for calculating a cycle of the time change of the narrow band power, and the classifying means is arranged to judge the frame of the accident sound calculation target by the judging means. When the signal is determined to be a signal corresponding to an accident sound, and the narrow band property is detected by the narrow band property detecting unit, if the period is not calculated by the periodicity calculating unit, the signal is classified as an accident sound and the periodicity is When calculated, it has a configuration that is not classified as an accident sound. With this configuration, instead of simply detecting the accident sound based on the power difference value between samples, the accident sound power is estimated using multiple frames and compared with the average power immediately before that Accident sounds with different times can be detected with high accuracy.The accident sound is detected based on the wideband acoustic signal and the narrowband acoustic signal, and the rising time is evaluated to evaluate the duration of the predetermined narrowband acoustic signal. Accident sounds with different times can be detected with high accuracy. Accident sounds are detected based on wideband and narrowband acoustic signals, and the periodicity of narrowband acoustic signals is evaluated. It is possible to realize an accident sound detection device that can detect siren sounds and the like and can be distinguished from sudden braking sounds, horn sounds, and the like.

Further, the accident sound detecting apparatus of the present invention is for calculating the narrow band power which is the signal level of the signal component occupying any frequency band in the audible frequency band of the analog acoustic signal detected by the microphone. Narrowband power data generating means, periodicity calculating means for calculating the cycle of the time change of the narrowband power, and the determining means determines that the signal of the accident sound calculation target frame corresponds to the accident sound. If the periodicity is not calculated by the periodicity calculation means, it is regarded as an accident sound, and if the periodicity is calculated, it is not classified as an accident sound. There is.
With this configuration, instead of simply detecting the accident sound based on the power difference value between samples, the accident sound power is estimated using multiple frames and compared with the average power immediately before that Accident sounds with different times can also be detected accurately, and accident sounds can be detected based on wideband acoustic signals and narrowband acoustic signals.
To evaluate the duration of a given narrowband acoustic signal,
Accident sounds with different rise times can be accurately detected. Accident sounds are detected based on a wideband acoustic signal and a narrowband acoustic signal, and the periodicity of the narrowband acoustic signal is evaluated. It is possible to realize an accident sound detection device which can detect the siren sound and the like and can be distinguished from the sudden braking sound and the horn sound.

Further, the accident sound detecting apparatus of the present invention comprises means for calculating the autocorrelation of the time change of the narrow band power based on the information of the time change of the narrow band power, and the periodicity calculating means, It has a configuration for calculating a period of time change of the narrow band power based on the information on the autocorrelation. With this configuration, instead of simply detecting the accident sound based on the power difference value between samples, the accident sound power is estimated using multiple frames and compared with the average power immediately before that Accident sounds with different times can be detected accurately,
Accident sound is detected based on a wideband acoustic signal and a narrowband acoustic signal, and the duration of a predetermined narrowband acoustic signal is evaluated, so even accident sounds with different rise times can be accurately detected. Accident sound is detected based on the wideband acoustic signal and the narrowband acoustic signal, and the periodicity of the narrowband acoustic signal is evaluated, so that the siren sound of an emergency vehicle can be detected, and sudden braking noise and An accident sound detection device that can be distinguished from horn sounds and the like can be realized.

Further, in the accident sound detecting apparatus of the present invention, the time range of the period to be calculated by the periodicity calculating means is 0.
It has a configuration that is any time range of 3 s or more and 10 s or less. With this configuration, instead of simply detecting the accident sound based on the power difference value between samples, the accident sound power is estimated using a plurality of frames,
By comparing with the average power immediately before that, it is possible to accurately detect even accident sounds with different rise times, and detect the accident sound based on the wideband acoustic signal and the narrowband acoustic signal to determine the predetermined narrowband. Since the duration of the acoustic signal is evaluated, accident sounds with different rise times can be detected with high accuracy. Accident sounds are detected based on the wideband acoustic signal and the narrowband acoustic signal, and the narrowband acoustic signal is detected. Since the periodicity of the signal is evaluated, it is possible to realize an accident sound detection device that can detect the siren sound or the like of an emergency vehicle and can be distinguished from the sudden braking sound or the horn sound.

Further, the accident sound detecting apparatus of the present invention has a structure provided with a discriminant analysis means for discriminating the accident sound based on a predetermined characteristic amount according to the accident sound for each accident type. With this configuration, instead of simply detecting the accident sound based on the power difference value between samples, the accident sound power is estimated using multiple frames and compared with the average power immediately before that Accident sounds with different times can be detected with high accuracy.The accident sound is detected based on the wideband acoustic signal and the narrowband acoustic signal, and the rising time is evaluated to evaluate the duration of the predetermined narrowband acoustic signal. Accident sounds with different times can be detected with high accuracy. Accident sounds are detected based on wideband and narrowband acoustic signals, and the periodicity of narrowband acoustic signals is evaluated. It is possible to detect siren sounds, etc., and also distinguish them from sudden braking sounds, horn sounds, etc. Therefore, it is possible to realize that can determine the type of accident sound, an accident sound detection device.

Further, in the accident sound detecting apparatus of the present invention, the discriminant analysis means discriminates the accident sound using a predetermined discriminant function and stores a discriminant coefficient necessary for forming the discriminant function. It has the composition which has. With this configuration, instead of simply detecting the accident sound based on the power difference value between samples, the accident sound power is estimated using multiple frames and compared with the average power immediately before that Accident sounds with different times can be detected with high accuracy.The accident sound is detected based on the wideband acoustic signal and the narrowband acoustic signal, and the rising time is evaluated to evaluate the duration of the predetermined narrowband acoustic signal. Accident sounds with different times can be detected with high accuracy. Accident sounds are detected based on wideband and narrowband acoustic signals, and the periodicity of narrowband acoustic signals is evaluated. Siren sound etc. can be detected,
To realize an accident sound detection device that can distinguish the kind of accident sound because it can be distinguished from sudden braking sound, horn sound, and the like, and when the accident sound is detected, the accident sound is discriminated and analyzed. You can

Further, in the accident sound detecting apparatus of the present invention, as a characteristic amount which is a basis for the discrimination by the discrimination analyzing means, the band of the inputted analog acoustic signal is divided into a plurality of bands and obtained. The configuration is such that band power, which is the level of the signal component, is used. With this configuration,
Accident sound power is not detected based on the difference in power between samples, but accident sound power is estimated by using multiple frames and compared with the average power immediately before that to detect accidents with different rise times. Sounds can be detected with high accuracy. Accident sounds are detected based on a wideband acoustic signal and a narrowband acoustic signal, and the duration of a predetermined narrowband acoustic signal is evaluated. Sound can be detected with high accuracy,
To detect an accident sound based on a wideband acoustic signal and a narrowband acoustic signal, and to evaluate the periodicity of the narrowband acoustic signal,
It is possible to detect the siren sound of an emergency vehicle, distinguish it from sudden braking sound, horn sound, etc. It is possible to realize an accident sound detection device that can be carried out.

Further, the accident sound detecting apparatus of the present invention further receives, as an input, analog acoustic signals from a plurality of microphones other than the first microphone having a directional characteristic different from that of the first microphone which is the microphone. A conversion unit for converting the corresponding analog acoustic signal into a digital signal composed of a frame for each microphone other than the microphone, and a wideband for generating information about the wideband power for each frame based on the digital signal. Power data generation means is further provided, and the determination means further has a sound source direction calculation means for calculating the arrival direction of the accident sound based on each piece of information about the wide band power generated by each of the wide band power data generation means. Have a configuration. With this configuration, instead of simply detecting the accident sound based on the power difference value between samples, the accident sound power is estimated using multiple frames and compared with the average power immediately before that Accident sounds with different times can be detected with high accuracy, and the range in which accident sounds are detected can be limited from the difference between the acoustic signals detected simultaneously using the directional microphone and the omnidirectional microphone. It is possible to realize an accident sound detection device capable of suppressing post-detection due to sound coming from outside the range.

[0021]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing the configuration of an accident sound detection device according to the first embodiment of this invention. In the present invention, the accident sound is a collision sound between vehicles, a collision sound between an object other than the vehicle and the vehicle, a sudden braking sound,
A sound that includes horn sounds.

In FIG. 1, the accident sound detecting apparatus 100 is
After performing a predetermined signal processing on the acoustic signal input to the acoustic signal input terminal 110, a converting unit 120 for converting the acoustic signal into a digital signal, and a wide band power of the acoustic signal based on the digital signal output from the converting unit 120. Wideband power data generating means 130 for calculating, and sound pressure fluctuation detecting means 140 for detecting a sound pressure fluctuation based on the digital signal output from the wideband power data generating means 130 and outputting it to the output terminal 150. To be done.

The conversion means 120 further includes the band limiting means 1
21 and analog / digital conversion (hereinafter, AD (Analog t
o Digital) conversion. ) Means 122. The band limiting unit 121 is a unit for performing band limiting on an acoustic signal that is an analog signal input to the acoustic signal input terminal 110 via a microphone installed near a road or the like, and outputting it to the AD conversion unit 122. is there.
Here, "acoustic signal" is an electrical signal, and "bandwidth limitation"
Means that the input acoustic signal is band-pass filtered.

By limiting the band in this way, the following effects can be obtained. First, low-frequency blocking filter processing can remove low-frequency noise due to wind and vibration. The low frequency band in this case refers to, for example, a frequency band equal to or lower than any frequency in the range of 200 Hz to 400 Hz. Secondly, aliasing associated with AD conversion can be prevented by performing a high-frequency blocking filter process.

Here, in the above high frequency blocking filter processing,
Known anti-aliasing filters can be used. Further, in addition to the above filter processing, it is effective to apply the A characteristic using a filter having a known A characteristic used in a sound level meter or the like to prevent erroneous detection. The AD conversion unit 122 is a unit that receives the analog acoustic signal output from the band limiting unit 121, converts the analog signal into a digital signal, and outputs the digital signal to the sound pressure fluctuation detection unit 140.

Next, the broadband power data generating means 130.
Further includes wideband power calculation means 131 and buffer 13.
It is composed of two. Broadband power calculation means 131
Receives the digital signal output from the AD conversion unit 122, performs a frame division process for dividing the input digital signal into a plurality of frames, calculates wide band power for each frame, and outputs the wide band power to the buffer 132. It is a means to do. Here, the broadband is, for example, 200
It refers to any frequency band in the frequency range from Hz to half the sample frequency. Of course, the sampling frequency at that time is higher than 400 Hz.

Considering a real-time operation by a DSP (Digital Signal Processor) or the like, a longer frame length is advantageous because the amount of calculation per unit time can be reduced.
On the other hand, since it is necessary to shorten the length in order to detect a short-time accident sound that suddenly occurs, the frame length is set in consideration of these.

Specifically, since the time required from the occurrence of the accident sound to the peak of the accident sound level is 5 to 10 ms in the shortest, the frame length is 5 ms.
The range of up to 50 ms is suitable, and about 10 ms is particularly suitable. Further, the interval between the preceding frame and the subsequent frame may be matched with the above frame length.

The broadband power can be calculated by adding the squared values of the amplitude values of the input signal at each sample point within the frame. Here, "within the frame" refers to a part or all of the signal within the frame. The buffer 132 sequentially accumulates the calculated values (data) of the broadband power output from the broadband power calculation means 131 for a predetermined time, and outputs the accumulated broadband power information to the sound pressure fluctuation detection means 140 at fixed time intervals. The size of the data that can be stored is appropriately a size corresponding to a time length of 0.5 s or more.

Next, the sound pressure fluctuation detecting means 140 is further composed of an accident sound power calculating means 141, a difference calculating means 142, a trend calculating means 143, a judging means 144, a variance calculating means 145, and a reference difference value generating means 146. To be done. The trend calculating means 143 is means for inputting the accumulated data of the wide band power output from the buffer 132, calculating a trend described later based on the accumulated data of the wide band power, and outputting the trend to the difference calculating means 142. The trend calculation is performed to calculate the average sound pressure immediately before the occurrence of a sudden event such as an accident sound.

The trend is calculated by a method of calculating the average value of the broadband power over a plurality of frames before the frame for which the accident sound power is to be calculated (not including the frame for which the accident sound power is to be calculated). In that case, the number of frames is preferably 10 to 100, and about 20 is particularly suitable.

The accident sound power calculation means 141 is a means for receiving the information of the wide band power accumulated in the buffer 132, calculating the accident sound power, and outputting it to the difference calculation means 142. The time from the occurrence of the accident sound to the peak of the accident sound level (hereinafter referred to as the rising time) greatly varies depending on the accident sound and is at least 5
There are variations in the range of ms to 100 ms.

Conventionally, the difference value between samples has been used as the accident sound power, but in the case of an accident sound having a short rise time, the accident sound can be detected based on the difference value. In case of a long accident sound,
Since the difference value between samples was small, it could not be detected well.

Since the rise time of the accident sound differs depending on the type of the accident sound, the accident sound power is estimated using a plurality of frames after the specific frame. There are two types of calculation methods for accident sound power. The first method is to set the maximum value of the broadband power in a plurality of frames after the accident sound power calculation frame (including the accident sound power calculation target frame) as the accident sound level. The second method is
In this method, the total value of the broadband powers in a single frame or a plurality of frames after the accident sound power calculation frame (including the accident sound power calculation target frame) is used as the accident sound power. In either method, it is appropriate that the number of frames used for calculation is about 1 to 10.

The difference calculating means 142 inputs the accident sound power output from the accident sound power calculating means 141 and the trend output from the trend calculating means 143, and calculates a difference value between the accident sound power and the trend. Judgment means 1
44 and the variance calculation means 145.

The above difference value is defined as the ratio of the accident sound power to the trend, and the decibel (d
B) Express with a value.

[Equation 1]

The variance calculating means 145 is the difference calculating means 14
Input the information of the difference value between the accident sound power output from 2 and the trend, calculate the average value and variance of the difference values,
It is a means for outputting to the reference difference value generating means 146. The number of frames used to calculate the mean and variance is 3
A number corresponding to a time length of 0 minutes to 1 hour is suitable. The average value and the variance are statistics calculated by a known statistical method.

The reference difference value generating means 146 receives the information on the average value of the difference value between the accident sound power and the trend output from the variance calculating means 145 and the information on the variance, and receives the information on the average value and the variance. It is a means for generating a reference difference value based on the information and outputting it to the determination means 144.

Two methods will be mentioned as the method of setting the reference difference value. The first method is a method of using a fixed value, and the range of the reference difference value is preferably 6 dB to 18 dB for practical use. The second method is a method of setting the reference difference value based on the information of the variance calculated by the variance calculating means 145, and the reference difference value is preferably about four times the variance.

The judging means 144 receives the information on the difference value between the accident sound power and the trend output from the difference calculating means 142 and the information on the reference difference value output from the reference difference value generating means 146, and receives the accident sound. It is determined whether or not the difference value between the power and the trend is equal to or greater than the reference difference value. When the difference value is equal to or greater than the reference difference value, a signal indicating an accident sound is output, and when the difference value is less than the reference difference value, it is not an accident sound. Output signal 15
It is a means for outputting to 0.

The smaller the reference difference value is, the less the mistaken detection of the accident sound is, but the more false detection is caused by the sound other than the accident sound. Further, the larger the reference difference value is, the more the accident sound detection error is, but the false detection due to the sound other than the accident sound is reduced. Therefore, the reference difference value is set appropriately according to the purpose.

FIG. 2 is a flow chart showing the flow of processing in the accident sound detecting method according to the first embodiment of the present invention. In step S221, the band limiting means 121
Performs band limitation on an acoustic signal that is an analog signal input to the acoustic signal input terminal 110 via a microphone installed near a road or the like.

In step S222, the AD conversion means 122
Converts the analog acoustic signal output from the band limiting means 121 into a digital signal. In step S231,
The wideband power calculation unit 131 performs a frame division process for dividing the digital signal output from the AD conversion unit 122 into a plurality of frames, and calculates the wideband power for each frame.

In step S232, the buffer 132
The calculated values (data) of the wide band power output from the wide band power calculating unit 131 are sequentially accumulated for a predetermined time, and the accumulated wide band power information is output to the sound pressure fluctuation detecting unit 140 at constant time intervals. In step S241, the trend calculation means 143 calculates a trend based on the wideband power accumulated data output from the buffer 132.

In step S242, the accident sound power calculation means 141 calculates the accident sound power based on the broadband power information accumulated in the buffer 132. Step S
In 243, the difference calculation means 142 calculates the difference value between the accident sound power output from the accident sound power calculation means 141 and the trend output from the trend calculation means 143.

In step S244, the variance calculation means 145
Calculates the average value and variance of the difference values of the difference values between the accident sound power output from the difference calculating means 142 and the trend. In step S245, the reference difference value generating means 1
46 receives as input information regarding the average value of the difference values between the accident sound power and the trend output from the variance calculating means 145 and information regarding the variance, and sets a reference difference value based on the information regarding the average value and the information regarding the variance. To generate.

In step S246, the determination means 144
Information on the difference value between the accident sound power output from the difference calculating means 142 and the trend, and the reference difference value generating means 146.
Based on the information of the reference difference value output from, it is determined whether the difference value between the accident sound power and the trend is greater than or equal to the reference difference value.

When it is determined in step S246 that the difference value between the accident sound power and the trend is equal to or greater than the reference difference value, the determining means 144 outputs a signal indicating that the accident sound is generated in step S251 and is less than the reference difference value. If it is determined that the signal is not an accident sound in step S252, the output terminal 15
Output to 0. Accident sound is detected by repeating the above frame processing operation.

As described above, the accident sound detecting apparatus and method according to the first embodiment of the present invention does not detect the accident sound based on the mere difference value of the powers between the samples, but a plurality of error sounds are detected. The accident sound power is estimated by using the frame No. 1 and compared with the average power immediately before the accident power, so that the accident sounds having different rise times can be accurately detected.

In the first embodiment of the present invention, the accident sound detecting method for performing the processing in each of the above steps S221 to S252 has been described, but the accident sound detecting operation including each of these steps will be described. An accident sound detection program to be executed is generated, and the computer executes these steps S221 to S2 on the basis of the program.
It is also possible to execute the accident sound detection operation including 52, and it is possible to obtain the same effect as above.

FIG. 3 is a block diagram showing the structure of an accident sound detecting apparatus according to the second embodiment of the present invention. In the present invention, the accident sound means a sound including collision sounds between vehicles, a collision sound between an object other than the vehicle and the vehicle, a sudden braking sound, a horn sound, and the like. Further, among the constituent means constituting the accident sound detecting apparatus 300, those having the same processing as those of the constituent means in the accident sound detecting apparatus 100 according to the first embodiment of the present invention are designated by the same reference numerals. However, the description is omitted.

In FIG. 3, the accident sound detecting device 300 is
In the accident sound detection device 100 according to the first embodiment of the present invention,
Narrowband power data generation means 330 and narrowband power data generation means 3 for calculating the narrowband power of the acoustic signal based on the digital signal output from the conversion means 120.
Based on the information output from 30, the narrow bandwidth detection means 340 for detecting the narrow bandwidth of the acoustic signal, and the information output from the sound pressure fluctuation detection means 140 and the narrow bandwidth detection means 340 The difference is that a classification means 360 for classifying accident sounds is provided.

The narrow band power data generating means 330 further includes a band limiting means 331 and a narrow band power calculating means 33.
2 and a buffer 333. The band limiting unit 331 is a unit for inputting the digital signal output from the converting unit 120, performing band limitation on the input digital signal, and outputting to the narrow band power calculating unit 332.

The purpose of the band limitation is to extract only the signal component of the outstanding horn and the sudden braking of the narrow band component. These accident sounds may have a long rise time of more than 1 s, and in such a case, they may not be detected by the method described in the first embodiment of the present invention. Detection can be performed by limiting the band as described in the second embodiment. The pass frequency of the band pass filter is 800
It is preferable to set it in the range from Hz to 2 kHz. A known IIR filter is preferably used as the bandpass filter, but other digital bandpass filters may be used.

The narrow band power calculation means 332 receives the digital signal output from the band limiting means 331 as input, performs a frame division process for dividing the input digital signal into a plurality of frames, and narrow band for each frame. It is a means for calculating the power and outputting it to the buffer 333.

The frame length and the frame interval in the frame division processing are the same as the values in the wide band power calculation described in the first embodiment of the present invention. Further, the calculation of the narrow band power can be performed by adding the squared values of the amplitude values of the input signal in each sample within the frame.

The buffer 333 outputs the calculated value (data) of the narrow band power output from the narrow band power calculating means 332.
Is sequentially stored for a predetermined time, and the stored wide band power information is output to the narrow band property detecting means 340 at fixed time intervals. The size of the data that can be stored is 0.5.
A size corresponding to a time length of s or more is suitable.

Next, the narrow band characteristic detecting means 340 further comprises a difference calculating means 341 and a duration calculating means 342. The difference calculation means 341 inputs the accident sound power output from the wideband power data generation means 130 and the information on the narrowband power accumulated in the buffer 333, and calculates the difference between the wideband accident sound power and the narrowband accident sound power. It is a means for calculating a value and outputting it to the duration calculating means 342.

The above difference value is defined as the ratio between the wideband accident sound power and the narrowband accident sound power, and is expressed in decibel (dB) values as shown in the following equation (2).

[Equation 2]

The duration calculating means 342 is means for inputting the information of the difference value output from the difference calculating means 341, calculating the duration of the narrowband accident sound signal, and outputting it to the classifying means 360. . The duration of the narrowband accident sound signal refers to the length of time during which the input difference value is equal to or greater than a predetermined value. Here, as the predetermined value, for example, a range of about −3 to −10 dB is suitable. The process performed by the difference calculating unit 341 and the duration calculating unit 342 is referred to as a process for detecting narrow band.

The classifying means 360 comprises sound pressure fluctuation detecting means 14
The accident sound is classified on the basis of the signal indicating whether or not the accident sound is output from 0 and the narrow band accident sound signal output from the duration calculating unit 342, and the signal corresponding to the type of the accident sound is output to the output terminal. It is a means for outputting to 150.

The classification performed by the classifying unit 360 is such that, when the sound pressure fluctuation detecting unit 140 determines that the sound is an accident sound, the narrow band accident sound included in consecutive frames including the frame that is the object of the judgment. If the duration of the signal is detected over a predetermined reference time, it may be classified as an accident sound due to sudden braking or horn, and in other cases, it may be classified as an accident sound in another category. Moreover, about 0.2 s to 2 s is suitable as the reference time to be compared with the duration of the narrowband accident sound signal.

FIG. 4 is a flow chart showing the flow of processing in the accident sound detecting method according to the second embodiment of the present invention. Among the steps constituting the accident sound detecting method, the same reference numbers are given to those performing the same processing as the steps constituting the accident sound detecting method according to the first embodiment of the present invention. The description is omitted.

In step S431, band limiting means 331
Performs band limitation on the digital signal output from the conversion unit 120. In step S432, the narrow band power calculating unit 332 performs a frame dividing process for dividing the digital signal output from the band limiting unit 331 into a plurality of frames, and calculates the narrow band power for each frame.

In step S433, the buffer 333
The calculated value (data) of the narrow band power output from the narrow band power calculating unit 332 is sequentially accumulated for a predetermined time, and the accumulated wide band power information is output to the narrow band property detecting unit 340 at constant time intervals. In step S441, the difference calculation unit 341 calculates the difference value between the accident sound power output in step S232 and the narrowband power information stored in the buffer 333.

In step S442, the duration calculating means 3
42 calculates the duration of the narrowband accident sound signal based on the difference value information output from the difference calculating means 341. In step S460, the classification unit 360 classifies the accident sound based on the signal indicating whether or not the sound pressure fluctuation detection unit 140 is the accident sound and the narrowband accident sound signal output from the duration calculation unit 342. Then, a signal corresponding to the type of accident sound is output to the output terminal 150. Accident sound is detected by repeating the above frame processing operation.

As described above, the accident sound detecting apparatus and method according to the second embodiment of the present invention detects an accident sound based on a wide band acoustic signal and a narrow band acoustic signal,
To evaluate the duration of a given narrowband acoustic signal,
Accident sounds with different rise times can also be accurately detected.

In the second embodiment of the present invention, the accident sound detecting method for performing the processing in each of the above steps S220 to S460 has been described, but the accident sound detecting operation including each of these steps will be described. An accident sound detection program to be executed is generated, and the computer executes these steps S220 to S4 on the basis of the program.
It is also possible to execute the accident sound detection operation including 60, and it is possible to obtain the same effect as the above.

FIG. 5 is a block diagram showing the structure of an accident sound detection system according to the third embodiment of the present invention. In the present invention, the accident sound means a sound including collision sounds between vehicles, a collision sound between an object other than the vehicle and the vehicle, a sudden braking sound, a horn sound, and the like. Further, among the constituent means constituting the accident sound detecting apparatus 500, those having the same processing as those of the constituent means in the accident sound detecting apparatus 300 according to the second embodiment of the present invention are designated by the same reference numerals. However, the description is omitted.

In FIG. 5, the accident sound detecting device 500 is
In the accident sound detection device 300 according to the second embodiment of the present invention,
A periodicity detection unit 540 for calculating the periodicity of the narrowband sound generated based on the information of the narrowband power output from the narrowband power data generation unit 330, and the sound pressure fluctuation detection unit 140 and the narrowband property detection. The difference is that a classifying unit 560 for classifying accident sounds based on output signals from the unit 340 and the autocorrelation calculating unit 541 is provided.

The periodicity detecting means 540 may erroneously detect the siren sound of an emergency vehicle or the like only by the narrow band evaluation in the second embodiment of the present invention. Provided to avoid. The emergency vehicle siren sound may be erroneously detected because it is a narrow band acoustic signal like the sudden braking sound and the horn sound, but is different from the sudden braking sound and the like in that it has periodicity. Therefore, the accident sound detection device 5 according to the third embodiment of the present invention
00 is for detecting an emergency vehicle siren sound or the like by paying attention to the periodicity of the narrow band signal.

The periodicity detecting means 540 further comprises an autocorrelation calculating means 541 and a cycle calculating means 542. The autocorrelation calculation unit 541 receives the narrowband power signal output from the narrowband power data generation unit 330 as input, calculates the autocorrelation of the input narrowband power signal, and cycles the calculated autocorrelation information. It is a means for outputting to the calculation means 542.

As a method of calculating the autocorrelation function, a known method of directly calculating in the time domain may be used, but the known method of once calculating the cross spectrum in the frequency domain and converting it into the time domain requires less calculation amount. That is desirable. In either method, it is desirable to secure 10 s or more as the lag length for calculating the autocorrelation function.

The cycle calculating means 542 is means for receiving the information of the autocorrelation output from the autocorrelation calculating means 541 as input, calculating the periodicity of the autocorrelation signal, and outputting it to 560. As a method of calculating the cycle, a method can be used in which the lag of the autocorrelation is changed within a predetermined time range and the lag at which the autocorrelation is a predetermined value or more and is maximum is set as the cycle. At that time, if the autocorrelation is a predetermined value or more and there is no maximum lag within the time range, the period is not detected. The cycle of the siren sound, etc.
For example, considering that it varies depending on the type of emergency vehicle such as ambulance and police car, the time range for changing the lag is 0.3 s.
It is preferable to set it to about 10 s.

The classifying means 560 is the sound pressure fluctuation detecting means 14
0 is an accident sound, a narrow band accident sound signal output from the duration calculating unit 342, and an information on the periodicity output from the periodicity detecting unit 540. And outputs a signal corresponding to the type of accident sound to the output terminal 150.

The classification performed by the classifying unit 560 is performed by the siren when the sound pressure fluctuation detecting unit 140 determines that the sound is an accident sound and when the period calculating unit 542 outputs a signal that the periodicity is detected. When the narrow band signal is detected in other cases, it may be classified as an accident sound such as a sudden braking sound or a horn sound. Further, when the accident sound is not classified into the above two categories, it may be classified into other accident sound categories.

FIG. 6 is a flow chart showing the flow of processing in the accident sound detecting method according to the third embodiment of the present invention. Among the steps constituting the accident sound detecting method, those having the same processing as the steps constituting the accident sound detecting method according to the second embodiment of the present invention are given the same reference numerals, The description is omitted.

In step S641, the autocorrelation calculation means 5
41 receives the signal of the narrow band power output from the narrow band power data generating means 330, and calculates the autocorrelation for the input narrow band power signal. Step S6
At 42, the cycle calculating means 542 is operated by the autocorrelation calculating means 54.
The periodicity of the autocorrelation signal output from 1 is calculated.

At step S660, the classification means 560
A signal indicating whether or not an accident sound is output from the sound pressure fluctuation detecting means 140, a narrowband accident sound signal output from the duration calculating means 342, and information regarding periodicity output from the periodicity detecting means 540. Accident sounds are classified based on. Here, the method of classifying the accident sound may be the method described above. Accident sound is detected by repeating the above frame processing operation.

As described above, the accident sound detecting apparatus and method according to the third embodiment of the present invention detects an accident sound based on a wide band acoustic signal and a narrow band acoustic signal,
Since the periodicity of the narrow band acoustic signal is evaluated, the siren sound or the like of the emergency vehicle can be detected and can be distinguished from the sudden braking sound or the horn sound.

In the third embodiment of the present invention, the accident sound detection method for performing the processing in each of the above steps S220 to S660 has been described. However, the accident sound detection operation including each of these steps will be described. An accident sound detection program to be executed is generated, and the computer executes these steps S220 to S6 based on the program.
It is also possible to execute the accident sound detection operation including 60, and it is possible to obtain the same effect as the above.

FIG. 7 is a block diagram showing the structure of an accident sound detection system according to the fourth embodiment of the present invention. In the present invention, the accident sound means a sound including collision sounds between vehicles, a collision sound between an object other than the vehicle and the vehicle, a sudden braking sound, a horn sound, and the like. Further, among the constituent means constituting the accident sound detecting apparatus 700, the same reference numerals are given to those performing the same processing as the constituent means in the accident sound detecting apparatus 100 according to the first embodiment of the present invention. However, the description is omitted.

In FIG. 7, the accident sound detecting device 700 is
In the accident sound detection device 100 according to the first embodiment of the present invention,
Based on the band power data generation means 730 for calculating the band power of the acoustic signal based on the digital signal output from the conversion means 120, and the output signals from the sound pressure fluctuation detection means 140 and the band power data generation means 730. Analysis / discrimination means 74 for analyzing and discriminating accident sounds by
The difference is that 0 is provided.

The band power data generating means 730 further includes an FFT calculating means 731, a band power calculating means 732.
And a buffer 733. The FFT calculation unit 731 receives the digital signal output from the conversion unit 120 as input, performs a frame division process for dividing the input digital signal into a plurality of frames, performs a known FFT for each frame, and outputs the band power. Calculation means 732
Is a means for outputting to.

The frame length and the frame interval in the frame division processing are the same as the values in the wide band power calculation processing described in the first embodiment of the present invention.
Further, the calculation of the narrow band power can be performed by adding the squared values of the amplitude values of the input signal in each sample within the frame. The FFT length needs to be equal to or larger than the number of samples forming one frame.

The band power calculation means 732 receives the signal after the FFT processing outputted from the FFT calculation means 731 as an input, calculates the band power for each frame of the inputted signal, and outputs it to the buffer 333. Is a means of. The purpose of the band power calculation is to calculate information that is used as a clue for discrimination used in the analysis discrimination unit 740, which will be described later, for discriminating the type of accident sound.

Bandwidth is 1 / n that is typically used in acoustic measurements.
A three octave bandwidth is preferred. For each band, the total power of the frequencies included in the band is calculated and used as the band power. The band power calculated here is represented by Li (i = 1 to N; N is the number of bands). The buffer 333 sequentially accumulates the calculated value (data) of the band power output from the band power calculation unit 732 for a predetermined time, and outputs the accumulated band power information to the analysis determination unit 740 at constant time intervals. It is a means.

The analysis / discrimination means 740 further includes a trigger detection means 741, an input variable setting means 742, a discrimination function coordinate calculation means 743, a discrimination coefficient holding means 744, a distance calculation means 745, a barycentric coordinate holding means 746, and a discrimination means 74.
It is composed of 7. The trigger detection means 741 receives a signal indicating whether it is an accident sound output from the sound pressure fluctuation detection means 140, determines whether or not it is an accident sound signal, and when it is determined to be an accident sound, a predetermined signal Input variable setting means 74
It is a means for outputting to 2.

The input variable setting means 742 is the buffer 73.
Calculated value (data) of band power output from 3
(I = 1 to N) and a predetermined signal output from the trigger detection means 741 as input, the full band level L shown in the following formula (3) is calculated, and the information of the full band level L is calculated as a discriminant function coordinate. It is a means for outputting to the means 743.

[Equation 3]

The discriminant function coordinate calculating means 743 receives the information of the entire band level L output from the input variable setting means 742 as input, and the discriminant function coordinate X _j (j = 1) shown in the following equation (4).
˜M) and outputs the information of the discriminant function coordinate X _j (j = 1 to M) to the distance calculating means 745.

[Equation 4]

Here, α _ij is a discrimination coefficient, which is a coefficient obtained in advance by analyzing a plurality of sample data for each type of accident sound. The analysis method is, for example, the well-known “discrimination based on Mahalanobis distance”. It is also possible to use the “method of generating coefficients”. Further, M is the number of dimensions and is represented by (the number of discriminant groups-1). The discrimination coefficient holding means 744 is a means for holding the discrimination coefficient α _ij described above, and is configured by a memory or the like.

The distance calculating means 745 receives the discriminant function coordinate X _j output from the discriminant function coordinate calculating means 743 as an input,
It is means for calculating the square distance D _k shown in the following equation (5) and outputting the information of the square distance D _k to the discriminating means 747.

[Equation 5]

Here, Y _k is a center coordinate, which is a coefficient obtained by previously analyzing a plurality of sample data for each type of accident sound, and the analysis method is, for example, based on the well-known “Mahalanobis distance. It is also possible to use the “discrimination coefficient generation method”. The center-of-gravity coordinate holding unit 746 uses the above-mentioned center coordinate Y.
It is a means for holding _k , and is configured by a memory or the like.

The discriminating means 747 receives the information of the square distance D _k output from the distance calculating means 745, determines _k having the minimum square distance D k as a discriminating group, and outputs a signal according to the result of the discrimination. It is a means for outputting to the terminal 150.

FIG. 8 is a flowchart showing the flow of processing in the accident sound detecting method according to the second embodiment of the present invention. Among the steps constituting the accident sound detecting method, the same reference numbers are given to those performing the same processing as the steps constituting the accident sound detecting method according to the first embodiment of the present invention. The description is omitted.

At step S831, the FFT calculation means 73.
In No. 1, the digital signal obtained by the AD conversion in step S220 is subjected to frame division processing for dividing into a plurality of frames, and known FFT processing is performed for each frame. In step S832, band power calculation means 732
Calculates the band power for each frame of the signal after the FFT processing.

In step S833, the buffer 333
The calculated values (data) of the band power calculated in step S832 are sequentially accumulated for a predetermined time, and the accumulated band power information is output to the analysis determination unit 740 at regular time intervals. In step S841, the trigger detection means 741 determines whether or not this signal is a signal indicating an accident sound based on the signal indicating whether or not the accident sound is generated in step S240, and determines that the signal is an accident sound. Outputs a predetermined signal to the input variable setting means 742.

In step S842, the input variable setting means 7
When the predetermined signal is input in step S841, 42 calculates the total band level L shown in the above equation (3).
In step S843, the discriminant function coordinate calculation means 743
Based on the information of the entire band level L calculated in step S842, the discriminant function coordinate X _j (j = 1) shown in the above equation (4).
~ M) is calculated.

In step S845, distance calculation means 745
Calculates the square distance D _k shown in the above equation (5) based on the discriminant function coordinate X _j calculated in step S843. In step S847, the determination unit 747 determines, based on the information on the square distance D _k calculated in step S845,
The _k having the smallest squared distance D _k is determined as the discrimination group. Accident sound is detected by repeating the above frame processing operation.

As described above, the accident sound detecting apparatus and method according to the fourth embodiment of the present invention discriminates and analyzes the accident sound when the accident sound is detected. can do.

In the fourth embodiment of the present invention, the accident sound detecting method for performing the processing in each of the above steps S220 to S847 has been described, but the accident sound detecting operation including each of these steps will be described. An accident sound detection program to be executed is generated, and the computer executes these steps S220 to S8 on the basis of the program.
It is also possible to execute the accident sound detection operation including 47, and it is possible to obtain the same effect as above.

FIG. 9 is a block diagram showing the structure of an accident sound detection system according to the fifth embodiment of the present invention. In the present invention, the accident sound means a sound including collision sounds between vehicles, a collision sound between an object other than the vehicle and the vehicle, a sudden braking sound, a horn sound, and the like. Further, among the constituent means constituting the accident sound detecting apparatus 900, those which perform the same processing as the constituent means in the accident sound detecting apparatus 100 according to the first embodiment of the present invention are designated by the same reference numerals. However, the description is omitted.

In FIG. 9, the accident sound detecting device 900 is
First and second microphones 90 arranged near the intersection 901
2, 903, the output signals from the first and second conversion means 120, the first and second conversion means 120, into which the acoustic signals detected by the first and second microphones 902 and 903, respectively, are input. Sound source direction calculation means 940 for calculating the sound source direction based on the output signals from the first and second wide band power data generation means 130 and the first and second wide band power data generation means 130, wide band power data Output signals from the sound pressure fluctuation detecting means 140, the sound pressure fluctuation detecting means 140, and the sound source direction calculating means 940, to which the output signal of the generating means 130 is input, are input to generate predetermined discrimination information regarding the occurrence of an accident. Discrimination means 960, surveillance camera 904 showing intersection 901, surveillance camera 904
AD conversion means 920 to which the video signal from
It is composed of a buffer 930 to which the digital image signal from the conversion means 920 is input, an image cutout means 970 to which the digital image signal from the buffer 930 and the discrimination information from the discrimination means 960 are inputted, and a storage means 980. Here, the first microphone 902 is a directional microphone, and its directional range is set to the range indicated by the quotation mark “905” in FIG. The second microphone is an omnidirectional microphone.

The sound source direction calculating means 940 further comprises a difference calculating means 941, a sound source direction judging means 942, and a judgment reference generating means 943. Difference calculating means 9
41 is the first and second wide band power data generating means 13
The means for calculating the difference between the wideband power of the unidirectional microphone 902 and the wideband power of the omnidirectional microphone 903 by inputting the wideband power data output from 0 as input, and outputting it to the sound source direction determining means 942. is there.

Here, in each frame, the difference between the above two values is expressed by a decibel value (dB) as shown in the following expression (6).
Calculate with.

[Equation 6] The correction value in the above equation (6) is a correction value for adjusting the sensitivity of the unidirectional microphone S902 and the sensitivity of the omnidirectional microphone S903, and one sound source directs the unidirectional microphone. When it is on the principal axis of the direction, the difference value shown in the above equation (6) is set to be zero.

The sound source direction judging means 942 receives the difference between the wide band power of the unidirectional microphone 902 and the wide band power of the omnidirectional microphone 903 outputted from the difference calculating means 941 as input, and makes the judgment reference generating means 943 beforehand. The set value and the above difference are compared, and the accident sound indicates the pointing range 905.
Is a means for determining whether or not the occurrence has occurred within the range and outputting the determination result to the determining means 960.

Here, since the directional range of the unidirectional microphone 902 is oriented toward the direction in which an accident sound is generated, such as in an intersection, the unidirectional microphone 902 responds to the accident sound coming from the directional range. Has high sensitivity, but has low sensitivity to sounds other than accident sounds such as construction site noise and airplane noise coming from other directions. On the other hand, the omnidirectional microphone 90
3 has high sensitivity in both directions. Therefore, by taking the difference between the two, when the difference between the two is small, the direction of sound generation is within the directivity range of the unidirectional microphone, and when the difference between the two is large, the direction of sound generation is other than that. It can be estimated to be the direction.

The method of estimating the sound source direction is, for example, that the sound source direction is unidirectional when the difference calculated by the above equation (6) is larger than the judgment reference value preset in the judgment reference generating means 943. It may be determined that the microphone S902 is within the directivity range 905, and if not, it may be determined to be outside the directivity range 905. The above judgment reference value is -3 dB to-
Any value in the 12 dB range is suitable.

The discrimination means 960 is the sound pressure fluctuation detection means 14
A signal indicating whether the sound is an accident sound from 0 and the sound source direction calculation means 9
Information regarding the sound source direction from 40 is input, and it is determined whether or not the accident sound is generated in the pointing range 905 based on the signal indicating whether the sound is an accident sound and the determination result of the sound source direction. It is a means for outputting a predetermined signal to the image cutout means 970 when it is determined that the accident sound has occurred. At that time, if it is not determined that the accident sound is generated within the pointing range 905, no signal is output to the pointing range 905.

The surveillance camera 904 continuously photographs the intersection, and the data obtained by the photographing is AD conversion means 920.
Is a means for outputting to. The AD conversion means 920
It is means for AD-converting the data output from the monitoring camera 904 to generate image data and outputting the image data to the buffer 930. The buffer 930 accumulates the image data obtained by the AD conversion for a predetermined time, and the accumulated image data is accumulated image data according to an output signal from an image extracting unit 970 described later. Is a means for outputting to.

The image cut-out means 970 receives the signal output when the result of the discrimination by the discrimination means 960 is an accident sound within the directivity range 905, and when it receives the signal from the discrimination means 960, the buffer 930. A predetermined signal is output to and the predetermined image data of the image data output from the buffer 930 is cut out according to the output signal.

The clipping may be performed within the pointing range 905 for about 10 seconds before and after the time when the accident sound is detected. The cut-out image may be stored in the storage unit 980. Here, the storage means 9
80 is preferably a hard disk or the like.

FIG. 10 is a flow chart showing the flow of processing in the accident sound detecting method according to the fifth embodiment of the present invention. Among the steps constituting the accident sound detecting method, the same reference numbers are given to those performing the same processing as the steps constituting the accident sound detecting method according to the first embodiment of the present invention. The description is omitted.

The conversion process performed in step S220 is
This is applied to the acoustic signals output from the unidirectional microphone 902 and the omnidirectional microphone 903. Step S1
At 020, the AD conversion means 920 performs AD conversion processing on the analog image data output from the surveillance camera 904 to generate digital image data.

The wideband power data generation processing performed in step S230 is performed on the digital signals output from both the first and second conversion means 120. In step S1030, buffer 930 accumulates the image data obtained by AD conversion for a predetermined time.

The processing in step S240 is performed on the signal output from the wide band power data generating means 130. In step S1041, the difference calculation means 941
Is based on the wideband power data output from the first and second wideband power data generating means 130, and the wideband power of the unidirectional microphone 902 and the omnidirectional microphone 903 of the unidirectional microphone 902 shown in the above equation (6). Calculate the difference from the broadband power.

In step S1042, the sound source direction determining means 942 compares the difference between the wideband power of the unidirectional microphone 902 and the wideband power of the omnidirectional microphone 903 calculated in step S1041 with a preset value. Then, it is determined whether or not the accident sound is generated within the pointing range 905.

At step S1060, the discrimination means 960.
Determines whether or not the accident sound is generated in the pointing range 905 based on the signal indicating whether or not the sound is the accident sound generated in step S240 and the information regarding the sound source direction generated in the determination in step S1042. When it is determined that the accident sound is generated within the pointing range 905, a predetermined signal is output as the image cutout unit 9.
It is a means for outputting to 70. At that time, the pointing range 9
No signal is output to the pointing range 905 unless it is determined that the accident sound is generated within 05.

In step S1070, the image cutting means 97.
0 cuts out predetermined image data from the image data output from the buffer 930 when the result of the determination in 1060 is an accident sound within the pointing range 905. In step S1080, the storage unit 980 stores the information in step S1070.
The image data cut out by is stored.

As described above, the accident sound detecting apparatus and method according to the fifth embodiment of the present invention is based on the difference between the acoustic signals detected simultaneously using the directional microphone and the omnidirectional microphone. Since the range in which the sound is detected can be limited, it is possible to suppress the post-detection by the sound coming from outside the desired range.

In the fifth embodiment of the present invention, the accident sound detecting method for performing the processing in each of the above steps S220 to S1080 has been described, but the accident sound detecting operation including each of these steps will be described. An accident sound detection program to be executed is generated, and the computer executes these steps S220 to S based on the program.
It is also possible to execute the accident sound detection operation including 1080, and it is possible to obtain the same effect as the above.

[0122]

As described above, the present invention improves the accuracy of accident sound detection in an accident sound detection device, reduces erroneous detection due to the siren sound of an emergency vehicle, and determines the type of detected accident sound. It is possible to provide an accident sound detection device, method, and program capable of making a determination and further reducing erroneous detection caused by sounds coming from directions other than the desired direction.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of an accident sound detection device according to a first embodiment of the present invention.

FIG. 2 is a flowchart showing a processing flow in the accident sound detection method according to the first embodiment of the present invention.

FIG. 3 is a block diagram showing a configuration of an accident sound detection device according to a second embodiment of the present invention.

FIG. 4 is a flowchart showing a flow of processing in the accident sound detection method according to the second embodiment of the present invention.

FIG. 5 is a block diagram showing a configuration of an accident sound detection device according to a third embodiment of the present invention.

FIG. 6 is a flowchart showing the flow of processing in the accident sound detection method according to the third embodiment of the present invention.

FIG. 7 is a block diagram showing a configuration of an accident sound detection device according to a fourth embodiment of the present invention.

FIG. 8 is a flowchart showing the flow of processing in the accident sound detection method according to the fourth embodiment of the present invention.

FIG. 9 is a block diagram showing a configuration of an accident sound detection device according to a fifth embodiment of the present invention.

FIG. 10 is a flowchart showing the flow of processing in the accident sound detection method according to the fifth embodiment of the present invention.

FIG. 11 is a block diagram showing the configuration of a conventional accident sound detection device.

[Explanation of symbols]

100, 300, 500, 700, 900, 1100
Accident sound detection device 110 Acoustic signal input terminal 120 Conversion means 121 Band limiting means 122 Analog / digital conversion means (AD conversion means) 130 Broadband power data generation means 131 Broadband power calculation means 132 Buffer 140 Sound pressure fluctuation detection means 141 Accident sound power Calculation means 142 Difference calculation means 143 Trend calculation means 144 Judgment means 145 Variance calculation means 146 Reference difference value generation means 150 Output terminal 330 Narrow band power data generation means 331 Band limiting means 332 Narrow band power calculation means 333 Buffer 340 Narrow bandwidth detection Means 341 Difference calculation means 342 Duration calculation means 360 Classification means 540 Periodicity detection means 541 Autocorrelation calculation means 542 Cycle calculation means 730 Band power data generation means 731 FFT calculation means 732 Band power calculation means 33 buffer 740 analysis discriminating means 741 trigger detecting means 742 input variable setting means 743 discriminant function coordinate calculating means 744 discriminant coefficient holding means 745 distance calculating means 746 barycentric coordinate holding means 747 discriminating means 901 intersections 902, 903 microphone 904 monitoring camera 905 pointing range 920 AD conversion means 930 Buffer 940 Sound source direction calculation means 941 Difference calculation means 942 Sound source direction determination means 943 Judgment reference generation means 960 Discrimination means 970 Image cutout means 980 Storage means 1102 Waveform shaping circuit 1103 AD converter 1104 Signal smoothing means 1105 Difference calculating means 1106 Comparison circuit 1107 Reference difference value

Claims (42)

[Claims] 1. A conversion means for converting an analog acoustic signal detected by a microphone into a digital signal composed of a frame, and based on the digital signal,
Wideband power data generating means for generating information about wideband power, which is a signal level of a signal component occupying a predetermined band of the analog audio signal, for each frame, and a predetermined number of frames past the accident sound calculation target frame The average value calculating means for calculating the average value of the wideband power obtained for each frame, and the signal of the frame of the accident sound calculation target is an accident based on the information of the wideband power and the information of the average value of the wideband power. An accident sound detection device, comprising: a judgment means for judging whether or not the signal corresponds to a sound. 2. The accident sound detection device comprises maximum value calculation means for calculating the maximum value of the wide band power obtained for each frame from the accident sound calculation target frame up to a predetermined number of past frames. , A difference value calculating means for calculating a difference value between the maximum value of the wideband power and the average value of the wideband power, the determining means, of the frame of the accident sound calculation target based on the information of the difference value. The accident sound detection device according to claim 1, wherein it is determined whether or not the signal corresponds to the accident sound. 3. The accident sound detection device, a total value calculation means for calculating a total value of the wide band power obtained for each frame from the frame of the accident sound calculation target to a predetermined number of past frames, A difference value calculating means for calculating a difference value between the total value of the wideband power and the average value of the wideband power, the determining means, the signal of the frame of the accident sound calculation target based on the information of the difference value. The accident sound detecting apparatus according to claim 1, wherein it is determined whether or not the signal corresponds to the accident sound. 4. The narrow band power data, wherein the accident sound detecting device calculates narrow band power which is a signal level of a signal component occupying any frequency band in an audible frequency band in an analog acoustic signal detected by the microphone. A signal of the frame of the accident sound calculation target corresponding to the accident sound is generated by the generation means, the narrow-bandwidth detection means for detecting narrowbandness based on the wideband power information and the narrowband power information, and the determination means. And a classifying unit that classifies the accident sound based on information as to whether or not the narrow band property is detected by the narrow band property detection unit. The accident sound detection device described in 1. 5. The narrow band characteristic detecting means calculates a difference value between the wide band power and the narrow band power, and detects the narrow band characteristic based on the difference value between the wide band power and the narrow band power. The accident sound detection device according to claim 4, wherein 6. The frequency band of any of the audible frequency bands to which the narrow band power data generating means calculates the narrow band power is a frequency band of any range in a frequency range of 800 Hz or more and 2 kHz or less. The accident sound detection device according to claim 5, wherein 7. The accident sound detecting device comprises a periodicity calculating means for calculating a cycle of time change of the narrow band power,
The classifying unit calculates the periodicity when the determination unit determines that the signal of the accident sound calculation target frame is a signal corresponding to an accident sound, and the narrowband property detecting unit detects narrowband property. The method according to claim 4, wherein when the cycle is not calculated by the means, the sound is classified as an accident sound, and when the periodicity is calculated, the sound is not classified as an accident sound.
Accident sound detector described. 8. The narrow band power data, wherein the accident sound detection device calculates narrow band power which is a signal level of a signal component occupying any frequency band in an audible frequency band of an analog acoustic signal detected by the microphone. In the case where the signal of the frame of the accident sound calculation target is determined to be a signal corresponding to the accident sound by the generation means, the periodicity calculation means for calculating the time change cycle of the narrow band power, and the judgment means, The accident sound detection according to claim 1, further comprising a classifying unit that classifies the sound as an accident sound when the cycle is not calculated by the periodicity calculating unit and the accident sound when the cycle is calculated. apparatus. 9. The accident sound detection device comprises means for calculating an autocorrelation of the temporal change of the narrowband power based on information of the temporal change of the narrowband power, and the periodicity calculation means is for the self-correlation. 9. The accident sound detecting device according to claim 7, wherein the cycle of the time change of the narrow band power is calculated based on the information about the correlation. 10. The time range of the cycle to be calculated by the periodicity calculation means is any time range of 0.3 s or more and 10 s or less. Accident sound detector described in. 11. The accident sound detecting device comprises a discriminant analysis means for discriminating the accident sound on the basis of a predetermined feature amount corresponding to the accident sound for each accident type. Accident sound detector. 12. The discriminant analysis means comprises means for discriminating an accident sound using a predetermined discriminant function and storing a discriminant coefficient necessary for forming the discriminant function. Accident sound detector described. 13. The band power, which is the level of the signal component of each band obtained by dividing the band of the input analog acoustic signal into a plurality of bands, is used as a feature amount to be a basis for the determination by the determination analysis unit. The accident sound detecting device according to claim 12, which is used. 14. The accident sound detection device further receives as input analog sound signals from a plurality of microphones other than the first microphone having a different directional characteristic from the first microphone which is the microphone, and receives the first microphone. For each microphone other than, conversion means for converting the corresponding analog acoustic signal into a digital signal composed of a frame, and wideband power data for generating information on the wideband power for each frame based on the digital signal The determining means further includes a sound source direction calculating means for calculating an arrival direction of the accident sound based on each piece of information about the wide band power generated by each of the wide band power data generating means. The accident sound detection device according to claim 1, which is characterized in that. 15. A conversion step of converting an analog acoustic signal detected by a microphone into a digital signal composed of frames, and a signal occupying a predetermined band of the analog acoustic signal for each frame based on the digital signal. Wideband power data generation step of generating information about wideband power that is a signal level of the component, and calculating an average value of the wideband power obtained for each frame from the frame of accident sound calculation to a frame past a predetermined number of frames. An average value calculation step,
A determination step of determining whether or not the signal of the frame of the accident sound calculation target is a signal corresponding to the accident sound based on the information of the wide band power and the information of the average value of the wide band power. Accident sound detection method. 16. The maximum value calculation step of calculating the maximum value of the wideband power obtained for each frame from the frame of the accident sound calculation target to a frame of a predetermined number of past frames in the accident sound detection method, , A difference value calculating step of calculating a difference value between the maximum value of the wideband power and the average value of the wideband power, the determining step, of the frame of the accident sound calculation target based on the information of the difference value. The accident sound detection method according to claim 15, wherein it is determined whether or not the signal corresponds to the accident sound. 17. The accident sound detection method includes a total value calculation step of calculating a total value of the wide band powers obtained for each frame from the frame of the accident sound calculation target to a predetermined number of past frames. A difference value calculation step of calculating a difference value between the total value of the wideband power and the average value of the wideband power, wherein in the determination step, the signal of the accident sound calculation target frame is based on the information of the difference value. The accident sound detecting method according to claim 15, wherein it is determined whether or not the signal corresponds to the accident sound. 18. The narrowband power data, wherein the accident sound detection method calculates narrowband power which is a signal level of a signal component occupying any frequency band in an audible frequency band of an analog acoustic signal detected by the microphone. A signal of the frame of the accident sound calculation target in the generation step, a narrow band property detection step of detecting a narrow band property based on the wide band power information and the narrow band power information, and the determination step corresponds to the accident sound. And a classification step of classifying the accident sound based on information on whether or not the narrowband property is detected in the narrowband property detection step. 15. Accident sound detection method described in 15. 19. In the narrowband property detecting step, a difference value between the wideband power and the narrowband power is calculated,
19. The accident sound detection method according to claim 18, wherein the narrow band property is detected based on a difference value between the wide band power and the narrow band power. 20. Any frequency band in the audible frequency band for which the narrow band power is calculated in the narrow band power data generation step is 800 Hz or more and 2 kHz.
20. The accident sound detection method according to claim 19, wherein the frequency band is any one of frequency bands in a frequency range of z or less. 21. The accident sound detection method comprises a periodicity calculation step of calculating a cycle of a time change of the narrow band power, wherein in the classification step, a signal of a frame of the accident sound calculation target is determined in the determination step. When the signal is determined to correspond to the accident sound, and the narrow band property is detected in the narrow band property detection step, if the period is not calculated in the periodicity calculation step, it is classified as an accident sound and the periodicity is calculated. 19. The method for detecting an accident sound according to claim 18, wherein the case is not classified as an accident sound. 22. The narrow-band power data, wherein the accident sound detection method calculates narrow-band power which is a signal level of a signal component occupying any frequency band in an audible frequency band in an analog acoustic signal detected by the microphone. When the generation step, the periodicity calculation step of calculating the cycle of the time change of the narrow band power, and the signal of the frame of the accident sound calculation target in the judgment step is a signal corresponding to the accident sound, 16. The accident sound detection according to claim 15, further comprising a classification step for classifying the sound as an accident sound when the cycle is not calculated in the periodicity calculation step and the accident sound when the cycle is calculated. Method. 23. The accident sound detection method comprises the step of calculating an autocorrelation of the temporal change of the narrowband power based on information of the temporal change of the narrowband power, and in the periodicity calculation step, the self-correlation is calculated. 23. The accident sound detection method according to claim 21 or 22, wherein the cycle of the temporal change of the narrow band power is calculated based on the information on the correlation. 24. The time range of the cycle to be calculated in the periodicity calculation step is any time range of 0.3 s or more and 10 s or less.
24. The accident sound detection method according to any one of 1 to 23. 25. The accident sound detection method according to claim 24, further comprising a discriminant analysis step of judging the accident sound based on a predetermined feature amount according to the accident sound for each accident type. Accident sound detection method. 26. The discriminant analysis step includes a step of discriminating an accident sound using a predetermined discriminant function and storing a discriminant coefficient necessary for forming the discriminant function. Accident sound detection method described. 27. The band power, which is the level of the signal component of each band obtained by dividing the band of the input analog acoustic signal into a plurality of bands, as a feature amount to be a basis for the determination in the determination analysis step. 25. The accident sound detecting method according to claim 24, wherein: 28. The accident sound detection method further inputs analog acoustic signals from a plurality of microphones other than the first microphone, which has a different directional characteristic, from the first microphone, which is the microphone, and receives the first microphone. For each microphone other than, a conversion step of converting the corresponding analog acoustic signal into a digital signal constituted by a frame, and based on the digital signal,
A wideband power data generation step of generating information on the wideband power for each frame, wherein the determination step includes the accident based on the respective pieces of information on the wideband power generated in the wideband power data generation step. The method further comprising a sound source direction calculation step for calculating a sound arrival direction.
5. Accident sound detection method described in 5. 29. A conversion step of causing a computer to convert an analog acoustic signal detected by a microphone into a digital signal composed of frames, and a predetermined band of the analog acoustic signal for each frame based on the digital signal. Wideband power data generation step for generating information about wideband power that is the signal level of the signal component occupying, and the average value of the wideband power obtained for each frame from the frame subject to accident sound calculation to a predetermined number of past frames And an average value calculation step for calculating, and determining whether the signal of the frame of the accident sound calculation target is a signal corresponding to the accident sound, based on the wide band power information and the wide band power average value information. Accident sound detection program characterized by executing a judgment step . 30. The accident sound detection program further calculates a maximum value for calculating the maximum value of the broadband power obtained for each frame from the accident sound calculation target frame to a frame past a predetermined number of frames. Causing the computer to execute a step and a difference value calculating step of calculating a difference value between the maximum value of the wideband power and the average value of the wideband power, and in the determining step, the accident sound based on the information of the difference value. 30. The accident sound detection program according to claim 29, wherein it is determined whether or not the signal of the frame to be calculated is a signal corresponding to the accident sound. 31. A total value calculating step for calculating a total value of the wideband powers obtained by the accident sound detection program for each frame from the accident sound calculation target frame up to a predetermined number of past frames; The difference value calculation step of calculating a difference value between the total value of the wideband power and the average value of the wideband power is executed by the computer, and in the determination step, the accident sound calculation target frame is based on the information of the difference value. 30. The accident sound detection program according to claim 29, wherein it is determined whether or not the signal is a signal corresponding to an accident sound. 32. Narrowband power data, wherein the accident sound detection program calculates narrowband power which is a signal level of a signal component occupying any frequency band in an audible frequency band of an analog acoustic signal detected by the microphone. A signal of the frame of the accident sound calculation target in the generation step, a narrow band property detection step of detecting a narrow band property based on the wide band power information and the narrow band power information, and the determination step corresponds to the accident sound. And a classification step of classifying the accident sound based on information on whether or not the narrowband property is detected in the narrowband property detection step, when the signal is determined to be The accident sound detection program according to claim 29. 33. In the narrowband property detecting step, a difference value between the wideband power and the narrowband power is calculated,
33. The accident sound detection program according to claim 32, wherein narrowband property is detected based on a difference value between the wideband power and the narrowband power. 34. Any frequency band in the audible frequency band for which the narrow band power is calculated in the narrow band power data generating step is 800 Hz or more and 2 kHz.
34. The accident sound detection program according to claim 33, wherein the accident sound detection program is a frequency band in any range within a frequency range of z or less. 35. The accident sound detection program causes a computer to execute a periodicity calculation step of calculating a cycle of time change of the narrow band power, and in the classification step, the accident sound calculation target frame in the determination step. Is determined as a signal corresponding to an accident sound, and when the narrow band property is detected in the narrow band property detection step, if the period is not calculated in the periodicity calculation step, it is classified as an accident sound, and the periodicity is determined. 33. The accident sound detection program according to claim 32, wherein when is calculated, it is not classified as an accident sound. 36. Narrowband power data, wherein the accident sound detection program calculates narrowband power, which is a signal level of a signal component occupying any frequency band in an audible frequency band of an analog acoustic signal detected by the microphone. When the generation step, the periodicity calculation step of calculating the cycle of the time change of the narrow band power, and the signal of the frame of the accident sound calculation target in the judgment step is a signal corresponding to the accident sound, 31. The accident according to claim 29, further comprising: a classification step of classifying the computer so that the sound is an accident sound when the cycle is not calculated in the periodicity calculation step, and the sound is not an accident sound when the periodicity is calculated. Sound detection program. 37. The accident sound detection program causes a computer to execute a step of calculating an autocorrelation of the temporal change of the narrow band power based on information of the temporal change of the narrow band power, and in the periodicity calculating step, 37. The accident sound detection program according to claim 35, wherein the cycle of the time change of the narrow band power is calculated based on the information on the autocorrelation. 38. The time range of the cycle to be calculated in the periodicity calculation step is any time range of 0.3 s or more and 10 s or less.
The accident sound detection program according to any one of 5 to 37. 39. The accident sound detection program further causes a computer to execute a discriminant analysis step of judging the accident sound based on a predetermined feature amount according to the accident sound for each accident type. Item 38. An accident sound detection program according to Item 38. 40. The discriminant analysis step includes a step of discriminating an accident sound using a predetermined discriminant function, and storing a discriminant coefficient necessary for forming the discriminant function. Accident sound detection program as described. 41. A band power, which is a level of a signal component of each band obtained by dividing a band of the input analog acoustic signal into a plurality of bands, as a feature amount to be a basis for the discrimination in the discriminant analysis step. 41. The accident sound detection program according to claim 40, wherein: 42. The accident sound detection program further receives analog acoustic signals from a plurality of microphones other than the first microphone, which has a different directional characteristic, as the first microphone, which is the microphone, and inputs the analog sound signals to the first microphone. For each microphone other than, a conversion step of converting the corresponding analog acoustic signal into a digital signal composed of a frame, and wideband power data for generating information on the wideband power for each frame based on the digital signal And a step of calculating a sound source direction for calculating the arrival direction of the accident sound based on each piece of information related to the wide band power generated in each of the wide band power data generation steps. Claims characterized by having Item 29: Accident sound detection program