KR20140020440A - Activity diagnosis apparatus and method of the same - Google Patents

Abstract

An activity diagnosis apparatus of the present invention is able to reliably diagnoses the activity level of a subject in an unsupervised way without a pre-training about a system by including: a clustering unit which classifies multiple groups based on similarity of feature points which indicate the operation of the subject; and a diagnosis unit which diagnoses the activity level of the subject from the features of each group or the number of the groups.

Description

ACTIVITY DIAGNOSIS APPARATUS AND METHOD OF THE SAME}

The present invention relates to an apparatus and method for diagnosing an activity, and an apparatus and a method for reliably diagnosing the activity level of a subject from the operation of a subject.

Increasing interest in health and attempting to diagnose the activity level of a subject is generally a method of estimating calories based on surveys such as the Short Form-36 Health Survey (SF-36).

The subjects' activities are surveyed for a certain period of time, and converted into energy consumption values and metabolic equivalent of task (MET) values of actual activities. Finally, calorie consumption of the subjects is estimated. After that, an automated product was commercialized. The processing process of these products is to convert the sensor signal measured by the sensor part that adopts acceleration into Fast Fourie Transform energy for each period and map it with MET to estimate the calorie consumption of the subject in real time.

The existing automation method mainly calculates the energy generated from the exercise by measuring the amount of exercise, then roughly classifies the activity based on the energy value, or omits it and maps the energy value with the MET one to one. Therefore, it is possible to calculate the momentum, but it is not possible to classify the activity of the subject, and even if the diarrhea classification is used, only the energy is used as the basis of judgment, which is inaccurate.

In order to solve this problem, there has been an attempt to classify the activity type as the motion recognition of the teacher learning using the measurement signal of the sensor. However, it has been attempted to limit the number of recognizable motions for achieving a certain level of recognition rate because the minimum unit of motion is ambiguous, It is difficult to collect data for teacher learning. Therefore, the existing technology has limitations in applying to the aging diagnosis, in which the type of activity is the main evaluation factor along with the exercise amount.

Korean Patent Laid-Open Publication No. 2001-0099250 discloses a technology for providing various kinds of activity information on a given situation in an integrated manner to users, and for providing information on behaviors by analyzing information according to the situation of other people. However, it is far from diagnosing the level of activity of the subject related to the subject's behavior.

Korean Laid-Open Patent Publication No. 2001-0099250

The present invention is to provide an activity diagnosis apparatus and method capable of reliably diagnosing the activity level of the subject from the subject's operation.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory and are not intended to limit the invention to the precise forms disclosed. Other objects, which will be apparent to those skilled in the art, It will be possible.

The activity diagnosis apparatus of the present invention is characterized in that the activity diagnosis apparatus comprises a clustering section for classifying feature points representing an operation of a subject into a plurality of clusters based on the degree of similarity and a clustering section for clustering the clusters from at least one of the characteristics of the clusters, And a diagnosis unit for diagnosing the activity level of the patient.

Another activity diagnosis apparatus of the present invention includes a setting position of a sensing unit mounted on the subject in order to acquire activity data indicating an operation of a subject, a setting period related to how long a characteristic point, which is a representative value of the activity data, An operation method applied to the activity data to calculate the feature points, and a setting section for determining at least one of overlapping periods indicating the degree of overlapping of the setting periods.

The activity diagnosis method of the present invention may include obtaining activity data indicating a subject's motion, classifying the activity data or a feature point representing a representative value of the activity data into a plurality of clusters based on a similarity, and the number of clusters. Diagnosing the activity level of the subject.

According to another aspect of the present invention, there is provided a method for diagnosing activity, comprising the steps of: installing a sensing unit on a plurality of body parts of a subject; calculating a plurality of minutiae candidates by applying different operations to the activity data acquired for each of the plurality of setting intervals, A ratio of the covariance between each comparison cluster to the covariance within each comparison cluster is calculated for at least one of overlapping sections indicating the overlapping degree of the set position, the setting section, the minutia point candidate, and the setting period of the sensing section, And setting the at least one of the setting position, the setting position, the minutiae point candidate, and the overlapping interval indicating the overlapping of the setting period.

The apparatus and method for diagnosing an activity of the present invention clusters data representing a subject's behavior, diagnoses the subject's activity by considering the classified cluster as a kind of activity, and performs the subject's activity in a comparative manner without prior learning of the system. Can diagnose the level.

1 is a block diagram showing an activity diagnosis apparatus of the present invention.
FIG. 2 is a schematic view showing characteristic points where clustering is performed by clustering units.
Fig. 3 is a schematic view showing the activity diagnosis method of the present invention.
4 is a schematic view showing a case where FR is large and a case where FR is small.
5 is a schematic diagram showing the degree of variance of the FR value of the redundant section, the setting period, and the minutiae candidate in the activity diagnosis apparatus of the present invention.
Fig. 6 is a schematic diagram showing the results of FR comparison between feature point candidates in the activity diagnosis apparatus of the present invention. Fig.
7 is a schematic view showing the result of FR comparison between the mounting positions of the sensing units in the activity diagnosis apparatus of the present invention.
8 is a schematic diagram showing an external configuration of the activity diagnosis apparatus of the present invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. The sizes and shapes of the components shown in the drawings may be exaggerated for clarity and convenience. In addition, terms defined in consideration of the configuration and operation of the present invention may be changed according to the intention or custom of the user, the operator. Definitions of these terms should be based on the content of this specification.

1 is a block diagram showing an activity diagnosis apparatus of the present invention.

The activity diagnostic apparatus shown in FIG. 1 includes a clustering unit 150 for classifying feature points representing the subject's behavior into a plurality of clusters based on the degree of similarity, and a diagnosis unit 150 for diagnosing the activity level of the subjects from the characteristics or the number of clusters of the clusters (Not shown).

The level of activity diagnosed by the diagnosis unit 170 can be used for diagnosis of aging, diagnosis of activity, and the like.

The aging diagnosis method based on the motion recognition of a subject has inherent limitations in the design of the diagnostic apparatus as well as the performance of the recognition rate.

First, the design of the recognizer for the teacher learning type is based on the setting of the class to be classified. In other words, it is not easy to classify human behavior into several classes.

Second, even if the behavior is divided into several units of motion, it is also difficult to group the various variants of the same behavior into the same class and generalize the signal-dimensional data obtained from the various subjects.

Finally, it is also difficult to collect data and to annotate large quantities of data in class units to learn the diagnostic device, even if the two problems mentioned above are resolved.

As a countermeasure against this, the present invention proposes a population model based on a comparative learning as a new activity level model, taking the level of activity in which the subject does not specifically define the activity as a measure of aging diagnosis.

The number of activities that the subject can use is assumed to be inversely proportional to the degree of aging. This premise is documented by the existing surveys for aging diagnosis, ADL and IADL.

It is difficult to visually distinguish the subject's movements, but it is easy to classify the subject's movements based on signal similarity through signal representations at the collected signal level.

Classification of signal patterns classified by signal similarity by clustering algorithm can be used to estimate the number of similar pattern clusters in the signal dimension and estimate it as the number of possible actions by the subject. As a result, the AL (activity level) is defined as the number of similar pattern clusters in the signal dimension, the characteristics of the cluster, the connection characteristics between the clusters, and the cluster distribution by specific feature points.

The subject's motion can be obtained in a variety of ways. At this time, the subject's motion includes a lot of data, and feature points representing the subject's motion can be used for efficient processing. The feature point expresses the subject's motion with minimum point, position information, speed information, and the like.

By using the minutiae points, the processing load of the post-processing unit such as the clustering unit 150 and the diagnosis unit 170 can be reduced, and more clear information can be provided.

The activity diagnosis apparatus may further include a sensing unit 110 and an extraction unit 130. [

The sensing unit 110 senses the motion of the subject and outputs activity data indicating the motion of the subject. The sensing unit 110 may include one or more sensors. The sensing unit may output the sensing result by each sensor, or may output the signal result obtained by the combination of the sensors. The sensing unit 110 may include an acceleration sensor. The activity data of the subject measured through the acceleration sensor may include the gravity value as the magnitude of the acceleration. The gravity value can be excluded because it is independent of the subject's movement. For this purpose, the activity diagnosis apparatus may include a preprocessing unit 120 for removing the gravity value included in the activity data output from the acceleration sensor.

The preprocessing unit 120 may also remove other noise other than the gravity value. For this, the pre-processing unit 120 may include a low-frequency band filter. Since the general activity acceleration of the subject is less than 20 Hz, the low-frequency band filter of the preprocessing unit 120 can remove signals of 20 Hz or more.

The preprocessing unit 120 may calculate the magnitude of the vector of the acceleration signal collected from the sensing unit 110 at the same time in the X, Y, and Z axes as a three-dimensional vector as shown in Equation 1, have.

Here, the absolute value A is activity data,

a _x is the acceleration with respect to the X axis,

a _y is the acceleration with respect to the Y axis,

a _z is the acceleration along the Z axis,

9.8 is the gravity value.

The extractor 130 may extract a feature point representing the subject's motion with a minimum number of points from the activity data. The minutiae can be extracted in various ways, and the methods described below recall the illustrative ones.

The minutiae include a mean, a variance, a skewness, a kurtosis, a median, a mean absolute deviation, an activity count, a frequency domain energy, a time domain energy, a signal magnitude area, a spectral entropy Lt; / RTI >

Mean is a measure of the arithmetic mean of the distribution of data and can generally be expressed as: " (2) "

Variance is a measure of the degree of variability of the distribution of data and can generally be expressed as Equation (3).

Skewness is a measure of the skewness of the distribution of data and can be represented by Equation 4.

Kurtosis is a measure of the sharpness of the distribution of data and can be generally expressed by Equation (5).

Median represents the median of the data and is insensitive to the effects of high-frequency noise compared to Mean. Median can be generally expressed by Equation (6).

Mean Absolute Deviation (MAD) is the mean absolute deviation and is insensitive to the effects of high frequency noise compared to the variance. MAD can be generally expressed by Equation (7).

Activity Count means to set a specific value as a threshold value in a certain interval, and to count the number of signals appearing above the threshold value. However, for convenience of calculation, the highest value is taken as the representative value in the data of the predetermined interval. Activity Count can be expressed by Equation (8).

Frequency domain energy is energy calculated in the frequency domain and may be represented by Equation 9.

Time domain energy is energy calculated in the time domain and may be represented by Equation 10.

The Signal Magnitude Area (SMA) is obtained by calculating the absolute value of the signal for each axis and then adding the signal.

The spectral entropy can be expressed by Equation (12) as the entropy obtained for the probability distribution of the frequency coefficients obtained by obtaining the spectrum of the signal through the FFT (fast Fourier transform) and taking the sum of the frequency coefficients as 1.

The extracting unit 130 generates a feature point vector X = (x _1, x ₂ , ..., x _d ) ^T by a combination of the feature points calculated in the above Equations (2) to (12). As the dimension of the feature point vector increases, the accuracy of classification and clustering in the clustering unit 150 increases.

The clustering unit 150 may classify activity data or feature points representing a subject's motion into a plurality of clusters based on similarity.

The goal of clustering is to make the samples in the same cluster close to each other and the distance between the samples belonging to different clusters to be far apart. Therefore, the concept of distance in clustering is important and there are several calculation methods. Depending on the selected calculation method, the result of clustering is different. Therefore, it is important to choose the appropriate distance measurement method for the given problem. Distance and similarity are the opposite concepts, and knowing one can easily calculate the other using the formula.

The clustering unit 150 may form or classify clusters of activity data or feature points based on similarity without prior knowledge of activity data or feature points. The clustering in the clustering unit 150 requires a distance measure and a clustering algorithm. Distance scale is inversely related to similarity, so distance scale can be expressed as similarity.

The similarity may be expressed by a mathematical expression such as a cosine similarity degree, a similarity degree of a binary characteristic vector, or the like.

The cosine similarity is used for document retrieval and the like, and a specific word is featured, and the appearance frequency of a specific word can be a feature point. Cosine similarity can be represented by the following equation (13).

The binary feature vector can be expressed by the following equation (14).

&Quot; (14) "

The similarity and distance can be transformed through the following equation (15).

s is the similarity, and d is the distance. The subscript max indicates the maximum value.

Clustering is a type of comparative company learning to classify and forms or classifies a cluster of samples such as feature points based on similarity without knowledge of samples such as feature points. Similarity is required for clustering. The degree of similarity is inversely related to the distance scale. Therefore, it is possible to apply a clustering algorithm that uses the distance scale after finding the similarity and converting it to the distance scale or the distance scale directly.

The distance scale may be obtained by applying Minkowski Street, Euclidean Street, Manhattan Street, Hamming Street, and the like.

In summary, the clustering unit 150 may calculate the similarity from the Minkowski distance, Euclidean distance, Manhattan distance, Hamming distance, etc. between each feature point.

In addition, the clustering unit 150 may cluster the feature points using a hierarchical clustering algorithm, a partitioned clustering algorithm, a neural network algorithm, a statistical search algorithm, or the like based on the similarity or the distance measure converted from the similarity.

..., c _2n } is the part of the cluster belonging to the different clusters C ₁ = {c ₁₁ , c ₁₂ , ..., c _1k }, and all the clusters c _2i of the clusters C ₂ = {c ₂₁ , c ₂₂ , When set, C ₂ is said to be included in C ₁ . For example, C ₂ = {{x ₁ , x ₃ , x ₆ }, {x ₂ }, {x ₄ , x ₅ }} gives C ₁ = {{x ₁ , x ₃ , x ₆ }, {x ₂ , x ₄ , x ₅ }}. At this time, n> k. The layer clustering algorithm outputs the clustering result as clustering results. Metrology clustering algorithms include a cohesive method of starting from small clusters and collecting them and a division method of dividing them from a large cluster.

The partition clustering algorithms include sequential algorithms, k-means algorithms, MST algorithms, and Gaussian mixture model (GMM) algorithms.

The sequential algorithm is an algorithm that finds the number of clusters. Instead, you need to set thresholds.

The k-means algorithm is often used because the number of clusters must be set, but it is easy to understand. If the number of clusters includes an item that evaluates the activity level, it is necessary to be parallel with other clustering algorithms to obtain the number of clusters.

The neural network algorithm includes SOM algorithm, ART algorithm, and so on.

The self organizing map (SOM) algorithm uses competitive learning as an algorithm that compares samples and organizes them by themselves.

The Minkowski distance d _ij represents a distance measure between two points and is expressed by equation (16). When p = 1, it is the Manhattan distance, and when p = 2, it is the Euclidian distance.

The Hamming distance uses the number of different bits as a distance measure that can be used for binary feature vectors.

FIG. 2 is a schematic view showing characteristic points where clustering is performed by the clustering unit 150. FIG.

Referring to FIG. 2, a plurality of feature points are displayed, and a state classified into five clusters by the clustering unit 150 is disclosed. Since each cluster at this time represents a specific operation of the subject, the number of clusters corresponds to the number of operations of the subject. Since the minutiae included in the respective clusters are similar patterns based on the similarity, the result output from the clustering unit 150 becomes the AL as described above.

The diagnosis unit 170 can diagnose the activity level of the subject from the characteristics or the number of clusters of the clusters formed in the clustering unit 150. Although the activity level can be determined only by the number of clusters, it is also possible to identify the activity level by the characteristics of each cluster. Of course, the activity level can be determined by using the characteristics of the clusters, the number of clusters, and the distribution of clusters by specific feature points.

The characteristics of the cluster can be defined by the area of each cluster, the distribution of feature points in each cluster, the feature point concentration in each cluster, the average energy of each cluster, and the energy deviation of each cluster.

According to the above-mentioned activity diagnosis apparatus, the activity classification is emphasized as compared with the existing diagnosis method, so that it is possible to diagnose the activity level more specifically. The number of the population resulting from clustering, the area information per cluster, , And the concentration of the signal. In addition, the calorie consumption meter, which is already released as a commercial product, calculates calorie consumption using energy-related feature point extraction, and thus can replace the existing commercial product.

3 is a schematic diagram illustrating an activity diagnosis method of the present invention and may be described as an operation of the activity diagnosis apparatus of FIG. 1. In FIG. 3, although the acceleration signal obtained by the acceleration sensor is shown, it is evident that the signal used may vary depending on the configuration of the sensing unit.

First, activity data indicating an operation of the subject is obtained (S 510). The sensing unit 110 may include an acceleration sensor. In this case, the activity data output from the sensing unit 110 may include an acceleration signal.

Then, the feature points, which are representative values of the obtained activity data or activity data, are classified into a plurality of clusters based on the degree of similarity. In operation S 530, the activity data or each feature point may be classified according to a signal pattern (S 520), and grouped into signal patterns classified in a time-series manner by operation performed by the clustering unit 150 (S 530). The clustering object may be the activity data output from the sensing unit 110. However, since the amount of data to be processed is large, it may be advantageous to use the characteristic point.

Next, the activity level of the subject is diagnosed by identifying (S540) the characteristics or the number of clusters formed by clustering. The activity level of the subject can be diagnosed by comparing the characteristics of the community or the number of clusters identified by the operation performed by the diagnosis unit 170 with the previously-examined data (S 550).

In the past, diagnosis of aging was made by understanding the number of motions of the subject through a survey, recognizing the motion using the acceleration signal obtained from the sensing unit 110, and grasping the number of motions immediately without a separate clustering. This, as mentioned above, faces various problems, but according to the present invention, such a problem can be solved.

The subject's actions can be represented by activity data or minutiae. At this time, it is advantageous to use feature points.

The method of extracting feature points from a plurality of activity data output from the sensing unit 110 may vary. Depending on the method of extracting the feature points, the efficiency of subsequent processing using the feature points may vary. By reducing the size of the activity data according to discovering the optimal feature point, it is possible to reduce the calculation time of the clustering unit 150 and the like and maximize the motion recognition performance. Therefore, it is necessary to extract the optimal feature point, which will be described below.

The activity diagnosis apparatus may include a setting unit 190 that sets various elements involved in the extraction of the feature points so that the extraction unit 130 extracts optimal feature points.

The element at this time may be the position of the sensing unit 110 mounted on the subject, the cycle of calculating the minutiae points, the calculation method applied to the minutiae point calculation, the degree of redundancy of the minutiae pointing cycle, and the like.

Specifically, the setting unit 190 sets the mounting position of the sensing unit 110 mounted on the subject in order to obtain the activity data indicating the subject's motion, and how long the characteristic points, which are representative values of the activity data, A setting period, an operation method applied to the activity data to calculate the minutiae, and an overlapping interval indicating the degree of overlap of the set period can be determined.

The sensing unit 110 may be mounted on various body parts of the subject. However, the efficiency of the feature point varies depending on the position of the sensing unit 110 placed on the subject.

For example, it is assumed that the subject's level of activity belongs to the young group (20 to 34 years old), the middle age group (35 to 49 years), or the elderly group (50 to 65 years) .

At this time, the young group, middle aged group, and mature group may be a comparative cluster that results from the diagnosis of the subject's activity.

In this case, the result of mounting the sensing unit 110 on the head and the arm of the subject may be different. For example, when the sensor 110 is mounted on the arm, the activity data shows a big difference according to the young, middle-aged, and middle-aged group, whereas when the sensor 110 is mounted on the head, the difference between each comparison cluster is different. It may appear small. In this case, it is advantageous to mount the sensing unit 110 on the arm.

In summary, the setting unit 190 can determine the optimum mounting position of the sensing unit 110 in the body part of the subject. For example, the setting unit 190 may determine the mounting position where the sensing unit 110 is mounted on the body 7 of the subject and then the optimal feature point can be extracted. The thus determined mounting position can be applied at the time of full-scale activity diagnosis.

The body 7 on which the sensing unit 110 is mounted may be the left waist of the subject, the main wrist, the main arm, the opposite wrist, the main thigh, the main ankle, and the pants pocket. In addition, the sensing unit 110 can be mounted on the right waist. At this time, the sensing unit 110 mounted on the right waist can collect data for analyzing the correlation between active energy and aging.

The main wrist and main arm are the right hand side for the right hand and the left hand side for the left hand. The main thigh and the ankle are on the left foot for right handed and on the right foot for left handed.

The pants pocket position is assumed to be a case where a portable terminal that can be stored in the pants pocket (the right bag pocket in the experiment) is configured as an activity diagnosis device.

The experiment process consisted of 18 non-aged subjects (Table 1) who were 20 to 60 years of age (Table 1), and the activity data generated in daily life for 4 hours per subject was collected by the acceleration sensor. At this time, the activity of the subject was recorded by taking a photograph at least once a minute using the camera in synchronization with the accelerometer signal to be collected.

20's 30s 40s 50s 60s 6 4 3 4 One sir Girl south Girl south Girl south Girl south Girl 4 2 3 One 2 One 2 2 0 One

Vision, various sensors, and RFID can be introduced for activity level or aging diagnosis. However, in this experiment, an acceleration sensor with excellent performance ratio and no restriction in place was used. At this time, the sensing unit 110 includes an acceleration sensor, a microprocessor, a Bluetooth module, and a digital clock. The acceleration sensor was a Freescale MMA7260QT, and the X, Y, and Z axis acceleration measurement ranges were set to ± 2g and the sampling frequency was fixed to 100Hz. For the analog-to-digital conversion of the acceleration sensor, ATmega8L of Atmel quantizes it to 8 bits and transmits data to the main server through the Bluetooth module. The main server has a storage unit for storing the transmitted data.

After collecting activity data for each part where the sensing part 110 is mounted, noise and gravity values are excluded through the preprocessing part 120. Then, the minutiae point candidates are calculated by applying the arithmetic formulas (2) to (12) to the output value of the preprocessing unit 120, respectively. Since the data output from the sensing unit 110 lasts for a certain period of time (4 hours in the experiment), it may be important to extract the characteristic points at certain intervals. Therefore, the above calculation method can be applied to at least one of Mean, Variance, Skewness, Kurtosis, Median, Mean Absolute Deviation, Activity Count, Frequency Domain Energy, Time Domain Energy, Signal Magnitude Area, Spectral Entropy, It can be one. In the experiment, the set period was set to 15 kinds of 1s, 5s, 10s, 15s, 30s, 45s, 60s, 75s, 90s, 105s, 120s, 135s, 150s, 165s and 180s. The overlapping sections were set to 6 types of 0%, 10%, 20%, 30%, 40% and 50%. For example, if the setting period is 1s and the overlapping interval is 0%, each calculation method is applied to the activity data collected for 1 second at intervals of 1 second. If the set period is 1s and the overlapping interval is 50%, the application of each calculation method to the activity data collected for one second is the same. 50% redundant activity data such as 0s ~ 1s, 0.5s ~ 1.5s, 1s ~ 2s, 1.5s ~ 2.5s are collected. That is, even if the same set period is used, the number of data to be applied to the operation method to be collected increases when the overlapping interval is increased.

In order to select feature points among the 11 minutiae candidates, the comparative clusters are classified into a young group (group 1), a middle-age group (group 2), and an adult group (group 3) ) And Fisher Ratio (FR).

FR is the ratio of covariance between societies (S _B ) to covariance (S _W ) in the community, and the larger the FR, the higher the difference between societies. When one-dimensional feature points are used, the variance of each data can be used instead of the covariance.

FR can be calculated by the following equation (14).

4 is a schematic diagram showing a case where the FR is large and small.

The left diagram in FIG. 4 shows a state in which the respective comparison communities can be clearly distinguished, and the right diagram shows a state in which it is difficult to distinguish the respective comparison communities. The computed FRs for each minutiae candidate are shown differently, and minutiae candidates representing the results close to the left figure can be taken as minutiae. At this time, the setting unit 190 may use the calculation method that calculates the minutia point candidates taken as minutiae points in the extracting unit 130. [

5 is a schematic diagram showing the degree of variance of the FR value of the redundant section, the setting period, and the minutiae candidate in the activity diagnosis apparatus of the present invention.

The following results were obtained based on the obtained FR values for the seven sensing unit 110 mounting positions, eleven feature point candidates, 15 set periods, and six overlapping sections.

First, we analyze the relative influence of 11 minutiae candidates, 15 setting periods, and 6 redundant sections on FR. FR value variation of 11 feature point candidates obtained without distinction of installation location / setup period / redundant section, 15 FR values (windowsize) obtained in 15 setup periods (windowsize) And the degree of variance of FR values occurring in six types of overlaps without discriminating between the mounting position / feature point candidate / setting period are shown in FIG. The difference in setting period is analyzed to have a small effect on overall cluster differentiation. However, as the length of the overlap period increases, the FR value increases.

Fig. 6 is a schematic diagram showing the results of FR comparison between feature point candidates in the activity diagnosis apparatus of the present invention. Fig.

The FR comparison between the feature point candidates obtained without the distinction between the mounting position, the overlapping interval, and the setting period shows the FR having a simple representative value for each set period of time series data rather than the energy series feature point or the statistical series feature point. Mean, Median, AC, and SMA are strong candidates, and the mean difference is also statistically significant.

7 is a schematic diagram showing the results of FR comparison between the mounting positions of the sensing unit 110 in the activity diagnosis apparatus of the present invention.

In the case of FR comparison for each mounting position obtained without distinguishing between redundant section, set period, and minutiae point, there is no statistical significance due to a large gap between feature points and overlapping sections, but positions of waist, ankle and pocket are suggested as candidates. Considering that the high FR of the ankle is unexpectedly accepted, but the individual characteristics are relatively low compared to the wrist or arm, the movement of the foot eventually discriminates human behavior characteristics and affects the entire community differentiation can see.

In summary, the setting unit 190 can determine at least one of a mounting position, a setting period, an operation type, and an overlapping interval of the sensing unit 110 that satisfies the set value of the ratio of the covariance between each comparison cluster and the covariance within each comparison cluster have.

The activity diagnostic apparatuses described above represent the subject 's motion in the signal dimension. This is done by presenting a method of extracting minutiae and suggesting an optimum mounting position of the sensing unit 110. [

The activity diagnosis apparatus may calculate a plurality of feature point candidates by applying different operations to activity data obtained for each of a plurality of setting sections from each of the sensing units 110 installed in the body part of the subject. Thereafter, the ratio of the covariance between each comparison cluster and the covariance within each comparison cluster is calculated for at least one of overlapping sections indicating the overlapping degree of the setting position, the setting interval, the minutia point candidate, and the setting period of the sensing unit 110, At least one of an installation position, a setting period, a minutiae point candidate, and an overlapping interval indicating the overlapping of the setting period may be set.

8 is a schematic diagram showing an external configuration of the activity diagnosis apparatus of the present invention.

The illustrated activity diagnostic apparatus may include a sensing unit 110 and a processing unit 180 for storing results measured by the sensing unit and for performing additional calculation.

The sensing unit 110 includes a sensor unit 111 for extracting an operation of the subject in real time, an RTC (Real Time Clock) unit 113 for setting the operation time of the sensor unit in real time, Unit 115 and a communication unit 117 for transmitting the measurement result to the processing unit. In addition, the sensing unit 110 may include a control unit 119 that manages whether the sensor unit is operated, RTC setting, management of measurement results stored in the storage unit, and whether the communication unit is operated.

The processing unit 180 may include the extraction unit 130, the clustering unit 150, the diagnosis unit 170, and the setting unit 190 described above.

The processing result of the processing unit 180 may be provided to the subject management server 160 of the medical institution or the protection facility via the wired / wireless communication network.

While the invention has been shown and described with reference to certain preferred embodiments thereof, it will be understood by those skilled in the art that various changes and modifications may be made without departing from the spirit and scope of the invention as defined by the appended claims. Accordingly, the true scope of the present invention should be determined by the following claims.

110 ... sensing unit 111 ... sensor unit
113 ... RTC unit 115 ... storage unit
115 ... storage unit 119 ... control unit
120 ... pre-
130 ... Extraction unit 150 ... Clustering unit
160.Subject Management Server 170 ... Diagnostic
180 ... processing unit 190 ... setting unit

Claims (10)

A clustering unit classifying a feature point representing a subject's motion into a plurality of clusters based on the similarity; And
A diagnosis unit for diagnosing the activity level of the subject from at least one of the characteristics of each cluster, the number of clusters, and a distribution of clusters for specific feature points;
Activity diagnostic device comprising a.
The method of claim 1,
The clustering unit calculates the similarity from at least one of Minkowski distance, Euclidean distance, Manhattan distance, and Hamming distance between the feature points.
And clustering the feature points using at least one of a hierarchical clustering algorithm based on the similarity, a segmentation clustering algorithm, a neural network algorithm, and a statistical search algorithm.
The method of claim 1,
The feature of the cluster may include at least one of the area of each cluster, the distribution of the feature points in the clusters, the concentration of the feature points in the clusters, the average energy of each cluster, and the energy deviation of each cluster.
The method of claim 1,
A detector configured to detect the movement of the subject and output activity data indicating the movement of the subject; And
And an extracting unit for extracting the feature points expressing an operation of the subject from the activity data to a minimum point.
5. The method of claim 4,
The sensing unit includes an acceleration sensor,
The activity data output from the acceleration sensor is a gravity value is applied,
And a preprocessor for removing the gravity value included in the activity data.
In order to calculate the feature point, the setting period related to the mounting position of the sensing unit mounted to the subject, how long the feature point, which is a representative value of the activity data, is to be acquired, and the feature point is obtained to obtain activity data indicating the operation of the subject. A setting unit configured to determine at least one of an operation scheme applied to activity data and an overlapping section indicating a degree of overlapping of the setting period;
Activity diagnostic device comprising a.
The method according to claim 6,
And the setting unit determines at least one of the mounting position, the setting period, the calculation method, and the overlapping interval in which the ratio of the covariance between the respective comparison clusters and the covariance in each comparison cluster satisfies a setting value.
The method according to claim 6,
The calculation method includes at least one of Mean, Variance, Skewness, Kurtosis, Median, Mean Absolute Deviation, Activity Count, Frequency Domain Energy, Time Domain Energy, Signal Magnitude Area, Spectral Entropy, One activity diagnostic device.
An activity diagnosis apparatus for classifying signal patterns classified based on similarity for each signal into a plurality of clusters through a clustering algorithm, and diagnosing an activity level of a subject by the number of classified similar pattern clusters,
Obtaining activity data indicative of the subject's motion;
Classifying a feature point, which is a representative value of the activity data or the activity data, into a plurality of clusters based on similarity; And
Diagnosing the activity level of the subject from the features of the cluster or the number of clusters;
Activity diagnostic method comprising a.

Images