JP2012168647A - Device and method for analysis of walking sound, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To determine walker's characteristics based on a walking sound.SOLUTION: A walking sound processing part 15A applies acoustic signal processing to a walking sound signal 22, and extracts a sound-generating position and sound-generating time where and when each footstep is generated. A parameter calculation part 15B calculates walking-state parameter values showing characteristics about a walking state of an object person, based on the sound-generating position and sound-generating time of each footstep. A category determination part 15C acquires a probability value corresponding to each walking-state parameter value of the object person from a probability distribution DB 14A by classification category, integrates the probability values, and calculates an evaluation value by classification category, and then determines a classification category, for which a highest evaluation value is obtained, as a classification category concerning the object person.

Description

  The present invention relates to a walking sound analysis technique for estimating characteristics of a target person and identifying and authenticating the target person by analyzing various sounds generated with walking.

  Conventionally, a technique for analyzing a walking sound detected by a microphone array and extracting a sound generation position and a sound generation time at which a footstep sound has been proposed (see, for example, Non-Patent Document 1 and Non-Patent Document 2). This is based on a technique for estimating a sound source position using a difference in arrival time of sound with respect to a microphone pair. Thereby, a pedestrian's step length, walking speed, etc. are computable.

Masanari Shoji, "Passive Acoustic Sensing of Walking", In Proceedings of the International Conference on Intelligent Sensors, Sensor Networks and Information Processing (ISSNIP '09), pp. 219-224, 2009. Shoji, “Study on footstep localization and tracking using microphone array,” Acoustical Society of Japan 2010 Spring Conference, Proceedings pp.727-730, 2010. Nomura et al., “Nearby Field Microphone Array”, Journal of the Acoustical Society of Japan, Vol.53, No.2, pp.110-116, 1997

In general, walking sounds, including footsteps that occur as a person walks, include characteristics unique to each person, such as differences in walking sounds caused by habits of walking, and characteristics associated with each person's clothes and belongings. It is known from daily experience. In some cases, a person's footsteps can be heard to intuitively determine who the person is.
However, with the above-described conventional technology, although the pedestrian's stride and walking speed can be quantitatively evaluated, physical features such as height, weight, sex, and age, and motor functional features such as gait disturbance and specific diseases, etc. In addition, it is impossible to determine a feature of a person from a walking sound such as a feature unique to each person in walking.

  The present invention has been made to solve such a problem, and an object thereof is to provide a walking sound analysis technique capable of determining a human characteristic based on a walking sound.

  In order to achieve such an object, the walking sound analysis apparatus according to the present invention performs acoustic signal processing on a walking sound signal including a footstep sound that occurs as the subject walks, thereby generating a sound generation position where each footstep sound is generated. And a walking sound processing unit for extracting the sound generation time, and a walking state parameter value indicating characteristics regarding the walking state of the subject based on the sound generation position and the sound generation time of each footstep obtained by the walking sound processing unit. A parameter calculation unit to be calculated and a probability distribution characteristic indicating a frequency distribution of walking state parameter values obtained from footsteps of a person belonging to the classification category for each classification category of the person set in advance according to the characteristics of the person The probability distribution database and the probability values corresponding to the walking state parameter values of the subject obtained by the parameter calculation unit are acquired from the probability distribution database for each classification category and integrated. And by, calculating an evaluation value for each classification category, and a determining category determination unit as classification categories related subjects the classification category which the highest evaluation value is obtained.

  At this time, for the footstep signal indicating the footsteps extracted from the walking sound signal, the walking sound processing portion calculates the feature values indicating the acoustic features relating to the footstep signal, and the parameter calculation portion obtains the walking sound processing portion. Statistical values for each feature value are calculated as the footstep parameter values of the target person, and the frequency distribution of each footstep parameter value obtained from the footsteps of the person belonging to the classification category is shown for each classification category in the probability distribution database. When storing probability distribution characteristics and calculating an evaluation value for each classification category in the category determination unit, obtain probability values corresponding to each footstep parameter value of the target person from the probability distribution database for each category category You may make it integrate each into the evaluation value of the classification division to do.

  In addition, for the footstep signal indicating the footsteps extracted from the walking sound signal in the walking sound processing unit, the feature values indicating the acoustic features related to the footstep signal are respectively calculated, and the parameter calculation unit obtained by the walking sound processing unit The statistical value for each feature value is calculated as the footstep parameter value of the target person, and the correspondence between the footwear type related to the footwear worn by the person and the footstep parameter value obtained from the footstep sound of the person wearing the footwear of the footwear type A footwear type database that further includes a footwear type determination unit that refers to the footwear type database and determines the footwear type corresponding to the footstep parameter value of the target person obtained by the parameter calculation unit as the footwear type of the target person. Prepared in the probability distribution database, the probability distribution indicating the frequency distribution for each footwear type for footwear worn by persons belonging to that category. The characteristics are further stored, and when the category determination unit calculates an evaluation value for each classification category, the probability value corresponding to the footwear type of the target person obtained by the footwear type determination unit is obtained from the probability distribution database for each classification category. It may be acquired and integrated into the evaluation value of the corresponding classification category.

  In addition, the walking sound processing unit collects the walking sound signal to generate observation point signals indicating the sound generated at the observation point for each observation point provided around the subject. For each point signal, an observation point feature value indicating an acoustic feature related to the observation point signal is calculated, and in the parameter calculation unit, a statistical value regarding each observation point feature value obtained by the walking sound processing unit is calculated for each observation point. Correspondence between the incident type for the incidental that accompanies the person and each observation point parameter value obtained from the observation point signal of the person who accompanies the incidental of the incident type An accessory type database indicating the relationship is further provided, and the accessory type corresponding to the observation point parameter value of the subject obtained by the parameter calculation unit is referred to as the accessory type of the subject with reference to the accessory type database. A probability distribution characteristic indicating a frequency distribution for each accessory type with respect to an accessory attached by a person belonging to the classification category for each classification category in the probability distribution database. When the evaluation value is calculated for each classification category by the category determination unit, the probability value corresponding to the target person's accessory type is obtained from the probability distribution database for each classification category and converted to the evaluation value of the corresponding classification category. Each may be integrated.

  Further, in the parameter calculation unit, the statistic on at least one of the step length of the subject, the walking speed, the opening width from the moving center line in the left-right direction, and the time interval of one step as the walking state parameter value. You may make it calculate the feature information which shows the acoustic feature of the said footstep signal extracted from the quantity or the spectrogram of the footstep signal.

  In addition, among the evaluation values calculated for each classification category by the category determination unit, the classification category in which the highest evaluation value is obtained from the evaluation values exceeding the preset threshold value for each classification category, or You may make it determine the classification division with the largest difference or ratio of the said evaluation value and a threshold value as a classification division regarding a subject.

  Further, the walking sound analysis method according to the present invention is a walking sound analysis method used in a walking sound analysis apparatus that analyzes a walking sound signal including footsteps generated along with walking of the subject and determines the characteristics of the subject. The walking sound processing unit extracts a sound generation position and a sound generation time at which each footstep sound is generated by performing an acoustic signal processing on the walking sound signal including the footstep sound generated with the walking of the subject. And a parameter calculating step in which the parameter calculating unit calculates a walking state parameter value indicating characteristics relating to the walking state of the subject based on the sound generation position and sound generation time of each footstep obtained by the walking sound processing unit; A probability distribution characteristic indicating a frequency distribution of walking state parameter values obtained from footsteps of a person belonging to the classification category for each classification category of the person set in advance according to the characteristics of the person The step of storing as a probability distribution database and the classification determining unit obtains and integrates the probability value corresponding to the walking state parameter value of the subject obtained by the parameter calculation unit from the probability distribution database for each classification category. A classification determination step of calculating an evaluation value for each classification category and determining the classification category with the highest evaluation value as the classification category for the subject.

  At this time, in the walking sound processing step, for the footstep signal indicating the footstep extracted from the walking sound signal, the feature value indicating the acoustic feature related to the footstep signal is calculated, respectively, and obtained in the parameter calculation step by the walking sound processing unit. Statistical values for each feature value are calculated as the footstep parameter values of the target person, and the frequency distribution of each footstep parameter value obtained from the footsteps of the person belonging to the classification category is shown for each classification category in the probability distribution database. When storing probability distribution characteristics and calculating evaluation values for each category in the category determination step, the probability values corresponding to each footstep parameter value of the target person are obtained from the probability distribution database for each category and You may make it integrate each into the evaluation value of the classification division to do.

  Further, in the walking sound processing step, for the footstep signal indicating the footstep extracted from the walking sound signal, a feature value indicating an acoustic feature regarding the footstep signal is calculated, and the parameter calculation step is obtained by the walking sound processing unit. Statistical values for each feature value are calculated as footstep parameter values of the target person, and each footstep parameter value obtained from the footwear type of the footwear worn by the person and the footsteps of the person wearing the footwear of the footwear type by the storage unit A step of storing a footwear type database indicating a correspondence relationship between the footwear type database and the footwear type determining unit with reference to the footwear type database to determine the footwear type corresponding to the footstep parameter value of the subject obtained by the parameter calculating unit. It further includes a footwear type determination step for determining the footwear type of the target person, and for each classification category in the probability distribution database, the classification category Further store probability distribution characteristics indicating the frequency distribution for each footwear type related to the footwear worn by the person to which the person belongs, and when calculating the evaluation value for each classification category in the category judgment step, the target person obtained by the footwear type judgment unit The probability value corresponding to the footwear type may be acquired from the probability distribution database for each category and integrated with the evaluation value of the corresponding category.

  Also, in the walking sound processing step, by collecting the walking sound signal, an observation point signal indicating the sound generated at the observation point is generated for each observation point provided around the subject person. For each point signal, an observation point feature value indicating an acoustic feature related to the observation point signal is calculated, and in the parameter calculation step, a statistical value regarding each observation point feature value obtained by the walking sound processing unit is calculated for each observation point. Each observation point parameter calculated from the observation point parameter value of the target person, and the storage unit obtained from the incidental type related to the incidental that accompanies the person and the observation point signal of the person accompanying the incidental of the incidental type A step of storing an accessory type database indicating a correspondence relationship with the value, wherein the accessory type determination unit refers to the accessory type database and observes the subject obtained by the parameter calculation unit An accessory type determination step for determining an accessory type corresponding to the parameter value as an accessory type of the subject is further provided, and in the probability distribution database, for each accessory category, an accessory attached by a person belonging to the class category, The probability distribution characteristic indicating the frequency distribution for each accessory type is further stored, and when the evaluation value is calculated for each classification category in the category determination step, the probability value corresponding to the target category of the target person is calculated for each category category. It may be acquired from the probability distribution database and integrated into the evaluation value of the corresponding classification category.

  Further, in the parameter calculation step, as the walking state parameter value, at least one of a subject's step length, walking speed, an opening width from the moving center line in the left-right direction, and a statistics regarding one or more time intervals of one step You may make it calculate the feature information which shows the acoustic feature of the said footstep signal extracted from the quantity or the spectrogram of the footstep signal.

  Further, among the evaluation values calculated for each classification category in the category determination step, among the evaluation values exceeding a preset threshold value for each classification category, the classification category having the highest evaluation value, or You may make it determine the classification division with the largest difference or ratio of the said evaluation value and a threshold value as a classification division regarding a subject.

  Moreover, the program concerning this invention is a program for functioning a computer as each part which comprises the walking sound analyzer as described in any one mentioned above.

  According to the present invention, on the basis of walking sounds including footsteps accompanying walking, the characteristics of the subject, for example, physical characteristics such as height, weight, weight, sex, age, and exercise such as walking disorders and specific diseases For functional features, it can be determined which of the classification categories preset for these features belongs.

It is a block diagram which shows the structure of the walking sound analyzer concerning 1st Embodiment. It is explanatory drawing which shows the stride and walking speed of a walk. It is a flowchart which shows the walk sound analysis process of the walk sound analyzer concerning 1st Embodiment. It is an example of a probability distribution characteristic regarding stride and height. It is an example of calculation of an evaluation value. It is a probability distribution characteristic example regarding a stride and a registrant. It is a flowchart which shows the walk sound analysis process of the walk sound analyzer concerning 2nd Embodiment. It is an example of calculation of an evaluation value. It is explanatory drawing which shows extraction of a footstep signal. It is an amplitude waveform diagram of a footstep signal. It is a spectrogram of a footstep signal. It is a flowchart which shows the walk sound analysis process of the walk sound analyzer concerning 3rd Embodiment. It is a block diagram which shows the walking sound analyzer concerning 4th Embodiment. It is a flowchart which shows the walk sound analysis process of the walk sound analyzer concerning 4th Embodiment. It is a block diagram which shows the walking sound analyzer concerning 5th Embodiment. It is explanatory drawing which shows a subject's observation area. It is a flowchart which shows the walk sound analysis process of the walk sound analyzer concerning 5th Embodiment.

Next, embodiments of the present invention will be described with reference to the drawings.
[First Embodiment]
First, a walking sound analyzer according to a first embodiment of the present invention will be described with reference to FIG. FIG. 1 is a block diagram illustrating a configuration of the walking sound analyzer according to the first embodiment.

This walking sound analysis device 10 is composed of an information processing device such as a personal computer or a server device as a whole, and receives a walking sound signal 22 including footsteps generated as the subject walks according to the characteristics of a person in advance. It has a function of determining the classification category to which the target person belongs from among the classification categories of the set person.
In FIG. 1, the sound collecting device 20 has a function of collecting output signals from a plurality of microphones 21 constituting a microphone array, performing A / D conversion processing, and outputting the signals to the walking sound analyzing device 10 as a walking sound signal 22. ing.

  In this embodiment, the sound generation position and sound generation time at which each footstep sound is extracted by performing acoustic signal processing on the footstep sound signal including the footstep footage that occurs as the subject walks, and each footstep sound obtained is obtained. On the basis of the sound generation position and sound generation time, a walking state parameter value indicating characteristics relating to the walking state of the target person is calculated, and for each person classification category set in advance according to the characteristics of the person, The probability distribution characteristic indicating the frequency distribution of the walking state parameter value obtained from the footsteps of the belonging person is stored, and the probability value corresponding to the obtained walking state parameter value of the target person is obtained from the probability distribution database for each classification category. By obtaining and integrating, an evaluation value is calculated for each classification category, and the classification category with the highest evaluation value is determined as the classification category for the subject.

Next, the configuration of the walking sound analyzer 10 according to the present embodiment will be described in detail with reference to FIG.
The walking sound analyzer 10 is provided with a signal input unit 11, an operation input unit 12, a screen display unit 13, a storage unit 14, and an arithmetic processing unit 15 as main functional units.

The signal input unit 11 includes a dedicated input / output interface circuit, and has a function of acquiring the walking sound signal 22 from the sound collecting device 20 and outputting it to the arithmetic processing unit 15.
The operation input unit 12 includes an operation input device such as a keyboard and a mouse, and has a function of detecting an operator operation and outputting the operation to the arithmetic processing unit 15.
The screen display unit 13 includes a screen display device such as an LCD, and has a function of displaying various data such as operation menus and analysis results output from the arithmetic processing unit 15 on the screen.

The storage unit 14 includes a storage device such as a hard disk or a semiconductor memory, and has a function of storing processing information and programs 14P used for various processing operations in the arithmetic processing unit 15. The program can be recorded on a recording medium or provided through a network.
As main processing information stored in the storage unit 14, there is a probability distribution database (hereinafter referred to as probability distribution DB) 14A. The probability distribution DB 14A shows a frequency distribution of arbitrary parameter values indicating the characteristics of the walking, obtained from the footsteps of the person belonging to the classification category, for each classification category of the person set in advance according to the characteristics of the person. Probability distribution characteristics are registered.

  For example, when a classification category of 160 cm to 180 cm is set as one of the height classification categories, parameter values can be extracted from the footsteps of a plurality of people belonging to this classification category. Specific examples of this parameter value are values specific to each person, such as stride and walking speed. Therefore, even if people belong to the same classification category, the stride and walking speed are different, but both have a certain degree of correlation. For example, a tall person has a relatively large stride. Therefore, the parameter values extracted from the footsteps of a plurality of people belonging to the same classification category are distributed with a certain peak.

  By normalizing such frequency distribution and integrating it for each section set as the parameter value, the parameter value can be expressed as a ratio of all persons belonging to the same classification category, that is, a probability value. . Therefore, if the probability values of the parameter values calculated from the footsteps of the subject person are acquired and compared for each category, it is possible to determine the likely category to which the subject belongs. In the present embodiment, probability values relating to a plurality of parameter values are integrated to determine the category classification of the target person. Details of these processes will be described later.

The arithmetic processing unit 15 includes a microprocessor such as a CPU and its peripheral circuits, and has a function of realizing various processing units by reading and executing the program 14P of the storage unit 14.
The main functional units realized by the arithmetic processing unit 15 include a walking sound processing unit 15A, a parameter calculation unit 15B, and a classification determination unit 15C.

  The walking sound processing unit 15A has a function of extracting the sound generation position and the sound generation time at which each footstep sound is generated by performing the acoustic signal processing on the walking sound signal 22 including the footstep sound generated with the walking of the subject. ing. As a sound generation position and sound generation time extraction technique used in the walking sound processing unit 15A, a known technique may be used (see, for example, Non-Patent Document 1 and Non-Patent Document 2).

Further, in the acoustic signal processing, the footstep signal related to the footsteps can be extracted from the walking sound signal 22 by performing the sound pickup processing / emphasis processing on the walking sound signal 22 based on the sound generation position and the sound generation time. Good. For these sound collection processing / enhancement processing, a known technique may be used (see, for example, Non-Patent Document 3).
The acoustic signal processing and sound collection processing / emphasis processing can be applied in principle regardless of the number of pedestrians.

The parameter calculation unit 15B has a function of calculating a walking state parameter value indicating characteristics regarding the walking state of the subject based on the sound generation position and sound generation time of each footstep obtained by the walking sound processing unit 15A. Yes.
In this embodiment, statistical values relating to the step length L, the walking speed V, the opening width B from the moving center line in the left-right direction, and the time interval T of one step are used as the walking state parameter values.

  FIG. 2 is an explanatory diagram showing the walking stride and walking speed. Here, an example is shown in which the right foot is stepped (grounded) at time T0 and the left foot is then stepped at time T1. Of these, the distance between the right foot and the left foot is the step length L, and the walking speed V (= L / T) is obtained by dividing the step length L by the time interval T (= T1-T0) between the right foot and the left foot. Further, the distance from the movement center line of the right foot and the left foot along the moving direction to the center of the left and right feet is defined as an opening width B.

  Since the step length L, the walking speed V, the opening width B, and the time interval T are obtained for each of the left and right steps, in the present embodiment, they are included in an observation section having a predetermined length cut out from the walking sound signal 22. The average value of values obtained for a plurality of steps is calculated as the actual stride length L, walking speed V, opening width B, and time interval T. Further, the standard deviation indicating the variation of the value obtained for each step with respect to the average value, that is, the stride standard deviation Lsd, the walking speed standard deviation Vsd, the opening width standard deviation Bsd, and the time interval standard deviation Tsd are also used as the walking state parameters. Calculated.

  About the step L regarding walking, the walking speed V, the opening width B, and the time interval T, it is a value in which the characteristic of a walking state appears most often. In the present embodiment, a case will be described as an example in which a total of eight statistics regarding these four types of values are used in combination as walking state parameters. Note that it is not necessary to use all of these eight walking state parameters, and at least one or a plurality of walking state parameters may be used in combination. In addition, you may use values other than these four types regarding a walk state as a walk state parameter.

  The classification determination unit 15C has a function of acquiring the probability value corresponding to each walking state parameter value of the subject obtained by the parameter calculation unit 15B from the probability distribution DB 14A for each classification category, and classifies each acquired probability value into classification categories. And the function of calculating the evaluation value for each classification category, and the function of determining the classification category with the highest evaluation value among these evaluation values as the classification category for the target person. .

[Operation of First Embodiment]
Next, the operation of the walking sound analyzer 10 according to the present embodiment will be described with reference to FIG. FIG. 3 is a flowchart illustrating a walking sound analysis process of the walking sound analysis apparatus according to the first embodiment.
The arithmetic processing unit 15 starts execution of the walking sound analysis process of FIG. 3 in response to an execution start instruction from the operator detected by the operation input unit 12.

  First, the walking sound processing unit 15A acquires the walking sound signal 22 via the signal input unit 11 (step 100), and performs acoustic signal processing on the walking sound signal 22, thereby generating a sound generation position where each footstep is generated and Each sound generation time is extracted (step 101).

  Subsequently, the parameter calculation unit 15B, based on the sound generation position and sound generation time of each footstep obtained by the walking sound processing unit 15A, the width of the subject from the stride, the walking speed, and the opening width in the left-right direction from the moving center line. , And statistics regarding the time interval of one step, that is, average step length L, step standard deviation Lsd, average walking speed V, walking speed standard deviation Vsd, average opening width B, opening width standard deviation Bsd, average time interval T, and time The interval standard deviation Tsd is calculated as the walking state parameter value of the subject person (step 102).

Next, the category determination unit 15C acquires the probability value corresponding to each walking state parameter value of the subject obtained by the parameter calculation unit 15B from the probability distribution DB 14A for each category (step 103). An evaluation value for each classification category is calculated by integrating the probability values for each classification category (step 104).
Thereafter, the category determination unit 15C determines the classification category in which the highest evaluation value is obtained among these evaluation values as the classification category related to the subject (step 105), and the determined classification category is input to the walking sound signal input. 22 is displayed on the screen display unit 13 as a result of analysis, and a series of walking sound analysis processing is terminated.

  FIG. 4 is an example of probability distribution characteristics relating to stride length and height, where the horizontal axis represents the average stride length L, and the vertical axis represents the probability value P (L) with respect to the average stride length L. Here, classification sections from H1 to H9 are set for height, and probability distribution characteristics are registered in advance in the probability distribution DB 14A for each of the classification sections H1 to H9.

  Specific examples of the classification are H1 = “80 cm or less”, H2 = “80-100 cm”, H3 = “100-120 cm”, H4 = “120-140 cm”, H5 = “140-160 cm”, H6 = “ 160-180 cm ”, H7 =“ 180-200 cm ”, H8 =“ 200-220 cm ”, and H9 =“ 220 cm or more ”. For example, when the average stride L = 50 cm, H3 indicates the highest probability value. Here, the probability distribution for each classification category is shown as a bell-shaped characteristic example similar to the normal distribution, but this is only a schematic diagram, and the probability distribution characteristic does not always have such a shape. .

  FIG. 5 is an example of evaluation value calculation. When the average step length L is calculated as the walking state parameter value of the subject by the parameter calculation unit 15B, the probability value P (L) is obtained for each of the classification sections H1 to H9 from the probability distribution characteristics shown in FIG. In the example of FIG. 5, “1.2%” for H4 registrants, “15.8%” for H5, “25.6%” for H6, “0.6%” for H7, H1, H2 other than these , H3, H8, and H9 all have a probability value of “0%”. In this way, probability values corresponding to the walking state parameters are obtained for each classification category, and the evaluation values E are calculated by integrating these probability values for each classification category.

When calculating the evaluation value E, weighting is performed on the walking state parameter. Generally expressed in mathematical formulas, if the probability value for the value j of the walking state parameter i in the classification category h is Pi (j, h) and the weight for the walking state parameter i is w (i), the classification category h is related. The evaluation value E (h) is obtained by the following equation (1).
E (h) = Σ _i w (i) Pi (j, h) (1)

  In this way, the evaluation value E (h) for each classification category h is calculated. In the example of FIG. 5, E (H4) is “0.4”, E (H5) is “13.2”, E (H6) is “26.1”, E (H7) is “0.1”, Other than these, E (H1), E (H2), E (H3), E (H8), and E (H9) are all “0”. Therefore, since the highest evaluation value is E (H6), it is determined that the classification category related to the height of the subject is H6, that is, 160 to 180 cm.

In FIGS. 4 and 5, height has been described as an example of the characteristics of a person. However, physical characteristics such as weight, weight, sex, and age, as well as motor functional characteristics such as gait disorders and specific diseases, etc. Can be applied in the same manner. Among these, for motor functional features such as walking disorders and specific diseases, the presence or absence of disabilities and diseases is determined by comparing preset threshold values with evaluation values, and displayed on the screen. Also good.
As for the analysis result, instead of displaying only the determined classification category on the screen, the evaluation value of each classification category may be displayed on the screen. In particular, the evaluation result can be displayed on the screen as a probability value distribution by normalizing the evaluation value.

[Effect of the first embodiment]
As described above, in this embodiment, the walking sound processing unit 15A performs acoustic signal processing on the walking sound signal 22, thereby extracting the sound generation position and sound generation time at which each footstep sound is generated, and the parameter calculation unit 15B. Then, based on the sound generation position and sound generation time of each footstep, a walking state parameter value indicating characteristics relating to the walking state of the subject person is calculated, and the classification determination unit 15C corresponds to each walking state parameter value of the subject person. By obtaining and integrating the probability values from the probability distribution DB 14A for each classification category, the evaluation value is calculated for each classification category, and the classification category with the highest evaluation value is determined as the classification category for the subject. It is a thing.

  Thereby, based on walking sounds including footsteps accompanying walking, the characteristics of the subject, for example, physical characteristics such as height, weight, weight, sex, age, and motor functional characteristics such as walking disorders and specific diseases It is possible to determine which of the classification categories set in advance for these features.

  In the present embodiment, as the walking state parameter value of the target person, any one of the step length of the target person, the walking speed, the opening width in the horizontal direction from the moving center line, and the time interval of one step is used. Therefore, it is possible to accurately capture the characteristics of walking as a parameter. Thereby, a classification division can be determined with high accuracy.

  Further, in this embodiment, when determining a classification category related to one feature of a person, the evaluation value is calculated by integrating the probability values related to a plurality of walking state parameters, so only one walking state parameter is calculated. Compared with the case where the evaluation value is calculated in step 1, the classification category can be determined without causing a large deviation.

  In the present embodiment, the case where there is one subject has been described as an example, but in principle, the number of subjects is not limited. For this reason, it is possible to evaluate the one-dimensional or two-dimensional flow of many people and the staying situation at each position including the characteristic information of each person in facilities such as large stores and exhibitions. This makes it possible to provide extremely effective information when optimizing the arrangement / shape of the store, venue, and passage in the facility and the arrangement of the product itself.

  In the present embodiment, in step 105 of FIG. 3, when the category determination unit 15C determines the target category based on the evaluation value calculated for each category, the evaluation calculated for each category The value is compared with the threshold value set in advance for each classification category, and among these evaluation values, the classification category that has the highest evaluation value among the evaluation values that exceed the threshold value, or the evaluation The classification category having the largest difference or ratio between the value and the threshold value may be determined as the classification category related to the subject. Further, when there is no evaluation value exceeding the threshold value, it may be determined that there is no corresponding or evaluation is impossible. Thereby, higher determination accuracy can be obtained for the classification classification determination of the subject person.

  Further, when determining the classification category based on the evaluation value, the difference or ratio between the highest evaluation value and the second highest evaluation value is set as a relative threshold value (dynamic threshold value) set in advance. If the difference or ratio exceeds the relative threshold value, the classification category with the highest evaluation value may be determined as the classification category for the subject. If the difference or ratio does not exceed the relative threshold value, it may be determined that there is no corresponding or evaluation. Thereby, even when there is not much difference between the highest evaluation value and the second highest evaluation value, higher determination accuracy can be obtained for the classification classification of the target person.

The relative threshold value is not necessarily a constant, and may be a dynamic threshold value defined as a function of the highest evaluation value, the next highest evaluation value, or a difference or ratio thereof.
In addition, in the classification determination processing based on the evaluation value, the configuration for introducing the threshold value or the relative threshold value is not limited to the present embodiment, and other embodiments described below are used. Is equally applicable.

[Second Embodiment]
Next, a walking sound analyzer 10 according to a second embodiment of the present invention will be described.
In the first embodiment, the case where the classification is a human physical feature or a motor functional feature has been described as an example. In the present embodiment, a case where the classification is an individual, that is, a case where the target person recognizes who will be described as an example.

  In the present embodiment, the probability distribution DB 14A of the storage unit 14 stores the frequency distribution of each walking state parameter value obtained from the footsteps of the registrant in the classification category for each classification category corresponding to the registrant registered in advance as a classification target. Is stored.

  FIG. 6 is an example of probability distribution characteristics regarding the stride and the registrant. The horizontal axis indicates the average stride L, and the vertical axis indicates the probability value P (L) with respect to the average stride L. Here, classification categories M1 to M5 are set as registrants, and probability distribution characteristics are registered in advance in the probability distribution DB 14A for each of the classification categories M1 to M5.

Specific examples of the classification categories are M1 = “Mr. A”, M2 = “Mr. B”, M3 = “Mr. C”, M4 = “Mr. D”, and M5 = “Mr. E”. For example, in the case of the average stride L = 80 cm, “Mr. E” of M5 indicates the highest probability value. Here, the probability distribution for each classification category is shown as a bell-shaped characteristic example similar to the normal distribution, but this is only a schematic diagram, and the probability distribution characteristic does not always have such a shape. .
In addition, about the other structure in the walking sound analyzer 10 concerning this Embodiment, it is the same as that of 1st Embodiment, and detailed description here is abbreviate | omitted.

[Operation of Second Embodiment]
Next, the operation of the walking sound analyzer 10 according to the second embodiment of the present invention will be described with reference to FIG. FIG. 7 is a flowchart showing the walking sound analysis process of the walking sound analyzer according to the second embodiment, and the same or equivalent parts as those in FIG.
The arithmetic processing unit 15 starts execution of the walking sound analysis process of FIG. 7 in response to an execution start instruction from the operator detected by the operation input unit 12.

  First, the walking sound processing unit 15A acquires the walking sound signal 22 via the signal input unit 11 (step 100), and performs acoustic signal processing on the walking sound signal 22, thereby generating a sound generation position where each footstep is generated and Each sound generation time is extracted (step 101).

  Subsequently, the parameter calculation unit 15B, based on the sound generation position and sound generation time of each footstep obtained by the walking sound processing unit 15A, the width of the subject from the stride, the walking speed, and the opening width in the left-right direction from the moving center line. , And statistics regarding the time interval of one step, that is, average step length L, step standard deviation Lsd, average walking speed V, walking speed standard deviation Vsd, average opening width B, opening width standard deviation Bsd, average time interval T, and time The interval standard deviation Tsd is calculated as the walking state parameter value of the subject person (step 102).

Next, the category determination unit 15C acquires the probability value corresponding to each walking state parameter value of the subject obtained by the parameter calculation unit 15B from the probability distribution DB 14A for each category, here for each registrant (step S1). 200), by integrating the acquired probability values for each registrant, an evaluation value for each registrant is calculated (step 201).
Thereafter, the category determination unit 15C determines that the registrant who obtained the highest evaluation value among these evaluation values is the target person (step 105), and the determined registrant is input to the input walking sound signal 22. As a result of the analysis, the screen display unit 13 displays the screen, and the series of walking sound analysis processing ends.

  FIG. 8 is an example of calculation of evaluation values. When the average step length L is calculated as the walking state parameter value of the subject by the parameter calculation unit 15B, the probability value P (L) is obtained for each of the classification sections M1 to M5 from the probability distribution characteristics shown in FIG. In the example of FIG. 8, the probability values of “4.9%” for M2, “13.8%” for M3, “26.2%” for M4, and “0%” for all other M1 and M5, respectively. Has been obtained. In this way, probability values corresponding to the walking state parameters are obtained for each classification category, and the evaluation values E are calculated by integrating these probability values for each classification category.

At this time, for the evaluation value E, the weight for the walking state parameter i may be calculated using the above-described equation (1) in consideration of w (i).
In this way, the evaluation value E (h) for each classification category h is calculated. In the example of FIG. 8, E (M1) is “1.6”, E (M2) is “5.9”, E (M3) is “13.8”, E (M4) is “25.9”, E (M5) is “3.4”. Therefore, since the highest evaluation value is E (M4), it is determined that the subject is Mr. D of M4.

  As for the analysis result, not only the determined classification category is displayed on the screen, but the evaluation value of each classification category may be displayed on the screen. In particular, the evaluation result can be displayed on the screen as a probability value distribution by normalizing the evaluation value.

[Effect of the second embodiment]
As described above, in this embodiment, in the probability distribution DB 14A of the storage unit 14, for each classification category corresponding to the registrant registered in advance as the classification target, each walking state parameter obtained from the footsteps of the registrant in the classification category. Since the probability distribution characteristic indicating the frequency distribution of values is stored, it is possible to determine which of the registrants the target person, and as a result, personal identification and authentication for the target person can be performed. It becomes possible.

[Third Embodiment]
Next, a walking sound analyzer 10 according to a third embodiment of the present invention will be described.
In the first embodiment, from the sound generation position and sound generation time of each footstep obtained by acoustic signal processing of the walking sound signal 22, the average stride L, the stride standard deviation Lsd, the average walk as the walking state parameter values. The case of calculating the speed V, the walking speed standard deviation Vsd, the average opening width B, the opening width standard deviation Bsd, the average time interval T, and the time interval standard deviation Tsd has been described as an example.
In the present embodiment, by analyzing the frequency spectrum of the walking sound signal 22, in addition to the walking state parameter value, a footstep parameter value is calculated as a feature value indicating an acoustic feature related to the footstep signal. A case where the classification classification of the target person is determined based on the parameter value will be described.

In the present embodiment, the walking sound processing unit 15A calculates a spectrogram as a feature value indicating an acoustic feature related to the footstep signal by performing frequency spectrum analysis on the footstep signal indicating the footstep extracted from the walking sound signal 22. It has a function.
The parameter calculation unit 15B obtains, from each feature value obtained by the walking sound processing unit 15A, that is, a spectrogram, a statistical value including an average spectral intensity value S and a spectral intensity value standard deviation Ssd related to a spectral intensity value at a specific frequency. Each footstep parameter value is calculated.

In the probability distribution DB 14A, a probability distribution characteristic indicating a frequency distribution of each footstep parameter value obtained from a footstep of a person belonging to the classification category is registered for each classification category.
When calculating the evaluation value for each classification category, the category determination unit 15C associates the probability value corresponding to each footstep parameter value of the target person from the probability distribution DB 14A for each classification category and the acquired probability value. And a function of integrating each into the evaluation value of the classification category.
In addition, about the other structure in the walking sound analyzer 10 concerning this Embodiment, it is the same as that of 1st Embodiment, and detailed description here is abbreviate | omitted.

  FIG. 9 is an explanatory diagram showing the cutout of the footstep signal. A person's footsteps usually have a duration of about 100 to 300 ms, and the walking sound signal 22 includes not only footsteps but also preceding and subsequent sections. For this reason, in order to acquire an acoustic feature related to footsteps, the entire range of the footstep duration Ta or only a part of the range required for the duration Ta is obtained from the walking sound signal 22 based on its amplitude and the like. May be extracted as a footstep signal.

  In the present embodiment, a statistic relating to a spectrum intensity value at a specific frequency is used as an acoustic feature related to a footstep signal. Therefore, first, a footstep signal is cut out from a plurality of frame signals having a time length Tf of, for example, about 50 ms while sequentially shifting the start time by a time length Ts of, for example, about 10 ms, and short-time Fourier transform is performed for each frame signal. Perform frequency spectrum analysis. As a result, a frequency spectrum distribution for each frame signal and its temporal change, that is, a spectrogram is obtained as a feature value of the footstep signal.

  FIG. 10A is an amplitude waveform diagram of a footstep signal, where the horizontal axis indicates the elapsed time in walking (unit: [sec]), and the vertical axis indicates the amplitude value (arbitrary unit [a.u.]) of the footstep signal. Here, footstep signals for five steps are shown, and each footstep is continuous at an interval of about 0.54 [sec].

  FIG. 10B is a spectrogram of a footstep signal, where the horizontal axis indicates the elapsed time in walking (unit: [sec]), the vertical axis indicates frequency (unit: [kHz]), and the color shading is the spectrum of the frequency component. Indicates intensity (unit: [dB]). Here, the frequency spectrum distribution obtained from each footstep signal for five steps is shown in correspondence with the footstep signal of FIG. 10A in time, and the footstep sound has a frequency component sound up to about 20 kHz. It can be seen that it is included at a certain spectral intensity.

  From the spectrogram of each frame signal thus obtained, for example, after removing the noise component spectrum evaluated from the data of the time when no walking sound exists, for example, the spectral intensity at the preset specific frequency f is removed. A value is acquired for each frame signal. Then, an average value of these spectrum intensity values, that is, an average spectrum intensity value S (f) may be calculated as a footstep parameter value. Further, the standard deviation indicating the variation of the spectrum intensity value for each frame signal with respect to the average value, that is, the spectrum intensity value standard deviation Ssd (f) may be calculated as the footstep parameter value.

  Further, the spectral intensity of the specific frequency f in one step of footsteps is not only evaluated by averaging the value of each frame signal, but more specifically, the time change St (f, t) in the time range of one step of footsteps. It can also be extracted and used as a footstep parameter. Also, feature information indicating the acoustic features of the footstep signal extracted from the spectrogram of the footstep signal, that is, an average spectrum intensity value S (f), a spectrum intensity value standard deviation Ssd (f), and a time change St (f, t) are obtained. Any combination may be used as one footstep parameter. In addition, this time change St (f, t) is used as a footstep parameter value in the other embodiments, like the average spectral intensity value S (f) and the spectral intensity value standard deviation Ssd (f). These may be arbitrarily combined and used as one footstep parameter.

Regarding the average spectral intensity value S (f), the spectral intensity value standard deviation Ssd (f), and the temporal change St (f, t) of the spectral intensity, the change in the sound generation position where the footstep sound is generated as necessary. After normalizing the data by correcting the variation in the intensity of the sound due to the sound, it is also possible to obtain an average value of a plurality of footstep signals obtained for each footstep signal and obtain a standard deviation with respect to the average value. . Alternatively, after normalizing the data of each footstep signal, the average spectral intensity value S (f) and the spectral intensity value standard deviation Ssd (f) are calculated for all the frame signals included in each footstep signal. Also good.
In addition, a plurality of frequencies are set as the specific frequency f, and the average spectral intensity value S (f), the spectral intensity value standard deviation Ssd (f), and the temporal change St (f, t) of the spectral intensity are set for each specific frequency. May be used as independent footstep parameter values, or a combination of these footstep parameters arbitrarily may be used as one footstep parameter value. This can also be applied to other embodiments.

[Operation of Third Embodiment]
Next, with reference to FIG. 11, operation | movement of the walking sound analyzer 10 concerning this Embodiment is demonstrated. FIG. 11 is a flowchart showing the walking sound analysis process of the walking sound analyzer according to the third embodiment, and the same or equivalent parts as those in FIG. 3 are given the same reference numerals.
The arithmetic processing unit 15 starts execution of the walking sound analysis process of FIG. 7 in response to an execution start instruction from the operator detected by the operation input unit 12.

First, the walking sound processing unit 15A acquires the walking sound signal 22 via the signal input unit 11 (step 100), and performs acoustic signal processing on the walking sound signal 22, thereby generating a sound generation position where each footstep is generated and Each sound generation time is extracted (step 101).
The walking sound processing unit 15A extracts a footstep signal indicating each footstep from the walking sound signal 22 as described above, and performs frequency spectrum analysis as a feature value indicating an acoustic feature related to the footstep signal. A spectrogram is calculated (step 300).

  Subsequently, the parameter calculation unit 15B, based on the sound generation position and sound generation time of each footstep obtained by the walking sound processing unit 15A, the width of the subject from the stride, the walking speed, and the opening width in the left-right direction from the moving center line. , And statistics regarding the time interval of one step, that is, average step length L, step standard deviation Lsd, average walking speed V, walking speed standard deviation Vsd, average opening width B, opening width standard deviation Bsd, average time interval T, and time The interval standard deviation Tsd is calculated as the walking state parameter value of the subject person (step 102).

  Further, as described above, the parameter calculation unit 15B calculates the average spectral intensity value S and the spectral intensity value standard deviation Ssd related to the spectral intensity value at the specific frequency based on the spectrogram obtained by the walking sound processing unit 15A as footsteps. It is calculated as a parameter value (step 102).

Next, the category determination unit 15C generates a probability distribution corresponding to each walking state parameter value of the subject obtained by the parameter calculation unit 15B and a probability value corresponding to each footstep parameter value for each category category. Obtained from the DB 14A (step 103), and by integrating the probability values corresponding to each walking state parameter value for each classification category, the evaluation value for each classification category is calculated, and the probability value corresponding to each footstep parameter value is calculated. And integrated into the evaluation value of the corresponding classification category (step 104).
Thereafter, when these evaluation values exceed a threshold value set in advance for each registrant, the classification determination unit 15C determines that the registrant is a target person (step 105), and determines the registered The person is displayed on the screen display unit 13 as an analysis result of the input walking sound signal 22, and the series of walking sound analysis processing is terminated. In this case, when there are multiple registrants whose evaluation value exceeds the threshold set in advance for each registrant, or who have the highest difference or ratio between the threshold and the evaluation value. The registrant who obtained the highest evaluation value among them may be determined as the target person. If there is no registrant exceeding the threshold, it may be determined that there is no registration or evaluation is impossible.

[Effect of the third embodiment]
As described above, according to the present embodiment, the walking sound processing unit 15A calculates the feature value indicating the acoustic feature related to the footstep signal for the footstep signal indicating the footstep extracted from the walking sound signal 22, and the parameter calculation unit. When the statistical value regarding each feature value obtained by the walking sound processing unit 15A is calculated as a footstep parameter value of the subject by 15B and the evaluation value is calculated for each classification category by the category determination unit 15C, Are obtained from the probability distribution DB 14A for each category and integrated with the evaluation values of the corresponding category.

  Thus, when determining which of the classification categories preset for the feature the feature of the subject belongs to, not only the walking state parameter value of the subject but also the acoustics included in the footsteps of the subject The footstep parameter value indicating the feature can also be considered, and the classification category of the subject can be determined in more detail.

[Fourth Embodiment]
Next, with reference to FIG. 12, a walking sound analyzer 10 according to a fourth embodiment of the present invention will be described. FIG. 12 is a block diagram showing a walking sound analyzer according to the fourth embodiment, and the same or equivalent parts as those in FIG.

  In the third embodiment, the case where the classification category of the subject person is determined in consideration of the footstep parameter value indicating the acoustic feature included in the footstep of the subject person has been described. Here, the footstep parameter value reflects not only the physical characteristics or motor functional characteristics of the subject but also the type of footwear worn by the subject. In this embodiment, paying attention to this, the footwear type of the footwear that the subject wears is determined from the footstep parameter value, for example, the footwear type such as leather shoes, high heels, sandals, sneakers, sandals, and the like. In consideration of this, the classification category of the target person is determined.

In the present embodiment, the walking sound processing unit 15A calculates a spectrogram as a feature value indicating an acoustic feature related to the footstep signal by performing frequency spectrum analysis on the footstep signal indicating the footstep extracted from the walking sound signal 22. It has a function.
The parameter calculation unit 15B obtains, from each feature value obtained by the walking sound processing unit 15A, that is, a spectrogram, a statistical value including an average spectral intensity value S and a spectral intensity value standard deviation Ssd related to a spectral intensity value at a specific frequency. Each footstep parameter value is calculated.

In addition, as shown in FIG. 12, the storage unit 14 includes, as new processing information, footwear types related to footwear worn by a person and footstep parameter values obtained from footsteps of a person wearing the footwear of the footwear type. A footwear type database (hereinafter referred to as footwear type DB) 14B indicating a correspondence relationship is stored.
The probability distribution DB 14A of the storage unit 14 registers a probability distribution characteristic indicating a frequency distribution for each footwear type regarding footwear worn by a person belonging to the classification category for each classification category. With respect to the probability distribution characteristics related to the footwear type, for example, individual probability values are registered for each combination of the classification category and the footwear type, not the continuous characteristics as shown in FIG. 4 described above.

Further, as shown in FIG. 12, the arithmetic processing unit 15 refers to the footwear type DB 14B of the storage unit 14 as a new processing unit, and corresponds to the footstep parameter value of the subject obtained by the parameter calculation unit 15B. A footwear type determination unit 15D that determines the footwear type to be used as the footwear type of the subject person is provided.
When the category determination unit 15C calculates an evaluation value for each classification category, the category determination unit 15C acquires a probability value corresponding to the footwear type of the target person obtained by the footwear type determination unit 15D from the probability distribution DB 14A for each classification category; And a function of integrating the acquired probability values into the evaluation values of the corresponding classification categories.

  The details of the walking sound processing unit 15A, the parameter calculation unit 15B, and the probability distribution DB 14A described above are the same as those in the third embodiment. The other configuration of the walking sound analysis apparatus 10 according to the present embodiment is the same as that of the first embodiment, and detailed description thereof is omitted here.

[Operation of Fourth Embodiment]
Next, the operation of the walking sound analyzer 10 according to the fourth embodiment of the present invention will be described with reference to FIG. FIG. 13: is a flowchart which shows the walk sound analysis process of the walk sound analyzer concerning 4th Embodiment, The same code | symbol is attached | subjected to the same or equivalent part as above-mentioned FIG.
The arithmetic processing unit 15 starts executing the walking sound analysis process of FIG. 13 in response to an execution start instruction from the operator detected by the operation input unit 12.

First, the walking sound processing unit 15A acquires the walking sound signal 22 via the signal input unit 11 (step 100), and performs acoustic signal processing on the walking sound signal 22, thereby generating a sound generation position where each footstep is generated and Each sound generation time is extracted (step 101).
The walking sound processing unit 15A extracts a footstep signal indicating each footstep from the walking sound signal 22 as described above, and performs frequency spectrum analysis as a feature value indicating an acoustic feature related to the footstep signal. A spectrogram is calculated (step 300).

  Subsequently, the parameter calculation unit 15B, based on the sound generation position and sound generation time of each footstep obtained by the walking sound processing unit 15A, the width of the subject from the stride, the walking speed, and the opening width in the left-right direction from the moving center line. , And statistics regarding the time interval of one step, that is, average step length L, step standard deviation Lsd, average walking speed V, walking speed standard deviation Vsd, average opening width B, opening width standard deviation Bsd, average time interval T, and time The interval standard deviation Tsd is calculated as the walking state parameter value of the subject person (step 102).

Further, as described above, the parameter calculation unit 15B calculates the average spectral intensity value S and the spectral intensity value standard deviation Ssd related to the spectral intensity value at the specific frequency based on the spectrogram obtained by the walking sound processing unit 15A as footsteps. It is calculated as a parameter value (step 102).
Thereafter, the footwear type determination unit 15D refers to the footwear type DB 14B of the storage unit 14, and determines the footwear type corresponding to the footstep parameter value of the subject obtained by the parameter calculation unit 15B as the footwear type of the subject. (Step 400).

  Next, the classification determination unit 15C has a probability value corresponding to each walking state parameter value of the subject obtained by the parameter calculation unit 15B and a probability corresponding to the footwear type of the subject determined by the footwear type determination unit 15D. Value is obtained from the probability distribution DB 14A for each classification category (step 103), and the probability value corresponding to each walking state parameter value is integrated for each classification category, thereby calculating an evaluation value for each classification category, The probability value corresponding to the footwear type is integrated into the evaluation value of the corresponding classification category (step 104).

  Thereafter, the category determination unit 15C determines that the registrant who obtained the highest evaluation value among these evaluation values is the target person (step 105), and the determined registrant is input to the input walking sound signal 22. As a result of the analysis, the screen display unit 13 displays the screen, and the series of walking sound analysis processing ends.

[Effect of the fourth embodiment]
As described above, according to the present embodiment, the walking sound processing unit 15A calculates the feature value indicating the acoustic feature related to the footstep signal for the footstep signal indicating the footstep extracted from the walking sound signal 22, and the parameter calculation unit. 15B, the statistical value regarding each feature value obtained by the walking sound processing unit 15A is calculated as the footstep parameter value of the subject person, and the footwear type determining unit 15D refers to the footwear type DB 14B, and the parameter calculating unit 15B When the footwear type corresponding to the footstep parameter value of the subject obtained in the above is determined as the footwear type of the subject and the evaluation value is calculated for each classification by the classification determination unit 15C, the footwear type determination unit 15D The probability value corresponding to the target person's footwear type is acquired from the probability distribution DB 14A for each classification category, and integrated into the evaluation value of the corresponding classification category. One in which the.

  Thus, when determining which of the classification categories preset for the feature the feature of the subject belongs to, not only the walking state parameter value of the subject but also the acoustics included in the footsteps of the subject The footwear type of the subject determined from the footstep parameter value indicating the feature can also be taken into consideration, and the classification of the subject can be determined in more detail.

  In the present embodiment, when the probability value corresponding to the footwear type is integrated into the evaluation value of the corresponding classification category, each walking state parameter value is actually as described in the above-described equation (1). The product-sum calculation of the probability value corresponding to and the probability value corresponding to the footwear type and the weight set in advance for each walking state parameter value and footwear type is performed. In this case, one common weight may be used as the weight corresponding to the footwear type, but a different weight may be used depending on the determination result of the footwear type.

  For example, when the determination result of the footwear type is “high heels”, the probability of being a woman is very high for the classification category of the subject's gender. On the other hand, when the determination result of the footwear type is “athletic shoes”, there is not much difference in the probability of male / female. For this reason, in such a case, an evaluation value may be calculated using a weight that is registered in the footwear type DB 14B in advance and that corresponds to the footwear type determination result. Thereby, it is possible to determine the classification category of the subject in more detail.

[Fifth Embodiment]
Next, with reference to FIG. 14, a walking sound analyzing apparatus 10 according to a fifth embodiment of the present invention will be described. FIG. 14 is a block diagram showing a walking sound analyzer according to the fifth embodiment, where the same or equivalent parts as those in FIG. 1 are given the same reference numerals.

  In the fourth embodiment, the case has been described in which the classification category of the subject is determined in consideration of the footwear type of the subject determined from the footstep parameter value indicating the acoustic feature included in the footstep of the subject. Here, the walking sound generated along with the walking of the subject is generated not only from the footsteps of the subject but also from accessories such as clothes and heels worn by the subject. In this embodiment, paying attention to this, in consideration of the incidental object of the subject determined from the observation point parameter value indicating the acoustic characteristics regarding the sound generated at the observation points provided around the subject, the subject The classification category is determined.

In the present embodiment, the walking sound processing unit 15A performs a sound collection process on the walking sound signal 22, thereby obtaining an observation point signal indicating the sound generated at the observation point for each observation point provided around the subject. And a function of calculating a spectrogram as an observation point feature value indicating an acoustic feature related to the observation point signal by performing frequency spectrum analysis for each observation point signal.
The parameter calculation unit 15B calculates, from each observation point feature value obtained by the walking sound processing unit 15A, that is, a spectrogram, a statistical value including an average spectral intensity value S and a spectral intensity value standard deviation Ssd related to a spectral intensity value at a specific frequency. Each observation point has a function of calculating the observation point parameter value of the subject person.

  FIG. 15 is an explanatory diagram showing the observation area of the subject. Here, as the observation area surrounding the subject, an area having a width of 0.6 m in each of the front, rear, left, and right directions and a height of 2.0 m from the floor in the height direction is set. Actually, this observation area is divided into a grid at intervals of 0.2 m to set 176 observation points, and an observation point signal indicating the sound generated at these observation points is generated. As a result, it is possible to obtain an observation point signal indicating a walking sound generated from an accessory such as the subject's clothes or personal belongings, and an incidental signal based on the observation point parameter value obtained from the observation point feature value indicating the acoustic feature. You can judge things.

For the observation point signal at each observation point, the above-described known sound collection processing / emphasis processing may be used (see, for example, Non-Patent Document 3). As a result, the friction sound of clothes, the vibrations of belongings, and the sound generated when hitting the body, such as front and rear, right and left of the subject's body, each part of the foot, near the waist, near the arm, near the head, etc. Can be obtained individually.
As for the observation area, the center position of the subject's body at each time can be estimated by calculating the walking speed from the sound generation position and sound generation time of the continuous footsteps. You only have to set it. Further, the size of the space between these observation areas and observation points is not limited to this.

  Further, as shown in FIG. 14, the storage unit 14 obtains, as new processing information, the incidental type related to the incidental that accompanies the person and the observation point signal of the person who accompanies the incidental of the incidental type. In addition, an accessory type database (hereinafter referred to as an accessory type DB) 14C indicating a correspondence relationship with each observation point parameter value is stored. As accessory types, there are clothes such as trousers, skirts and coats, fabric materials thereof, belongings such as bags and bags, and shapes and sizes thereof.

  Further, the probability distribution DB 14A of the storage unit 14 registers probability distribution characteristics indicating the frequency distribution for each accessory type related to the accessory attached by the person belonging to the classification category for each classification category. For example, the probability distribution characteristics related to the accessory types are not continuous characteristics as shown in FIG. 4 described above, but individual probability values are registered for each combination of the classification category and the accessory type.

Further, as shown in FIG. 12, the arithmetic processing unit 15 refers to the incidental type DB 14C as a new processing unit, and corresponds to the observation point parameter value of the subject obtained by the parameter calculation unit 15B. An accessory type determination unit 15E that determines the object type as an accessory type of the subject is provided.
When the category determination unit 15C calculates an evaluation value for each classification category, the category determination unit 15C acquires, from the probability distribution DB 14A, the probability value corresponding to the target category of the subject obtained by the accessory type determination unit 15E for each classification category. A function and a function of integrating the acquired probability value into the evaluation value of the corresponding classification category.

[Operation of Fifth Embodiment]
Next, with reference to FIG. 16, the operation of the walking sound analyzer 10 according to the fifth embodiment of the present invention will be described. FIG. 16 is a flowchart showing the walking sound analysis process of the walking sound analyzer according to the fifth embodiment, and the same or equivalent parts as those in FIG. 7 are given the same reference numerals.
The arithmetic processing unit 15 starts execution of the walking sound analysis process of FIG. 16 in response to an execution start instruction from the operator detected by the operation input unit 12.

First, the walking sound processing unit 15A acquires the walking sound signal 22 via the signal input unit 11 (step 100), and performs acoustic signal processing on the walking sound signal 22, thereby generating a sound generation position where each footstep is generated and Each sound generation time is extracted (step 101).
In addition, the walking sound processing unit 15A generates an observation point signal at each observation point from the walking sound signal 22 as described above, and performs frequency spectrum analysis, so that the observation point signal is generated for each observation point signal. A spectrogram is calculated as a feature value indicating an acoustic feature related to (step 500).

  Subsequently, the parameter calculation unit 15B, based on the sound generation position and sound generation time of each footstep obtained by the walking sound processing unit 15A, the width of the subject from the stride, the walking speed, and the opening width in the left-right direction from the moving center line. , And statistics regarding the time interval of one step, that is, average step length L, step standard deviation Lsd, average walking speed V, walking speed standard deviation Vsd, average opening width B, opening width standard deviation Bsd, average time interval T, and time The interval standard deviation Tsd is calculated as the walking state parameter value of the subject person (step 102).

Further, as described above, the parameter calculation unit 15B observes the average spectral intensity value S and the spectral intensity value standard deviation Ssd related to the spectral intensity value at the specific frequency based on the spectrogram obtained by the walking sound processing unit 15A. A point parameter value is calculated (step 102).
Thereafter, the accessory type determination unit 15E refers to the accessory type DB 14C of the storage unit 14 and determines the accessory type corresponding to the observation point parameter value of the subject obtained by the parameter calculation unit 15B. It is determined as an object type (step 501).

  Next, the classification determination unit 15C corresponds to the probability value corresponding to each walking state parameter value of the subject obtained by the parameter calculation unit 15B and the accessory type of the subject determined by the accessory type determination unit 15E. Are obtained from the probability distribution DB 14A for each classification category (step 103), and the evaluation values for each classification category are calculated by integrating the probability values corresponding to each walking state parameter value for each classification category. At the same time, the probability value corresponding to the accessory type is integrated into the evaluation value of the corresponding classification category (step 104).

  Thereafter, the category determination unit 15C determines that the registrant who obtained the highest evaluation value among these evaluation values is the target person (step 105), and the determined registrant is input to the input walking sound signal 22. As a result of the analysis, the screen display unit 13 displays the screen, and the series of walking sound analysis processing ends.

[Effect of Fifth Embodiment]
As described above, in the present embodiment, the walking sound processing unit 15A performs sound collection processing on the walking sound signal 22 to indicate the sound generated at the observation point for each observation point provided around the subject. An observation point signal is generated, an observation point feature value indicating an acoustic feature related to the observation point signal is calculated for each observation point signal, and each observation obtained by the walking sound processing unit 15A is performed by the parameter calculation unit 15B. The statistical value related to the point feature value is calculated as the observation point parameter value of the subject for each observation point, and the accessory type determination unit 15E refers to the accessory type DB 14C to obtain the target obtained by the parameter calculation unit 15B. The accessory type corresponding to the observation point parameter value of the person is determined as the accessory type of the target person, and when the category determination unit 15C calculates the evaluation value for each classification category, it corresponds to the accessory type of the target person. Probability values, is obtained from a probability distribution DB14A each classification category, in which so as to integrate each to the evaluation value of the corresponding classification categories.

  Thereby, when determining which of the classification categories set in advance for the feature the subject has, the walking sound generated around the subject as well as the walking state parameter value of the subject It is also possible to consider the incidental type of the subject determined from the observation point parameter value indicating the acoustic feature included in the, and to determine the classification of the subject in more detail.

  In the present embodiment, a case has been described in which a statistical value including the average spectral intensity value S and the spectral intensity value standard deviation Ssd regarding the spectral intensity value at a specific frequency is used as the observation point parameter value. May also be included in the observation point parameter value. Thereby, when determining the accessory type in the accessory type determination unit 15E, it is possible to consider the position where each accessory exists in the observation area of the subject, and determine the type of the accessory in more detail. can do.

  In the present embodiment, when the probability value corresponding to the accessory type is integrated into the evaluation value of the corresponding classification category, each walking state parameter as described in Expression (1) is actually used. The sum of products of the probability value corresponding to the value and the probability value corresponding to the accessory type and the weight set in advance for each of the walking state parameter value and the accessory type is calculated. In this case, one common weight may be used as the weight corresponding to the accessory type, but a different weight may be used depending on the determination result of the accessory type.

  For example, when the determination result of the accessory type is “handback”, the probability of being a woman is extremely high for the classification category of the subject's gender. On the other hand, when the determination result of the accessory type is “rucksack”, there is not much difference in the probability of male / female. For this reason, in such a case, the evaluation value may be calculated using the weight according to the determination result of the incidental type registered in advance in the incidental type DB 14C. Thereby, it is possible to determine the classification category of the subject in more detail.

[Extended embodiment]
The present invention has been described above with reference to the embodiments, but the present invention is not limited to the above embodiments. Various changes that can be understood by those skilled in the art can be made to the configuration and details of the present invention within the scope of the present invention. Each embodiment may be implemented in any combination.

  DESCRIPTION OF SYMBOLS 10 ... Walking sound analyzer, 11 ... Signal input part, 12 ... Operation input part, 13 ... Screen display part, 14 ... Memory | storage part, 14A ... Probability distribution DB, 14B ... Footwear type DB, 14C ... Incident type DB, 15 ... arithmetic processing unit, 15A ... walking sound processing unit, 15B ... parameter calculation unit, 15C ... segment determination unit, 15D ... footwear type determination unit, 15E ... accessory type determination unit, 20 ... sound collector, 21 ... microphone, 22 ... walking sound signal.

Claims (13)

A walking sound processing unit that extracts a sound generation position and a sound generation time at which each footstep sound is generated by performing acoustic signal processing on a walking sound signal including a footstep sound that occurs as the subject walks, and
A parameter calculation unit that calculates a walking state parameter value indicating characteristics related to the walking state of the subject based on the sound generation position and sound generation time of each footstep obtained by the walking sound processing unit;
A probability distribution database that stores a probability distribution characteristic indicating a frequency distribution of the walking state parameter value obtained from a footstep of a person belonging to the classification category for each preset classification category of the person according to the characteristics of the person,
By obtaining and integrating the probability values corresponding to the walking state parameter values of the subject obtained by the parameter calculation unit from the probability distribution database for each classification category, an evaluation value is calculated for each classification category. And a classification determination unit that determines a classification category having the highest evaluation value as a classification category related to the target person. The walking sound analyzer according to claim 1,
The walking sound processing unit calculates a feature value indicating an acoustic feature related to the footstep signal for the footstep signal indicating the footstep extracted from the walking sound signal,
The parameter calculation unit calculates a statistical value regarding each feature value obtained by the walking sound processing unit as a footstep parameter value of the subject,
The probability distribution database stores a probability distribution characteristic indicating a frequency distribution of each footstep parameter value obtained from a footstep of a person belonging to the category for each category.
The category determination unit obtains a probability value corresponding to each footstep parameter value of the target person from the probability distribution database for each category category when calculating the evaluation value for each category category. A walking sound analyzer characterized by being integrated into the evaluation values of the classification categories. In the walking sound analyzer according to claim 1 or 2,
The walking sound processing unit calculates a feature value indicating an acoustic feature related to the footstep signal for the footstep signal indicating the footstep extracted from the walking sound signal,
The parameter calculation unit calculates a statistical value regarding each feature value obtained by the walking sound processing unit as a footstep parameter value of the subject,
A footwear type database showing a correspondence relationship between footwear types related to footwear worn by a person and each footstep parameter value obtained from footsteps of a person wearing the footwear of the footwear type,
With reference to the footwear type database, further comprising a footwear type determination unit that determines the footwear type corresponding to the footstep parameter value of the target person obtained by the parameter calculation unit as the footwear type of the target person,
The probability distribution database further stores, for each classification category, a probability distribution characteristic indicating a frequency distribution for each footwear type regarding footwear worn by persons belonging to the classification category,
When the category determination unit calculates the evaluation value for each of the classification categories, a probability value corresponding to the footwear type of the target person obtained by the footwear type determination unit is calculated for each classification category in the probability distribution database. The walking sound analyzer is characterized by being acquired from the above and integrated with the evaluation values of the corresponding classification categories. In the walking sound analyzer according to any one of claims 1 to 3,
The walking sound processing unit generates an observation point signal indicating sound generated at the observation point for each observation point provided around the subject by performing sound collection processing on the walking sound signal, For each observation point signal, an observation point feature value indicating an acoustic feature related to the observation point signal is calculated,
The parameter calculation unit calculates a statistical value regarding each observation point feature value obtained by the walking sound processing unit as an observation point parameter value of the subject for each observation point,
An accessory type database that indicates a correspondence relationship between an accessory type related to an accessory attached to a person and each observation point parameter value obtained from the observation point signal of a person attached to the accessory of the accessory type ,
An accessory type determination unit that refers to the accessory type database and determines an accessory type corresponding to the observation point parameter value of the subject obtained by the parameter calculation unit as the subject type of the subject; Prepared,
The probability distribution database further stores, for each classification category, a probability distribution characteristic indicating a frequency distribution for each accessory type related to an accessory attached to a person belonging to the classification category,
When calculating the evaluation value for each classification category, the category determination unit obtains a probability value corresponding to the accessory type of the target person from the probability distribution database for each classification category, and corresponds to the classification category. A walking sound analyzer characterized in that it is integrated with each evaluation value of each category. In the walking sound analyzer according to any one of claims 1 to 4,
The parameter calculation unit relates to any one or more of at least the step length of the subject, the walking speed, the opening width from the movement center line in the left-right direction, and the time interval of one step as the walking state parameter value. A walking sound analyzer characterized by calculating feature information indicating a statistic or an acoustic feature of the footstep signal extracted from a spectrogram of the footstep signal. In the walking sound analyzer according to any one of claims 1 to 5,
The category determination unit is a category category having the highest evaluation value among the evaluation values calculated for each category category and exceeding the preset threshold value for each category category. Alternatively, the walking sound analyzing apparatus characterized in that the classification category having the largest difference or ratio between the evaluation value and the threshold value is determined as the classification category related to the subject. A walking sound analysis method used in a walking sound analysis apparatus that analyzes a walking sound signal including footsteps generated along with walking of a target person and determines characteristics of the target person,
A walking sound processing step in which a walking sound processing unit extracts a sound generation position and a sound generation time at which each footstep sound is generated by performing acoustic signal processing on a walking sound signal including a footstep sound that occurs as the subject walks. ,
A parameter calculating step for calculating a walking state parameter value indicating characteristics relating to the walking state of the subject based on a sound generation position and a sound generation time of each footstep obtained by the walking sound processing unit;
A probability distribution characteristic indicating a frequency distribution of the walking state parameter value obtained from a footstep of a person belonging to the classification category for each of the classification categories of the person set in advance according to the characteristics of the person as a probability distribution database Memorizing step;
The classification determination unit acquires the probability value corresponding to the walking state parameter value of the subject obtained by the parameter calculation unit from the probability distribution database and integrates the classification value for each classification category. A walking sound analysis method, comprising: a classification determination step of calculating an evaluation value and determining a classification category with the highest evaluation value as a classification category related to the subject. The walking sound analysis method according to claim 7,
The walking sound processing step calculates a feature value indicating an acoustic feature related to the footstep signal for the footstep signal indicating the footstep extracted from the walking sound signal,
The parameter calculating step calculates a statistical value regarding each feature value obtained by the walking sound processing unit as a footstep parameter value of the subject,
The probability distribution database stores a probability distribution characteristic indicating a frequency distribution of each footstep parameter value obtained from a footstep of a person belonging to the category for each category.
The category determination step obtains a probability value corresponding to each footstep parameter value of the target person from the probability distribution database for each category category when calculating the evaluation value for each category category. A walking sound analysis method characterized by integrating each of the classification values into the evaluation values. In the walking sound analysis method according to claim 7 or 8,
The walking sound processing step calculates a feature value indicating an acoustic feature related to the footstep signal for the footstep signal indicating the footstep extracted from the walking sound signal,
The parameter calculating step calculates a statistical value related to each feature value obtained by the walking sound processing unit as a footstep parameter value of the subject,
The storage unit further includes a step of storing a footwear type database indicating a correspondence relationship between footwear types related to footwear worn by a person and the footstep parameter values obtained from footsteps of a person wearing the footwear of the footwear type,
The footwear type determining unit refers to the footwear type database, and determines the footwear type determining step that determines the footwear type corresponding to the footstep parameter value of the target person obtained by the parameter calculating unit as the footwear type of the target person. In addition,
The probability distribution database further stores, for each classification category, a probability distribution characteristic indicating a frequency distribution for each footwear type regarding footwear worn by persons belonging to the classification category,
In the category determination step, when calculating the evaluation value for each classification category, the probability value corresponding to the footwear type of the target person obtained by the footwear type determination unit is calculated for each classification category in the probability distribution database. The method for analyzing a walking sound, which is obtained from the above and integrated into the evaluation values of the corresponding classification categories. In the walking sound analysis method according to any one of claims 7 to 9,
The walking sound processing step generates an observation point signal indicating the sound generated at the observation point for each observation point provided around the subject by collecting the walking sound signal. For each observation point signal, an observation point feature value indicating an acoustic feature related to the observation point signal is calculated,
The parameter calculating step calculates a statistical value regarding each observation point feature value obtained by the walking sound processing unit as an observation point parameter value of the subject for each observation point,
The storage unit is an accessory that indicates a correspondence relationship between an accessory type related to an accessory attached to a person and each observation point parameter value obtained from the observation point signal of a person attached to the accessory of the accessory type. Further comprising the step of storing a type database;
The accessory type determination unit refers to the accessory type database and determines the accessory type corresponding to the observation point parameter value of the subject obtained by the parameter calculation unit as the subject type of the subject. The method further includes an accessory type determination step,
The probability distribution database further stores, for each classification category, a probability distribution characteristic indicating a frequency distribution for each accessory type related to an accessory attached to a person belonging to the classification category,
In the category determination step, when calculating the evaluation value for each classification category, a probability value corresponding to the subject accessory type is acquired from the probability distribution database for each classification category, and the corresponding classification is obtained. A walking sound analysis method, characterized by integrating each of the evaluation values of the categories. The walking sound analysis method according to any one of claims 7 to 10,
The parameter calculating step relates to any one or more of at least the subject's step length, walking speed, width from the movement center line in the left-right direction, and a time interval of one step as the walking state parameter value. A walking sound analysis method characterized by calculating statistical information or feature information indicating an acoustic feature of the footstep signal extracted from a spectrogram of the footstep signal. The walking sound analysis method according to any one of claims 7 to 11,
The category determination step includes a category category in which the highest evaluation value is obtained from among the evaluation values calculated for each category category, and an evaluation value exceeding a preset threshold value for each category category. Alternatively, the walking sound analysis method characterized in that the classification category having the largest difference or ratio between the evaluation value and the threshold value is determined as the classification category related to the subject.   The program for functioning a computer as each part which comprises the walking sound analyzer of any one of Claims 1-6.

Images