JP2002108385A - Signal detecting method and device, and recording medium recorded with program for the method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To enable a signal detecting device to search for a similar place at a high speed even when a very long storage signal is processed. SOLUTION: Feature histograms at respective points of time are generated by deriving a feature series from the storage signal and stored (storage signal feature histogram series storage part 3), several places of interest are set for the stored feature histogram series, and the similarity between the storage signal feature histograms at the respective places of interest and a target signal feature histogram generated from a target signal is calculated (similarity calculation part 7) and compared with previously set target similarity to decide the similarity to target signals at the respective places of interest, thereby detecting a place which is similar to the target signal as an object to be detected.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and an apparatus for detecting a signal which are suitable for searching for a location of a signal similar to a signal registered in advance from a signal sequence, and a method for recording and recording the same. It relates to a recording medium.

[0002]

2. Description of the Related Art A time at which a specific commercial is broadcast is detected from a sound signal of a broadcast and automatically recorded, or a specific theme song is detected and video recording is started or stopped. For this purpose, an acoustic signal detection technique is used. With the above-described acoustic signal detection technique, it is possible to determine the position of a signal similar to a signal registered in advance from a certain signal sequence. It is also possible to automatically monitor the time when a clapping sound is emitted from the broadcast, the time when a laughter is emitted, or search for a specific scene. A typical example of the above-described acoustic signal detection technology is disclosed in Japanese Patent No. 3065314 registered on May 12, 2000.
No. "High-speed signal search method, device and recording medium".

[0003]

However, according to the above-mentioned prior art, since the accumulated signal characteristic histogram is created while performing the signal detection processing, the signal detection processing is greatly performed when the accumulated signal is extremely long. It takes a long processing time, and it is not possible to search for a similar target part at high speed. The present invention has been made in view of the above circumstances, and a means for newly accumulating a stored signal feature histogram sequence is added to eliminate a processing time involved in creating a histogram when performing signal detection processing, so that an extremely long time is obtained. It is an object of the present invention to provide a signal detection method and apparatus capable of searching for a similar part at high speed even when performing signal processing on an accumulated signal, and a recording medium on which the method is programmed and recorded.

[0004]

According to a first aspect of the present invention, there is provided a signal detecting method for searching a previously stored signal for a portion similar to a target signal. A feature sequence is derived from the accumulated signal, a feature histogram at each time point is created and stored, and several points of interest are set for the stored feature histogram sequence. The similarity between the accumulated signal feature histogram at the location and the target signal feature histogram created from the target signal is calculated, and the similarity is compared with a preset target similarity at each target location. It is characterized in that similarity with the target signal is determined, and a portion similar to the target signal to be detected is detected.

According to a second aspect of the present invention, in the signal detection method according to the first aspect, in setting several points of interest to the accumulated feature histogram series, the calculated similarity is calculated based on the similarity. A similarity upper limit value near the current position of the point of interest with respect to the accumulation signal is obtained, and a position of the next point of interest set based on the upper limit value is obtained.

According to a third aspect of the present invention, a feature sequence is derived from a previously stored signal, a time window of a predetermined length is set in the derived feature sequence, and the feature sequence within the time window is classified and counted. A first step of creating a feature histogram at each point in time of the accumulated signal, a second step of accumulating the created feature histogram sequence, and a method of extracting a feature histogram from a pre-registered target signal for signal detection. Derive a sequence, set a time window of a predetermined length in the derived characteristic sequence of the target signal, classify and count the characteristic sequence within the time window, and create a characteristic histogram at each time point of the target signal. And setting a point of interest with respect to the stored feature histogram series. A fourth step of calculating the similarity of the statistical signal feature histogram, and performing the similarity calculation on several points of interest of the accumulated signal, and comparing the similarity at the point of interest with the preset similarity. A fifth step of determining a similarity between the position of the accumulation signal and the target signal.

According to a fourth aspect of the present invention, there is provided a signal detecting apparatus for searching a stored signal stored in advance for a portion similar to a target signal. An accumulated signal extraction system that derives a feature sequence from a signal, creates a feature histogram at each point in time, and accumulates the feature histogram; sets a number of points of interest for the accumulated feature histogram sequence; , And the similarity between the target signal feature histogram created from the target signal and the target signal feature histogram calculated from the target signal. A comparison operation system for judging similarity with the signal and detecting a portion similar to the target signal to be detected. That.

According to a fifth aspect of the present invention, in the signal detection device according to the fourth aspect, the comparison operation system sets the number of points of interest in the accumulated feature histogram sequence. A calculating unit that obtains a similarity upper limit in the vicinity of the current position of the point of interest with respect to the accumulated signal from the calculated similarity, and obtains a position of the next point of interest set based on the upper limit. And

According to a sixth aspect of the present invention, there is provided an accumulated signal characteristic calculating means for deriving a characteristic sequence from an accumulated signal stored in advance, a time window of a predetermined length is set in the derived characteristic sequence, and the time window is set within the time window. Stored signal feature histogram creating means for creating a feature histogram at each time point of the accumulated signal by classifying and counting the feature series of the accumulated signal; accumulated signal feature histogram sequence accumulating means for accumulating the created feature histogram series; Target signal feature calculation means for deriving a feature sequence from the target signal for signal detection, setting a time window of a predetermined length in the feature sequence of the derived target signal, and classifying and counting the feature sequence within the time window. Signal histogram creating means for creating a feature histogram at each time point of the target signal The attention point setting means for setting a target position relative to the stored feature histogram series,
A similarity calculating means for calculating the similarity between the stored signal feature histogram and the target signal feature histogram at the noted location, and performing the similarity calculation on some noted locations of the stored signal, By comparing the similarity with a preset similarity,
A similarity determination unit configured to determine a similarity between the location of the accumulation signal and the target signal.

According to a seventh aspect of the present invention, in the signal detection device according to the sixth aspect, the accumulated signal characteristic histogram creating means and / or the target signal characteristic histogram creating means is the accumulated signal characteristic calculating means. And / or reads a time series of feature vectors output from the target signal feature calculation means, divides each element of the feature vector into n sections, and when the number of dimensions of each feature vector is m, n Classification into one of the m-th section, a time window of an arbitrary length is set at the head of the accumulated signal and / or the target signal, and the number of feature vectors for each classification within the time window is determined. The above operation is repeated while sequentially moving the time window in the time axis direction, and a time series of a histogram is created. According to the above configuration, a means for newly accumulating a histogram series of accumulated signal features is added, and processing time related to histogram creation when signal detection is performed is eliminated.
Even when it is desired to perform signal detection for an extremely long accumulated signal, high-speed signal detection is possible.

According to an eighth aspect of the present invention, in the signal detection device according to the sixth or seventh aspect, the attention point setting means obtains the movable width ω of the attention point by calculating the following equation. It is characterized by.

(Equation 3) Here, it is assumed that θ is the similarity threshold, S is the accumulated signal similarity value, D is the total frequency of the feature amount sequence, and floor (•) represents the rounding down to the decimal point.

According to a ninth aspect of the present invention, in the signal detection device according to any one of the sixth to eighth aspects, the similarity calculating section obtains the similarity by calculating the following equation.

(Equation 4) Here, L indicates the total number of feature vector classifications, D indicates the total frequency of the histogram, and h _Rl and h _Il indicate the number of feature vectors included in the l-th classification of the target signal feature histogram and the accumulated signal feature histogram, respectively. Shall be.

According to a tenth aspect of the present invention, there is provided a computer readable program for recording a signal detection program used in a signal detection device for detecting a portion similar to a target signal from a stored signal stored in advance. In the recording medium, the signal detection program includes a first step of deriving a feature sequence from a stored signal stored in advance, a time window having a predetermined length set in the derived feature sequence, and a feature sequence within the time window. A second step of creating a feature histogram at each time point of the accumulated signal by classifying and counting the data, a third step of accumulating the created feature histogram sequence, and a pre-registered object for signal detection. A fourth step of deriving a feature sequence from the signal; A fifth step of setting a time window of the target signal and classifying and counting the feature sequence within the time window to generate a feature histogram at each point of the target signal; , A seventh step of calculating the degree of similarity between the accumulated signal feature histogram at the point of interest and the target signal feature histogram, and accumulation by repeating the processing of the sixth and seventh steps. The similarity calculation is performed for some attention points of the signal, and the similarity between the attention signal and the target signal is compared by comparing the similarity at the attention point with a preset similarity. A signal detection program for causing a computer to execute the determining step and the eighth step is recorded.

According to an eleventh aspect of the present invention, in the recording medium according to the tenth aspect, the sixth step is a step of determining whether or not the current point of interest with respect to the stored signal is based on the similarity calculated in the seventh step. The method further includes a step of obtaining a similarity upper limit value near the position, and a step of obtaining a position of a target point to be set next based on the upper limit value.

[0015]

FIG. 1 is a block diagram showing an embodiment of a signal detection device according to the present invention. Here, a description will be given of a case where a signal detection target is an acoustic signal. The signal detection device according to the present invention can be roughly classified into a stored signal feature extraction system A and a comparison operation system B that performs a comparison operation between the stored signal and the target signal. The accumulated signal feature extraction system A
It comprises an accumulated signal feature calculation unit 1, an accumulated signal feature histogram creation unit 2, and an accumulated signal feature histogram sequence accumulation unit 3. The comparison operation system B includes a target signal feature calculation unit 4 and
It is composed of a target signal feature histogram creating unit 5, an attention point setting unit 6, a similarity calculating unit 7, and a similarity determining unit 8.

The signal detection apparatus of the present invention comprises a target signal, that is, an audio signal to be searched as a sample, and a stored signal, that is, a predetermined long-term sound having a possibility of including the target signal as a part thereof. A signal is obtained as an input, and the degree of similarity with the target signal is set to a preset value θ (search threshold).
And outputs an arbitrary point in the accumulated signal exceeding the value. The signal detection device illustrated in FIG. 1 is specifically configured by a computer device including a CPU and a peripheral LSI therefor. This computer device functions as each block shown in FIG. 1 when a CPU reads and executes a control program recorded on a predetermined recording medium (semiconductor memory, magnetic disk, optical memory, or the like). This control program can be distributed via a communication line.

In FIG. 1, the accumulated signal feature calculation unit 1
A feature sequence is derived from an externally obtained accumulated signal. The accumulated signal feature histogram creating section 2 includes an accumulated signal feature calculating section 1
A time window of a predetermined length is set in the feature sequence of the accumulated signal output by the above, and a feature histogram at each time point of the accumulated signal is created by classifying and counting the feature sequence in the time window. , And outputs it to the accumulation signal feature histogram sequence accumulation unit 3. The accumulated signal feature histogram sequence accumulation unit 3 accumulates the sequence of the feature histogram created by the accumulated signal feature histogram creating unit 2.

The target signal feature calculation unit 4 derives a feature sequence from a previously registered target signal. The target signal feature histogram creating unit 5 sets a predetermined time window of a predetermined length in the feature sequence of the target signal output by the target signal feature calculation unit 4, and classifies and counts the feature sequence within the time window. Thus, a histogram of the feature related to the target signal is created and output to the attention point setting unit 6.

As will be described later, the point-of-interest setting section 6 has a built-in operation section (not shown), and sets the point-of-interest for the accumulated signal feature histogram sequence stored in the accumulated signal feature histogram sequence accumulating section 3. Then, the stored signal feature histogram at the point of interest is output to the similarity calculator 7.
The similarity calculation unit 7 includes a target signal feature histogram creation unit 5
With respect to the target signal feature histogram created in step (1) and the accumulated signal feature histogram at the point of interest set by the point of interest setting unit 6, the similarity indicating the degree of similarity between them is calculated and output to the similarity determination unit 8. At the same time, the result of the similarity calculation is output to the attention point setting unit 6 to set the next attention point. The similarity determination unit 8 performs the similarity calculation for some points of interest of the accumulated signal, and compares the similarity at the points of interest with a preset similarity (θ: search threshold), A similarity between the location of the accumulation signal and the target signal is determined and output as a signal detection result.

FIG. 2 is a flowchart cited for explaining the operation of the embodiment of the present invention. Hereinafter, the operation of the embodiment of the present invention shown in FIG. 1 will be described in detail with reference to the flowchart shown in FIG. First, the accumulated signal feature calculation unit 1 reads the given accumulated signal (step S11). Next, feature extraction is performed on the read stored signal (step S12). Here, since a spectral feature is used as the feature, the feature extraction can be performed by, for example, a band-pass filter.

For example, when it is desired to search for an acoustic signal of about 15 seconds, an expected result can be obtained if the specific setting of feature extraction is as follows. That is, seven band-pass filters are used, their center frequencies are set at equal intervals in the range of 300 Hz to 3500 Hz on a logarithmic axis, and a time window having a time length of about 60 ms is moved by 10 ms. Meanwhile, the average value of the square of the output of each band-pass filter within this time window is calculated, and the obtained seven values are grouped into a seven-dimensional feature vector. In this case, one feature vector is obtained every 10 ms. In addition, by using the image feature, it can be used for searching for an image. In this case, the result expected to divide one frame image of the video into four equal parts in the horizontal direction and three equal parts in the vertical direction, and to obtain a total of 36-dimensional feature vectors with the average of the RGB values in each division as a feature can get.

First, the accumulated signal feature histogram creating section 2 reads the time series of the feature vectors of the accumulated signal output from the accumulated signal feature calculating section 1 and creates a feature vector histogram (step S13).
Specifically, first, feature vectors are classified. For example,
The value of each element of the feature vector is divided into several bins (sections). That is, for example, if each element is divided into three sections and the number of dimensions of each feature vector is 7, the total number of bins, that is, the number of sections 3 arranged on the horizontal axis of the histogram is the seventh power. . Therefore, each feature vector is classified into any one of the 3 7 sections. Next, a time window having the same length as the target signal is set at the head of the accumulated signal, and the number of feature vectors for each classification in the time window is counted, thereby creating a histogram of feature vectors.

The creation of the histogram is repeated while sequentially moving the time window in the direction of the time axis of the accumulated signal, thereby obtaining a time series of the histogram. Here, assuming that the movement width of the time window is z, z may be, for example, a time corresponding to one feature vector. The accumulated signal feature histogram sequence created in this way is output to the accumulated signal feature histogram sequence accumulation unit 3. Note that the method of classifying feature vectors is not limited to the above-described method, and it is also possible to use a vector quantization technique. Further, the length of the time window does not necessarily have to be the same as the length of the target signal. Further, the moving width z of the time window does not necessarily need to be one by one feature vector, and may be moved by a time corresponding to a plurality of feature vectors at a time so as not to impair the accuracy of signal detection. .

Stored signal feature histogram sequence storage unit 3
Reads a series of accumulated signal feature histograms output from the accumulated signal feature histogram creating unit 2, records and stores the series on a built-in recording medium (step S1).
4). As described above, it is a main object of the present invention to record and accumulate a series of accumulated signal feature histograms on a recording medium prior to signal detection processing for a specific target signal,
As a result, there is a unique effect that the signal detection processing after the specific target signal is given is remarkably speeded up. Details will be described later.

First, the target signal feature calculator 4 reads a target signal which is a sample of an audio signal to be detected (step S15). Next, feature extraction is performed on the read target signal (step S16). The feature extraction performs the same operation as previously performed in the accumulated signal feature calculation unit 1. First, the target signal feature histogram creating unit 5 reads the time series of feature vectors for the target signal output from the target signal feature calculating unit 4, and creates a histogram of the feature vector (step S17).
The creation of the histogram is performed in the same manner as the accumulated signal feature histogram creation unit 2.

At the beginning of the signal detection process, the point-of-interest setting section 6 sets the beginning of the accumulated signal feature histogram accumulated in the accumulated signal feature histogram sequence accumulation section 3 as:
A point of interest is set, and the accumulated signal feature histogram at the point of interest is output to the similarity calculator 7 (step S).
18). In similarity calculation unit 7, as described later, the stored signal feature histogram in the target location, but the degree of similarity is calculated S _RI of the target signal wherein the histogram (Step S
19) At that time, the attention point setting unit 6 receives the calculated similarity value _SRI from the similarity calculation unit 7 (Step S).
20). Based on the similarity value, the attention point setting unit 6
The skippable width ω, which is a value indicating at which position the point of interest can be set at the next time point, is calculated (step S21).
Built-in arithmetic unit. The skippable value ω calculated by the operation unit is obtained by the following equation.

[0027]

(Equation 5) However, the unit of the skippable value ω is the number of feature vectors, that is, the time in units of the time interval between adjacent feature vectors in the feature vector sequence.
(•) indicates devaluation. Therefore, when the accumulated signal feature histogram creating unit 2 sets z = 1 (that is, creates one histogram for each feature vector), the value of ω is not changed and the number of histograms is used as a unit. It indicates the amount of movement of the point of interest at this time.

When the histogram is created while moving by a plurality of feature vectors, it is easily understood that a value obtained by appropriately converting the value of the stepable width ω indicates the movement amount of the point of interest. For example, if one histogram is created for every n feature vectors, the movement amount of the point of interest is fl
oor (ω / n) +1. On the other hand, if S _RI ≧ θ, ω = 1 (full search) to find a local peak of similarity. That is, the attention point setting unit 6 includes the similarity calculation unit 7
The similarity upper limit in the vicinity of the current position of the point of interest with respect to the stored signal is obtained from the similarity calculated by the above, and the position of the next point of interest set next is obtained based on the upper limit.

The attention point setting unit 6 outputs the position of the next attention point determined as described above to the similarity calculation unit 7 and thereafter sequentially reaches the end of the accumulated signal (step S22). ), The similarity calculation process by the similarity calculation unit 7 (step S19), and the process of calculating the skippable width ω by the attention point setting unit 6 (step S20). The similarity calculation unit 7 first reads the target signal feature histogram output from the target signal feature histogram creation unit 5 and the accumulated signal feature histogram output from the attention point setting unit 6 and indicated by the attention point. Target signal feature histogram H _R outputted from the target signal feature histogram forming unit 5, a storage signal feature histogram showing the attention point output from the attention point setting unit 6 and H _I. Here, R and I indicate that the histogram is created from the feature vector of the target signal and the feature vector of the accumulated signal, respectively.

The calculation of the similarity _SRI can be performed by the following equation.

(Equation 6) Here, L is the total number of feature vector classifications (here, 3 to the seventh power), D is the total frequency of the histogram (total number of feature vectors derived from one target signal), and h _RI , h
_Il represents each H _R, the number of feature vectors included in l th classification H _I (degrees). The similarity calculation unit 7 outputs the value of the similarity to the attention point setting unit 6 and simultaneously outputs the value to the similarity determination unit 8. When the value of the similarity exceeds a preset value (threshold), the similarity determination unit 8 means that the target signal is present in the accumulated signal. The number of the target signal and the current position of the point of interest for the stored signal are output (step S23).

FIG. 3 shows an experimental example of the operation of the signal detecting device according to the present invention. FIG. 3 is a diagram showing a required time for signal detection when the present invention is applied in a table format for comparison with a conventional example. Here, a sound signal for 20 hours is set as a storage signal, 200 kinds of 10-second sound signals are set as target signals, and a time required for detecting whether or not the target signal exists in the stored signal is set as a personal time. Measured by computer. FIG. 3 shows the time required for signal detection based on the presence or absence of the accumulated signal feature histogram sequence accumulation unit 3. That is, (a) shows a conventional example having no accumulated signal feature histogram sequence accumulation unit 3,
(B) shows the time required for signal detection according to the present invention having the accumulated signal feature histogram sequence accumulation unit 3. As is apparent from FIG. 3, according to the present invention, a high-speed processing of 5 times or more is realized.

As described above, according to the present invention, means for newly accumulating a histogram series of accumulated signal features is added, and the processing time involved in creating a histogram when performing signal detection is eliminated. Even when it is desired to perform signal detection for an extremely long accumulated signal, high-speed signal detection is possible. In the present invention, only the acoustic signal has been exemplified as a target of signal detection. However, the present invention is not limited to this. In the same configuration as the case of the acoustic signal, the accumulated signal feature calculating unit 1 and the accumulated signal feature histogram creating unit 2
In, it is also possible to search for a video signal by using video features such as color features based on the video.

[0033]

As described above, according to the present invention, a means for newly accumulating the accumulated signal characteristic histogram sequence is added, and the processing relating to the histogram creation at the time of performing the signal detection processing is eliminated. Even when performing signal processing on a long accumulated signal, a similar part can be searched at high speed. As a result, it is possible to detect the time when a specific commercial aired from the audio signal of the broadcast and automatically record it, or to start and stop video recording by detecting a specific theme song become. It is also possible to automatically monitor the time when a clapping sound is emitted from the broadcast, the time when a laughter is emitted, and the like, and to search for a specific scene. Further, the present invention can be applied to detection of not only an acoustic signal but also a general signal such as an image pattern.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating an embodiment of a signal detection device according to the present invention.

FIG. 2 is a flowchart showing the operation of the embodiment of the present invention.

FIG. 3 is a diagram showing a time required for signal detection when the present invention is applied in a table format for comparison with a conventional example.

[Explanation of symbols]

 Reference Signs List 1 accumulated signal feature calculation unit 2 accumulated signal feature histogram creation unit 3 accumulated signal feature histogram sequence accumulation unit 4 target signal feature calculation unit 5 target signal feature histogram creation unit 6 attention point setting unit 7 similarity calculation unit 8 similarity determination unit

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat ゛ (Reference) G10L 3/00 531C (72) Inventor Hiroshi Murase 2-3-1 Otemachi, Chiyoda-ku, Tokyo Nippon Telegraph and Telephone Telephone Co., Ltd. F-term (reference) 5C025 BA11 DA08 5C026 DA27 DA28 5D015 AA06 EE00 HH04

Claims (11)

[Claims] 1. A signal detection method for searching for a location similar to a target signal from a stored signal stored in advance, wherein a feature sequence is derived from the stored signal to create a feature histogram at each time point. Then, a number of points of interest are set for the stored feature histogram sequence, and a target signal characteristic histogram created from the target signal and a stored signal characteristic histogram at each of the target points is set. The similarity is calculated, and the similarity is compared with a preset target similarity to determine the similarity with the target signal at each point of interest, and a portion similar to the target signal to be detected is determined. A signal detection method characterized by detecting. 2. When setting some points of interest in the stored feature histogram series, a similarity upper limit near the current position of the point of interest with respect to the accumulated signal is obtained from the calculated similarity. 2. The signal detecting method according to claim 1, wherein a position of a target portion set next is determined based on the upper limit value. 3. A feature sequence is derived from a stored signal stored in advance, a time window having a predetermined length is set in the derived feature sequence, and the feature sequence within the time window is classified and counted to thereby calculate the value of the stored signal. A first step of creating a feature histogram at each point in time, a second step of storing the created feature histogram sequence, and a feature sequence derived from a preregistered target signal for signal detection. A third step of setting a time window of a predetermined length in the feature sequence of the target signal and creating a feature histogram at each time point of the target signal by classifying and counting the feature sequence in the time window; Attention points are set for the feature histogram series obtained, and the accumulated signal feature histogram and the target signal feature histogram at the attention points are set. A fourth step of calculating the similarity of the stored signals; and performing the similarity calculation on some points of interest of the accumulated signal, and comparing the similarity at the point of interest with a preset similarity. A fifth step of determining a similarity between the location of the stored signal and the target signal. 4. A signal detecting device for searching a point similar to a target signal from a stored signal stored in advance, wherein a characteristic sequence is derived from the stored signal to generate a characteristic histogram at each time point. An accumulation signal extraction system for accumulating the same, setting a number of points of interest with respect to the stored feature histogram sequence, an accumulation signal characteristic histogram at each of the points of interest, and an object generated from the object signal. The similarity with the signal feature histogram is calculated, the similarity is compared with a preset target similarity to determine the similarity with the target signal at each point of interest, and the target signal to be detected And a comparison operation system for detecting a portion similar to the above. 5. The comparison operation system according to claim 1, further comprising: setting a plurality of points of interest with respect to the accumulated feature histogram sequence, based on the calculated similarity, in a vicinity of a current position of the point of interest with respect to the accumulated signal. 5. The signal detection device according to claim 4, further comprising a calculation unit that obtains a similarity upper limit value, and obtains a position of a target portion set next based on the upper limit value. 6. A stored signal feature calculating means for deriving a feature sequence from a previously stored stored signal, a time window having a predetermined length is set in the derived feature sequence, and the feature sequence within the time window is classified and counted. Storing signal feature histogram creating means for creating a feature histogram at each time point of the accumulated signal; storing signal feature histogram sequence storing means for storing the created feature histogram sequence; Target signal feature calculation means for deriving a feature sequence from a target signal, and setting a time window of a predetermined length in the feature sequence of the derived target signal, and classifying and counting the feature sequence within the time window to thereby calculate the target signal. Target signal feature histogram creating means for creating a feature histogram at each point in time; Attention point setting means for setting an attention point for the histogram sequence; similarity degree calculation means for calculating the similarity between the accumulation signal feature histogram and the target signal characteristic histogram at the attention point; The similarity calculation is performed on the point of interest, and the similarity at the point of interest is compared with a preset similarity to determine the similarity between the point of the accumulated signal and the target signal. Means, and a signal detection device. 7. The accumulated signal feature histogram creating unit and / or the target signal feature histogram creating unit may include a time series of a feature vector output from the accumulated signal feature calculating unit and / or the target signal feature calculating unit. Is read, and each element of the feature vector is divided into n sections. When the number of dimensions of each feature vector is m, m of n
Classification into one of the multiplied sections, setting a time window of an arbitrary length at the beginning of the accumulated signal and / or the target signal, and counting the number of feature vectors for each classification within the time window 7. The signal detection apparatus according to claim 6, wherein the above operation is repeated while sequentially moving the time window in the time axis direction to generate a time series of a histogram. 8. The signal detection device according to claim 6, wherein the point of interest setting means obtains a movable width ω of the point of interest by calculating the following equation. (Equation 1) Here, it is assumed that θ is the similarity threshold, S is the accumulated signal similarity value, D is the total frequency of the feature amount sequence, and floor (•) represents the rounding down to the decimal point. 9. The signal detecting apparatus according to claim 6, wherein the similarity calculating unit obtains the similarity by calculating the following equation. (Equation 2) Here, L indicates the total number of feature vector classifications, D indicates the total frequency of the histogram, and h _Rl and h _Il indicate the number of feature vectors included in the l-th classification of the target signal feature histogram and the accumulated signal feature histogram, respectively. Shall be. 10. A computer-readable recording medium in which a signal detection program used for a signal detection device for detecting a portion similar to a target signal out of pre-stored stored signals is recorded. The program derives a feature sequence from a stored signal stored in advance.
A second step of setting a time window of a predetermined length in the derived feature sequence, and creating a feature histogram at each time point of the accumulated signal by classifying and counting the feature sequence within the time window. A third step of accumulating the created feature histogram sequence, a fourth step of deriving a feature sequence from a pre-registered target signal for signal detection, and a predetermined step in the derived feature signal sequence of the target signal. A fifth step of setting a long time window and classifying and counting the feature sequence within the time window to create a feature histogram at each time point of the target signal; A sixth step of setting a location; a stored signal feature histogram and the target signal feature histogram at the noted location. A seventh step of calculating the similarity of the stored signals; and repeating the processing of the sixth and seventh steps to perform the similarity calculation for some target points of the accumulated signal, A signal detection program for causing a computer to execute an eighth step of determining the similarity between the location of the stored signal and the target signal by comparing the degree of similarity and a preset degree of similarity. 11. The sixth step is a step of obtaining a similarity upper limit near the current position of the point of interest with respect to the accumulation signal from the similarity calculated in the seventh step, based on the upper limit. 11. The recording medium according to claim 10, further comprising the step of: determining a position of a target point to be set next.