KR20130039625A - Method and apparatus for circadian life pattern modeling and anomaly detection for living alone - Google Patents

Abstract

PURPOSE: A method for detecting abnormality and modeling a life pattern of a person living alone and a device thereof are provided to confirm the abnormality of a person living alone by modeling the life pattern of the person based on sensing information and calculating an abnormal level of activities of the person based on the life pattern. CONSTITUTION: A home monitoring module transmits measurement data sensed through one or more sensors at home to a server computer(S100). The server computer models a life pattern of a person living alone based on the measurement data(S200). The server computer calculates the abnormality of the person based on generated life pattern modeling information and detects an abnormality level by comparing the calculated abnormality with a set threshold value(S300). [Reference numerals] (AA) Start; (BB) End; (S100) Measuring a position and the time of staying at home, the time of leaving, and sleeping hours and transmitting to a server computer; (S200) Modeling a living pattern of 24-hour cycle; (S300) Detecting the generation of abnormality by comparing a calculated degree of abnormality with a threshold value; (S400) Transmitting abnormality generation data to a carer terminal when abnormality is generated;

Description

Life pattern modeling and anomaly detection method and device for living alone {Method and apparatus for circadian life pattern modeling and anomaly detection for living alone}

The present invention relates to a method and apparatus for modeling a living pattern of a single person and performing abnormal detection.

As the recent UN report points out, the aging of lower birth rates and life expectancy is a global phenomenon. Moreover, if Korea's low fertility rate continues, one of the three countries (Japan, Macau and Korea) with the highest aging population (more than 40% of elderly population over 60 years old) in 2050 will be compared to other developed countries. It is expected to be. This rapid aging will bring about national problems in all aspects of politics, economy and society. In particular, the financial burden of medical expenses for the elderly is expected to increase significantly. Therefore, various fields are trying to solve these problems. Among them, due to the reduced effect of elderly medical expenses using u-Healthcare technology using IT (currently, the cost of maintaining the health of healthy elderly for a long time is much lower than the cost of treatment after illness). Development is being done.

In particular, as the nuclear family grows faster, the number of people living alone in their families will increase. Although single people are very likely to be seriously injured by various accidents (falls, etc.) in their homes, social interactions are relatively low compared to other generations, resulting in more serious injuries due to delays in accident discovery. There may even be deaths. Therefore, behavior monitoring technology that installs several sensors in the house and processes the detected data to automatically detect such accidents and notify appropriate providers (family, welfare workers, etc.), u-Healthcare field as well as situational awareness ( Context-aware) Since it is a necessary technology for many applications of computing, intensive R & D has recently been made.

Among these behavior monitoring techniques, a method of detecting daily activities and deriving a living pattern from them has been proposed. The life pattern detecting technique includes information detected by the home sensing system (location / movement of a resident, whether a specific object is used, and a basis). Attention to periodicity, repeatability, etc.

Various sensing systems and technologies are researched and developed for this purpose. Some research groups in Japan, France, and the United States have researched and developed a method of modeling the elderly's life patterns and detecting abnormalities based on environmental sensors. And as shown in the following non-patent documents 1 and 2, G. Virone proposed a series of methods of using the motion sensor to identify the residents' movement and to create a 24-hour life pattern model.

Virone noted that most people's daily life repeats on a daily basis (ie, a 24-hour cycle) and that certain daily activities are performed repeatedly at similar times. A modeling method called Circadian Activity Rhythm (CAR) is proposed.

However, the above-described conventional methods do not accurately reflect time information on various daily activities when performing the life pattern detection, and precisely how the daily activities differ from the usual based on the detected life pattern information. There was a problem in that it could not be monitored quickly and thus could not cope quickly with emergencies.

Patent Publication No. 10-2009-109775 2009. 10. 21. Patent Registration No. 10-0917605 2009. 09.09. Patent Publication No. 10-2010-26850 2010.03.10.

 G. Virone, N. Noury, J. Demongeot, "A system for automatic measurement of circadian activity deviations in telemedicine," Biomedical Engineering, IEEE Transactions on, vol. 49, no.12, pp. 1463-1469, Dec. 2002.  G. Virone, M. Alwan, S. Dalal, S.W. Kell, B. Turner, J.A. Stankovic, R. Felder, "Behavioral Patterns of Older Adults in Assisted Living," Information Technology in Biomedicine, IEEE Transactions on, vol. 12, no. 3, pp. 387-398, May. 2008.

An object of the present invention is to recognize daily activities through remote monitoring equipment installed in a single resident's house, modeling a 24-hour life pattern that reflects the repeatability of the daily activities based on this information. An object of the present invention is to provide a life pattern modeling, anomaly detection method, and an apparatus, which are independent to detect anomaly based on modeling information.

Another object of the present invention is to model life patterns as a fuzzy set that reflects the time repeatability of feature components including staying, going out, sleeping, etc. in each room constituting the daily life pattern of a single person, based on this modeling information. The present invention is to provide a method and apparatus for life pattern modeling and abnormal detection, which is the only one that can check abnormality by calculating a degree of anomaly, which is a measure of how different daily activities of a particular day are from usual.

Life pattern modeling and anomaly detection method, the living alone according to an embodiment of the present invention for achieving this object, (1) in the premises monitoring module, through one or more sensors installed in each room, toilet, entrance, kitchen in the home Transmitting the detected measurement data to the server computer through a network communication network, and (2) the server computer based on the measurement data received from the in-house monitoring module through the network communication network. Generate sleep time, and create fuzzy sets for each feature, including stay time in each room, toilet, and kitchen, outgoing behavior, and sleep behavior based on the resident's home location and residence time, outgoing time, and sleep time. And modeling the life patterns of the living alone in the 24-hour period by the logical product of the fuzzy sets for each feature, and (3) Based on the life pattern modeling information generated in step (2), the server computer calculates a degree of anomaly, which is a measure of the difference between the living pattern of a single person and a daily living pattern, and the calculated abnormality. And detecting whether an abnormality occurs by comparing the degree with a preset threshold.

In addition, the life pattern modeling and abnormality detection method of a single person according to an embodiment of the present invention, (4) the server computer monitors the registered provider terminal when the abnormality of the single person is detected as a result of the processing of step (3) The method may further include transmitting data about an abnormal occurrence of the target reader as text or voice data.

In addition, the single life pattern modeling and abnormal detection device according to an embodiment of the present invention, by connecting a plurality of unspecified communication lines to each other to perform data communication related to the life pattern modeling and detection of abnormal occurrences of a single person In-house monitoring module that transmits the measurement data detected by the communication network and one or more sensors installed in each room, toilet, entrance, and kitchen of the home to the server computer through the network communication network, and the measurement received from the home monitoring module through the network communication network. Based on the data, the resident's home location and dwelling time, time to go out, sleep time is generated, and based on the resident's home location and dwell time, time to go out, sleep time in each room, toilet and kitchen, Create fuzzy sets for each feature, including outing and sleep behaviors, Create fuzzy sets for each feature, including residence time, outing behavior, and sleeping behavior in the room and kitchen, and model living patterns of the living alone in the 24-hour period by the logical product of fuzzy sets for each feature. Based on the pattern modeling information, the degree of abnormality, which is a measure of the difference between a single daily life pattern and a daily life pattern, is calculated, and the abnormality is detected by comparing the calculated abnormality with a preset threshold. It can include a server computer.

In addition, the single life pattern modeling and abnormality detection device according to an embodiment of the present invention, the communication provider terminal receiving data about the occurrence of the abnormal target of the monitoring target alone from the server computer via a network communication network as text or voice data It can be configured to include.

As described above, according to the life-pattern modeling and abnormal detection method and apparatus of the present invention, the life-pattern is modeled on the basis of the sensory information on the daily activities of the household, and based on this, the daily activity of a specific day By calculating the degree of abnormality that indicates how different is the presence of a single person, it is possible to easily monitor the activity status of a single person remotely.

In addition, if an abnormality of a single person is found, it is effective to promptly respond to a crisis situation by promptly notifying an emergency situation to a provider (a public service officer such as a family member or a welfare worker or a manager).

1 is a block diagram schematically showing the configuration of the life pattern modeling and abnormality detection apparatus, the living alone according to an embodiment of the present invention;
2 is a block diagram showing in detail the configuration of the indoor monitoring module of FIG.
3 is a block diagram showing in detail the configuration of the server computer of FIG.
Figure 4 is a flow chart showing in detail the operation of the life pattern modeling and abnormal detection method alone living according to an embodiment of the present invention,
5 and 6 are flowcharts showing in detail the operation of each subroutine of FIG.
7 is a view showing an example of a fuzzy set based on a 5-day feature value for the outgoing feature component of a particular elderly person to which the present invention is applied;
8 is a view showing an example of the installation plan of the house plan and home monitoring equipment of a specific elderly people to which the present invention is applied,
9 is a view showing an example of the outgoing fuzzy set of three elderly people of FIG.
FIG. 10 is a diagram illustrating an example of models of six feature components of the elderly A of FIG. 8.

Hereinafter, with reference to the accompanying drawings will be described in detail the life-pattern modeling and abnormal detection method and apparatus thereof.

1 is a block diagram schematically showing the configuration of the life pattern modeling and abnormality detection apparatus alone in accordance with an embodiment of the present invention, Figures 2 and 3 are in detail the configuration of the indoor monitoring module and server computer of Figure 1 The block diagram shown.

As shown in FIG. 1, the apparatus of the present invention includes a network communication network 100, an indoor monitoring module 200, a server computer 300, a provider terminal 400, and the like.

The network communication network 100 is a variety of communication networks including wired / wireless Internet. The network communication network 100 connects communication lines between a home monitoring module 200 and a server computer 300, a server computer 300, and a provider terminal 400 to each other. Data communication related to life pattern modeling and anomaly detection of a single person (eg, elderly person, patient, etc.) is performed.

The indoor monitoring module 200 is composed of a motion sensor, a sleep sensor, a door sensor, and installed in each room, toilet, entrance, and kitchen of the home, and the measured data detected through each sensor is transmitted through the network communication network 100. Send to computer 300.

The server computer 300 generates the resident's home location and dwell time, outing time, and sleep time based on the measurement data transmitted from the premises monitoring module 200 through the network communication network 100, and generates the resident's home. Create fuzzy sets for each feature, including dwelling time in each room, bathroom, and kitchen, going out, and sleeping, based on location, dwelling time, outing time, and sleeping time, and a logical product of the fuzzy sets for each feature. Model the life patterns of the resident's 24-hour cycle. Based on the life pattern modeling information, the degree of abnormality, which is a measure of the difference between the living pattern and the daily life pattern of the living person, is calculated, and the calculated abnormality is compared with a preset threshold to determine whether an abnormality has occurred. Detect.

In addition, the server computer 300 transmits a voice or text message indicating that the problem has occurred to the single owner to the registered provider terminal 400 when the problem occurs.

The provider terminal 400 is a communication terminal possessed by a provider such as a family or welfare worker, a public service officer such as a manager, and when an abnormality occurs in a single person, an abnormality from the server computer 300 through the network communication network 100. Receive data about the occurrence as text or voice data.

FIG. 2 is a detailed diagram illustrating the configuration of the indoor monitoring module 200 of FIG. 1 and includes a power supply unit 210, a sensor unit 220, a control unit 230, a data communication unit 240, and the like.

The power supply unit 210 supplies power for the operation of the device.

The sensor unit 220 is composed of a motion sensor, a door sensor for detecting the opening / closing of a door, and a sleep sensor using a plurality of force sensitive resistors (FSRs), and are installed in each room, toilet, entrance, and kitchen of the home. Measure the position of the phosphorus, going out, and sleep.

The controller 230 controls the measurement data measured by the sensor unit 220 to be transmitted to the server computer 300 through the data communication unit 240.

The data communication unit 240 transmits the measurement data measured by the sensor unit 220 to the server computer 300 under the control of the controller 230.

3 is a diagram illustrating the configuration of the server computer 300 of FIG. 1 described above in detail, and includes a data communication unit 310, a control unit 320, a life pattern modeling unit 330, an abnormality detection unit 340, and an abnormality occurrence processing unit. 350, the database 360, and the like.

The data communication unit 310 receives measurement data from the indoor monitoring module 200 through the network communication network 100, and transmits abnormality occurrence data to the provider terminal 400.

The control unit 320 generates and stores the home location and residence time of the resident, outing time, sleep time based on the measurement data transmitted from the home monitoring module 200 through the data communication unit 310, and the life pattern modeling unit. Generation and modeling of a fuzzy set that reflects time repeatability based on feature values for a plurality of days for each feature component performed at 330, and a specific day of the relevant single person performed by the abnormality detector 340. For abnormality detection through the calculation of the abnormality degree, which is a measure of the difference between the life pattern and the usual life pattern, and comparison of the abnormality value and the threshold value as a reference for occurrence of the abnormality, and the abnormality detection performed by the abnormality occurrence processing unit 350. Control the initiation provider terminal 400 notification of related data.

The life pattern modeling unit 330 generates a fuzzy set reflecting time repeatability based on feature values of a plurality of days for each feature component based on the control of the controller 320, and performs a corresponding multiplication by performing a logical product of the fuzzy sets. Model the life patterns of phosphorus in the 24-hour cycle.

The abnormality detection unit 340 checks the normality of a specific day of each characteristic component of the corresponding living person based on the life pattern modeling information generated by the life pattern modeling unit 330 based on the control of the control unit 320, and is normal. After normalizing the degree values, using the average and standard deviation of the normalized values of the specific days used to generate the fuzzy set for each feature of the individual The degree of abnormality, which is a measure of the difference between the patterns, is calculated, and the abnormality value is compared with a threshold value that is a criterion for occurrence of abnormality.

In this case, the threshold value used as a criterion for occurrence of abnormality is preferably determined in advance based on the accumulated experiment experience.

When the abnormality detection unit 340 detects an abnormality of the single resident through the abnormality detection unit 340 based on the control of the control unit 320, the abnormality occurrence processing unit 350 may provide information about the provider who is responsible for the single resident already registered in the database 360. And the data related to the detection of abnormality is transmitted to the corresponding provider terminal 300 as text or voice data.

The database 360 stores the provider information (for example, communication terminal number, name, address, etc.) of the monitoring targets alone based on the control of the control unit 320, and measures received from the indoor monitoring module 200. Storage for each feature component of the data, storage of life pattern modeling information of each living person generated through the life pattern modeling unit 330, abnormality detection information of the abnormality detection unit 340 and abnormality transmitted to the provider terminal 400 Process the storage of detection data.

If described in detail with reference to Figures 4 to 7 an embodiment of the life-pattern modeling and abnormal detection method that is alone in accordance with the present invention configured as described above.

4 to 6 are flowcharts showing in detail the operation process of the life pattern modeling and abnormal detection method, which is a single life according to an embodiment of the present invention, and FIG. 5 illustrates an example of a fuzzy set based on five feature values.

First, the indoor monitoring module 200 transmits the measurement data detected through one or more sensors installed in each room, toilet, entrance, and kitchen of the home to the server computer 300 through the network communication network 100 (S100).

Step S100 tracks the location of the resident in the home, finds the time spent in the separated space (that is, the time of stay), and checks the behavior and time of going out, sleeping, and the like. In the study of Virone, the motion sensor was installed in the space where the elderly stayed mainly during daily activities. In this case, it is detected when the single person is in the space where daily activities are performed. Modeled. This method detects the presence of a single person in the main activity space, so the frequency of sensor data is increased and accuracy is increased. However, as the detection frequency of the sensor is increased, the battery life is increased and the operating life is shortened.

Accordingly, in the present invention, unlike the conventional method, the motion sensor is installed at a position (entrance door or aisle) that distinguishes a main area of a house (a bedroom, a small room, a toilet, a kitchen, a porch, etc.). As a result, only the case where the living alone is detected, the result is transmitted to the server computer 300 and stored. Therefore, the overall detection frequency is lowered, which has the advantage of longer operation. However, since data is only detected in position movement, it is difficult to obtain residence time information in each space. In the present invention, a position tracking method using motion sensor data generated during movement is developed and applied.

The basic idea of this method is very simple: if a sensor is installed between spaces A and B, a detection signal is generated when moving from A → B or B → A. Can be. Moreover, since the present invention uses sleep sensor data, more accurate initial position can be estimated. However, in the experiment, it was shown in the experiment that the detection is not one time but only several times during space movement. In this case, it is necessary to distinguish whether it is a simple overlapping detection or a repetitive movement (ie, moving back and forth). Time information was used to distinguish this. In other words, we developed and applied a filter to make the same sensor-occurrence data existing in this window occur once at the first occurrence time using the 2-minute time interval window in the sense data sequence.

In addition, when analyzing the actual experiment (elderly housing experiment) data which will be described separately separately, there was a case where an irrelevant motion sensor installed in the middle of the movement was detected. This is because the sensing areas of other sensors overlap in the space of the movement path. This detection occurs with a very low probability, so it is not necessary to consider the position tracking of the present invention, but to remove it, the occurrence of another sensor signal in the middle or very close time of the same sensor data sequence is regarded as a false detection and eliminated. Was developed and applied.

As such, when the measurement data is transmitted from the indoor monitoring module 200 to the server computer 300 through the step S100, the server computer 300 is based on the measurement data transmitted from the indoor monitoring module 200 through the network communication network 100. Create a single person's home location and time of stay, time to go out, sleep time, and based on the home alone location and time of stay, time to go out, time to sleep, time spent in each room, toilet and kitchen, outing behavior, A fuzzy set for each characteristic component including sleep behaviors is generated, and a life pattern of a 24-hour period of the living alone is modeled as a logical product of the fuzzy sets for each characteristic component (S200).

Referring to FIG. 5, the server computer 300 determines whether measurement data is received from the indoor monitoring module 200 (S210), and when the measurement data is received from the indoor monitoring module 200. , Based on the measurement data, the resident's home location and dwelling time, outing time, sleep time is generated, and the staying in each room, toilet and kitchen based on the resident's home location and dwelling time, outing time, sleep time The feature value for each feature component including time, outing behavior, and sleep behavior is stored for a plurality of preset days (S220).

The server computer 300 generates a fuzzy set that reflects time repeatability based on feature values for a plurality of days for each feature component including residence time, outing behavior, and sleep behavior in each room, bathroom, and kitchen ( S230), by performing a logical multiplication of the fuzzy sets reflecting the time repeatability for the plurality of days for each feature component generated (S240) is modeled as a life pattern of the living alone 24 hours.

Step S200 is a method of modeling daily life repeated in a fuzzy set using sensor data collected by the home monitoring module 200 in the server computer 300 as a fuzzy set. Therefore, it can be considered that the more repetition of the daily actions in the same time zone, the higher the representativeness of the pattern is. In Virone's study of prior art non-patent literatures 1 and 2, to find such repeatability, the residence time in each room was calculated for one hour per hour on a daily basis, and the average of these values over several days. A simple and relatively repeatable pattern representation method (life pattern model) was proposed. This conventional method is a representation of the repetition of a person's daily activities as a time and time of stay in a residential space (Circadian Activity Rhythm; CAR).

The life pattern modeling method of the present invention includes not only residence time in each room but also daily activities such as going out and sleeping. As such, staying, going out, and sleeping in each room are regarded as the constituents that make up the pattern of daily activities. If the time information of these constituents is the feature value, the information of when and when they went out when they went out It becomes the characteristic value of a characteristic component.

First, the pattern model proposed by Virone of the non-patent literatures 1 and 2, which is a prior art, assumes that an elderly person goes out at around 8:12 am and enters at 10:22. The method of constructing is to check the presence / absence of outing behavior at the interval of 0 hour and 1 hour, and calculate how much time it takes if there is. In this case, the value is 48 minutes in the 8-9 o'clock section, 60 minutes in the 9-10 hour section, and 22 minutes in the 10-11 hour section. Therefore, the feature vector is represented by a vector having 24 components, as shown in the following equation.

x _i = [0, ..., 0,48,60,22,0, ..., 0]

Where i is It is an index indicating a characteristic component (outing).

In this way, a feature vector is obtained for all feature components, and all components are collected to form a CAR matrix for a specific day. The CAR matrix, which is obtained in units of 24 hours (day), is obtained over several days, and the mean value matrix and the standard deviation matrix of these values become the life pattern model of an elderly person. In fact, if you stay in the same place at the same time zone, the average stay time in that time zone increases.

In the present invention, a modeling method based on fuzzy theory is proposed, and a model is generated by fuzzy sets for each feature. Using a simple example, FIG. 7 is a graph of feature values for outing features obtained from actual data for five days of a particular elderly. The small sized graph with the date shown in the upper part of FIG. 7 is a graph (characteristic value) indicating when the person went out (marked with 1) for 24 hours from 00:00. From the graph, you can easily see when and how many times you went out. In this way, the fuzzy set for the outgoing component is calculated using the feature data of 5 days. The calculation method is to overlap all five graphs. However, for each overlap, add 1/5 (1 / n with n-day data). In other words, every five days out of the interval has a value of 1, once only 1/5 value. In this case, the more repeatability of the time interval that goes out, the closer to 1. In conclusion, the calculated result graph is defined as a fuzzy set with elements ranging from 00:00 to 23:59. Here, the degree of belonging of a particular time indicates the repeatability of the feature component. Therefore, if the same daily activities are repeated at the same time, the degree of belonging value is close to one. The fuzzy set graph of the outing component of FIG. 7 shows that the behavior of going out between 10 and 11 o'clock and coming home beyond 12 o'clock is very regular and repetitive. Thus, one fuzzy set is made based on several days of data for individual feature components.

That is, the fuzzy set based on the feature values for the plurality of days for each feature component generated in step S230 is generated by the following equation.

Where p is an index for identifying a single person, i is an index representing a characteristic component, and all fuzzy sets are defined for a set U representing a 24-hour range from 00:00 to 23:59 in minutes.

Then, fuzzy sets are calculated by using feature values of several days for individual feature components, and the living pattern of the living person is modeled by logical AND of these fuzzy sets.

In other words, the life pattern modeling of the 24-hour period generated in step S240 is generated by the Dium equation.

Where Nc is the number of feature components.

Now, after generating a fuzzy set for each feature through the step S200, and modeling the living pattern of the living alone 24 hours period by the logical product of the fuzzy set for each feature, the server computer 300 modeling the life pattern Based on the information, the degree of abnormality, which is a measure of the difference between the living pattern and the daily life pattern, is detected, and the abnormality is detected by comparing the calculated abnormality with a preset threshold value (S300). ).

Referring to FIG. 6, the server computer 300 determines a degree of normality for a specific day of each feature component of the living person based on the life pattern modeling information generated through the step S200. Check (S310), and normalizes the determined normality value by dividing the maximum value by 1440, which indicates the maximum value of 24 hours in minutes (S320).

Subsequently, the server computer 300 uses the normalized value and the standard deviation of the normality values of the specific days used when generating the fuzzy set for each characteristic component of the living person, and the standard deviation of the living alone in the step S320. The degree of abnormality which is a measure of the difference between the life pattern and the usual life pattern is calculated (S330).

The abnormality value calculated in step S330 is compared with a threshold value that is a criterion for occurrence of abnormality (S340), and it is determined whether the abnormality value exceeds the threshold as a result of the comparison (S350). In this case, the threshold value used as a reference for occurrence of abnormality in step S340 is preferably determined in advance based on the accumulated experience of the experiment.

If the abnormality value exceeds the threshold as a result of the determination in step S340, the server computer 300 detects that an abnormality has occurred to the living alone and performs step S400 (S360).

In step S300, it is determined whether a pattern different from the normal pattern from which the daily living pattern of the resident is derived is detected, and how to find the abnormal detection from the normal pattern.

First, Virone's method of the prior art non-patent literature 1 calculates the degree of abnormality by comparing the derived model (mean and standard deviation CAR matrix) with a specific daily CAR matrix. The value of each element of a specific CAR matrix is compared with the range of the following equation to distinguish between two small and large sides. This range is shown in the following equation.

Small variation range: [m-μ ₁ σ, m + μ ₁ σ]

Large variation range: [m-μ ₂ σ, m + μ ₂ σ]

Where m is the mean, sigma is the standard deviation, and μ ₁ , μ ₂ is the weight (0 <μ ₁ <μ ₂ ). When each element of the CAR matrix falls within this range, four kinds of variations (less than / over and large / small) are recorded and the number of all variations that occur during the day is summed to calculate the outlier.

The method of the present invention calculates the correspondence between the model constructed using the general two-way intersection computation method and the specific daily activities. Using the model for the outing of a particular elderly person of FIG. 7, if a certain elderly person goes out from 9:36 to 12:19 on a certain day (k), a fuzzy set of outing features of the day may be calculated. The model derived for a particular day is the non-fuge set (1 for outing time, 0 otherwise). The area of the fuzzy set obtained by performing the intersection operation of these two fuzzy sets shows the result reflecting how the characteristic values (time and time) of the outgoing components of a particular day match the accumulated model. In the same way, the sum of the area of the resultant fuzzy set obtained by performing the intersection operation and the result of a specific day for all the characteristics of the elderly (stay time, go-out, sleep behavior, etc.) reflects the normality of the specific day. do. This value is defined as normal.

That is, the normality of the specific day generated in step S310 is generated by the following equation.

Where M ^p _i is a fuzzy set constructed using a plurality of day data of feature _i , and C ^p _i is the feature of i It is a non-fuge set, and I ^p _N (k) is the sum of the area A ^p _i of the intersection of the two fuzzy sets.

In order to normalize this normal value, the maximum value should be considered. This maximum value is 1440 if the elderly are to stay in one space for 24 hours, so the normality value is the sum of 1 every 1 minute for 24 hours a day. Therefore, as shown in the following equation, the normality value calculated in step S310 is divided by 1440 and normalized.

r ^p (k) = I ^p _N (k) / 1440

This value can be used to define anomalies. That is, the degree of abnormality generated in step S330 is as follows.

,

,

Where I ^p _A (k) is the degree of abnormality and m ^p _r and δ ^p _r are the multiple days (N _FS) used to compute the fuzzy set M ^p _i for the feature component i Is the mean and standard deviation of the normal values.

As described above, the abnormality is not simply defined as the inverse of the normality value, but is calculated as a ratio with the standard deviation of the difference from the average of the normality values of the days used when calculating the model. The reason is that the absolute value of normality varies greatly even though it is normalized according to the life patterns of the elderly. Therefore, in order to apply a simple technique of setting a specific threshold and reporting an abnormality when the threshold is exceeded, the mathematical equation of the degree of abnormality is applied. This will be described in more detail in the following actual experiment results.

Lastly, when the abnormality of the sole person is detected as a result of the process of step S300, the server computer 300 transmits data on the occurrence of the abnormality of the monitoring subject alone to the registered provider terminal 400 as text or voice data. To perform a service taking an appropriate action (S400).

Next, with reference to Figures 8 to 10 will be described for the actual experimental results for the elderly by applying the method of the present invention.

FIG. 8 is a view showing an example of a floor plan of a particular elderly person and an installation environment of home monitoring equipment to which the present invention is applied. FIG. 9 is a view showing an example of a fuzzy set of three elderly people of FIG. 8. FIG. 8 is a diagram illustrating an example of models of six feature components of the elderly A in FIG. 8.

In order to verify the performance of the method of the present invention, an in-house monitoring module was installed in a house where three people (two males (A, B), one female (C) and an average age of 76 years) were installed and a long-term operation experiment was performed. It was. The operating period of the system is 13 months for A old man, about 2 months for B old man and 3 months for C old man. First of all, the three senior citizens' houses are apartments. In particular, B and C elderly people live in the same apartments. Accordingly, two kinds of apartment floor plans are shown in FIGS. 8A and 8B. In the case of FIG. 8A, the house is composed of three rooms, a living room and a kitchen, and a toilet. In FIG. 8B, the house is composed of two rooms (a bedroom, a small room), a toilet, a hall, and a kitchen. However, one room was excluded from the experiment because it was rarely used.

All three houses had six motion sensors, one door sensor, and a sleep sensor. The blue and red ranges shown in the plan view show the approximate detection area of the installed motion sensor. The motion sensor shown in ellipse is installed as a motion sensor for spatial division, and the motion sensor displayed in triangle shape is a verification sensor for verifying position tracking results. Since it is installed sideways as possible, as shown in the drawing has a wider detection range.

Location tracking is performed using sensor data for each elderly person. The location tracking performance was checked for accuracy by using in-lab verification experiments and verification motion sensors installed in actual elderly houses. The test result detected the position movement without error. A fuzzy set type model is created using the feature values extracted here. All three subjects have common features. There are 6 ingredients in the living room, small room, restroom and kitchen. First, fuzzy sets (M ^p _i ) for each feature were constructed using 15 days of data for all three patients. These fuzzy sets were created using data of N _fs = 15 days.

First, the fuzzy set for the outing behavior is shown in FIG. 9. What can be seen in Figure 9 is the degree of repetition for the outing behavior. C elderly people (solid line) shows a large degree of belonging during lunch time (11 o'clock to 13:00), so it can be seen that going out during this time is very repetitive and regular. On the other hand, the elderly A (dashed line) has a small degree of belonging at lunch and dinner time, so it shows that they rarely go out. B elderly people also have relatively regular outings during lunch and dinner. Based on the proposed method, it is easy to grasp the repeatability of certain daily activities (main stay space).

FIG. 10 is a diagram showing a fuzzy set for each characteristic component of an elderly A. The graph of FIG. 10 shows a fuzzy set for three components of residence time (solid line), restroom (dotted line), and sleep activity (dashed line). The graph below shows the residence time purge set of the remaining three components: small room (dotted line), kitchen (solid line), and toilet (dashed line).

The fuzzy set of sleep activity shows that sleep activity of A-A is very regular and repetitive. They show that they sleep a little late and get up late, and there is a sleep interruption due to the use of the toilet around 3 am. In addition, when looking at the fuzzy set of the four rooms, it can be seen that the elderly A spends most of the time in the room from the fuzzy set. On the other hand, small rooms rarely stayed, and restrooms were used equally throughout the day. Finally, the kitchen was observed to stay long in the evening.

Now, the normality and abnormality values are calculated using the monitoring results in real life for the elderly A and the elderly C, and the possibility of abnormality detection using the same will be described.

To calculate outlier values, the normality values of the individual days used to create the fuzzy set for each feature must be calculated, and the mean and standard deviation of these values calculated.

The results are shown in Tables 1 and 2 below.

mean (A ^p _i ) stdev (A ^p _i ) percentage Inner room 696.1 39.5 83.7% Small room 3.3 4.7 0.4% restroom 3.1 2.6 0.4% Kitchen 122.5 51.2 14.7% outing 6.6 7.7 0.8% I ^p _N (k) 831.6 37.0 100% r ^p (k) 0.577 0.02571

mean (A ^p _i ) stdev (A ^p _i ) percentage Inner room 430.5 63.2 67.4% Small room 0.1 0.4 0.0% restroom 7.9 2.7 1.2% Kitchen 92.6 50.5 14.5% outing 107.9 46.0 16.9% I ^p _N (k) 639.0 45.4 100% r ^p (k) 0.444 0.0315

Table 1 and Table 2 show the mean and standard deviation values of the normality values calculated from the 15-day data of A and C elderly, respectively. Not only the values of normality (I ^p _N (k) and r ^p (k)) but also the mean and standard deviation values of the area values (A ^p _i ) for each of the five components constituting these values are also shown. The reason for displaying the value and ratio ('percentage' column) for each component is to know what is the main characteristic component that determines the value of normality. The results in Tables 1 and 2 show that this ratio varies among the elderly and that it is easy to identify which constituents are important in life patterns. Experimental results show that A and C elderly people rarely use small rooms. In particular, the elderly in C rarely use a small room, so it is not considered to have a significant effect even if this feature is excluded from the entire model. As such, the method of the present invention has the advantage of very clearly indicating which features are important in everyday life. Lastly, the absolute size of the normal value shows that there is a big difference according to the living pattern of the elderly (A senior = 831, C senior = 639).

Now look at the anomaly results for a particular day using these values. Table 3 below shows the abnormality values for specific 7 days of elderly A. If the threshold is set to 2, the abnormal day is 4 out of 7 days (20, 21, 23, 25 days). In the case of 25 days, the highest value was due to the relatively short stay in the room and the relatively long time spent in the kitchen, unlike the life pattern of the old man who spends most of the time in the room. The 20th was also judged abnormal for the same reason. The 21st and 23rd days seem to be due to the difference in the model of residence time in the kitchen.

19th 20 days 21st 22nd 23rd 24th 25th Inner room 689 576 633 661 719 802 534 Small room 0 5.7 0 1.7 0.2 0 0 restroom 0.5 0.5 0.7 0.2 0 2.3 0 Kitchen 116 155 98.3 103 35.9 45.9 183 outing 7.3 0 0 0.6 0 1.3 0 I ^p _N (k) 813 737 732 767 755 852 717 r ^p (k) 0.57 0.51 0.51 0.53 0.53 0.59 0.50 I ^p _A (k) 0.50 2.54 2.68 1.73 2.05 0.56 3.07

Even for C elderly, the abnormality values were calculated for the same 7 days and are shown in Table 4. The results show that there are more than three days (21, 23, 24) exceeds the threshold. First, in case of 21 days, the value of normality is much smaller than the average value, because it is not going out. In the case of the 23rd day, it was judged to be due to the stay in the kitchen which was much higher than the residence time in the home room, and in the case of the 24th day, the residence time in the kitchen was particularly short.

19th 20 days 21st 22nd 23rd 24th 25th Inner room 442 318 506 469 309 405 369 Small room 2.2 12.6 0 2.6 2.4 3.4 5 restroom 0 0 0.7 0 0 0 0 Kitchen 93 119 98.3 25 108 22 36 outing 127 105 0 97.7 122 112 154 I ^p _N (k) 665 556 732 594 541 543 565 r ^p (k) 0.46 0.39 0.51 0.41 0.38 0.38 0.39 I ^p _A (k) 0.59 1.82 2.68 0.97 2.15 2.10 1.62

As shown in the above, it was possible to find a day in which the activity was performed differently from the comparison with a simple threshold value by using the method of calculating an abnormality value using the fuzzy set model of the present invention. However, further research will be needed on how to determine common thresholds or whether elderly thresholds should be introduced.

Herein, while the present invention has been described with reference to the preferred embodiments, those skilled in the art will variously modify the present invention without departing from the spirit and scope of the invention as set forth in the claims below. And can be changed.

100: network communication network 200: home monitoring module
210: power supply unit 220: sensor unit
230: control unit 240: data communication unit
300: server computer 310: data communication unit
320: control unit 330: life pattern modeling unit
360: database 400: provider provider terminal

Claims (14)

(1) in the indoor monitoring module, transmitting measurement data detected through one or more sensors installed in each room, toilet, entrance, and kitchen of the home to a server computer through a network communication network;
(2) The server computer generates the home location and dwell time, the outgoing time, and the sleep time of the living alone, based on the measurement data transmitted from the home monitoring module through the network communication network, Based on residence time, outing time and sleep time, create fuzzy sets for each feature including stay time, outing and sleep behavior in each room, toilet, and kitchen. Modeling life patterns of the phosphorus's 24-hour cycle, and
(3) The server computer calculates a degree of anomaly, which is a measure of the difference between the living pattern of a single person and the usual living pattern, based on the living pattern modeling information generated in step (2). And detecting whether an abnormality occurs by comparing the calculated abnormality with a preset threshold.
Living pattern modeling and anomaly detection method including the living alone. The method of claim 1,
(4) The server computer transmits, as text or voice data, data on the occurrence of an abnormality of the monitoring target alone to the registered provider terminal when the abnormality of the sole is detected as a result of the processing in step (3). Life pattern modeling and abnormal detection method, which further comprises the steps. The method of claim 1,
The step (2)
(2-1) the server computer determining whether measurement data is received from the indoor monitoring module;
(2-2) When the measurement data is received from the indoor monitoring module as a result of the determination in the step (2-1), the server computer generates the home location, dwell time, outing time, and sleep time of the living person based on the measured data. And a plurality of days for setting characteristic values for each characteristic component, including residence time in each room, toilet, and kitchen, outing behavior, and sleeping behavior, based on the home location, dwelling time, outing time, and sleeping time of the living alone. Saving each step,
(2-3) The server computer calculates feature values for a plurality of days for each feature component, including residence time, outing behavior, and sleep behavior in each of the rooms, toilets, and kitchens stored in step (2-2). Generating a fuzzy set based on time repeatability, and
(2-4) The server computer performs a logical product of fuzzy sets reflecting the time repeatability for a plurality of days for each feature component generated in step (2-3), thereby living the 24-hour period of the living alone. Modeling with patterns
Living pattern modeling and anomaly detection method including the living alone. The method of claim 3, wherein
The fuzzy set based on feature values for a plurality of days for each feature component generated in step (2-3),

(Where p is the index identifying the individual, i is the index representing the characteristic component, and all fuzzy sets are defined for a set U representing the 24-hour range from 00:00 to 23:59 in minutes) Life pattern modeling and anomaly detection method, which are generated by living alone. The method of claim 3, wherein
The life pattern modeling of the 24-hour period generated in the step (2-4),

Where Nc is the number of feature components. The method of claim 1,
Step (3),
(3-1) checking, by the server computer, a degree of normality for a specific day of each feature component of the living person based on the life pattern modeling information generated through step (2);
(3-2) the server computer normalizing the normality value identified in step (3-1) by dividing the normality value by 1440 indicating the maximum value of 24 hours in minutes;
(3-3) The server computer uses the value normalized in the step (3-2), the average and the standard deviation of the normality values of the specific days used to generate the fuzzy set for each feature component of the individual. Calculating the degree of abnormality, which is a measure of the difference between the living pattern of a single person and the daily living pattern,
(3-4) the server computer comparing the abnormality value calculated in the step (3-3) with a threshold value that is a criterion for occurrence of abnormality and determining whether the abnormality value exceeds the threshold value, and
(3-5) If the abnormality value exceeds the threshold as a result of the determination in step (3-4), the server computer detects that an abnormality has occurred in the living alone.
Living pattern modeling and anomaly detection method including the living alone. The method according to claim 6,
The degree of normality for a specific day generated in the step (3-1),

Where M ^p _i is a fuzzy set constructed using a plurality of day data of feature _i , and C ^p _i is the feature of i The non-fuzzy set, I ^p _N (k), is the sum of the area (A ^p _i ) of the intersections of the two fuzzy sets. Detection method. The method according to claim 6,
The abnormality generated in the step (3-3),

,

(Where I ^p _A (k) is at least approximately, m ^p _r and δ _r ^p is a plurality of days was used to calculate the fuzzy set M ^p _i of the feature component i (N _FS Life pattern modeling and anomaly detection method, which is generated by the equation of the mean and standard deviation of The method according to claim 6,
The threshold value used as a reference for occurrence of abnormality of step (3-4) is
Life pattern modeling and anomaly detection method, which are determined by living alone based on accumulated experiments. A computer-readable recording medium having recorded thereon a computer program capable of executing the method of any one of claims 1 to 9. Network communication network that connects a plurality of unspecified communication lines so that data communication related to life pattern modeling and anomaly detection of a single person can be performed.
In-house monitoring module, which transmits the measurement data detected by one or more sensors installed in each room, toilet, entrance, and kitchen of the house to the server computer through the network.
Based on the measurement data transmitted from the home monitoring module through the network communication network, the home location and dwelling time, outing time, sleep time of the living alone is generated, Based on this, we generate fuzzy sets for each feature, including residence time, outing behavior, and sleep behavior in each room, toilet, and kitchen. Modeling, calculate the abnormality level, which is a measure of the difference between the living pattern and the daily life pattern of the living person, based on the life pattern modeling information, and compares the calculated abnormality level with the preset threshold. To detect whether a server computer
Life pattern modeling and abnormal detection device including a single. The method of claim 11,
Single life pattern modeling and abnormal detection device further comprises a provider terminal for receiving data about the abnormal occurrence of the monitoring targets alone from the server computer through the network communication network as text or voice data. The method of claim 11,
The indoor monitoring module,
Power supply unit for supplying power for the operation of the device,
It is installed in each room, toilet, entrance, kitchen of the house, and the sensor unit which measures the location, the outing, and the sleep of a single person,
A control unit which controls to transmit the measurement data measured by the sensor unit to the server computer; and
A data communication unit for transmitting the measurement data measured by the sensor unit to the server computer under the control of the controller;
Life pattern modeling and abnormal detection device including a single. The method of claim 11,
The server computer,
A data communication unit receiving measurement data from the indoor monitoring module and transmitting abnormality data to the provider terminal;
Based on the measurement data received from the indoor monitoring module through the data communication unit, the generation and storage of a resident's home location, dwelling time, outing time, and sleeping time, and based on feature values of a plurality of days for each feature component The criteria of generation of fuzzy sets and life pattern modeling reflecting time repeatability, abnormality calculation and abnormality value and occurrence of abnormality, which are the measure of the difference between the daily life pattern and the daily life pattern, A control unit for controlling an abnormality detection through comparison of a threshold value, and a notification of a caller's terminal of data related to the abnormality detection;
A life pattern modeling unit for generating a fuzzy set reflecting time repeatability based on feature values for a plurality of days for each feature component, and modeling a living pattern of a 24-hour period of a living alone by performing a logical product of the fuzzy sets;
Based on the life pattern modeling information generated by the life pattern modeling unit, the normality of a specific day of each characteristic component of the individual is checked, and the normalized value is normalized, and then each normalized value and each characteristic of the individual Using the mean and standard deviation of the normality values of the specific days used to generate the fuzzy set for each component, the abnormality value, which is a measure of the difference between the living pattern and the daily life pattern, is calculated. An abnormality detection unit for detecting that an abnormality has occurred in a single living person when the abnormality value exceeds the threshold value as compared with a threshold value that is a criterion for occurrence of abnormality;
An abnormality occurrence processing unit for transmitting the data related to the abnormality detection to the registered provider terminal as text or voice data when the abnormality of the relevant individual is detected through the abnormality detection unit, and
Stores the information on the provider of monitoring the living alone, stored for each feature component of the measurement data received from the home monitoring module, storage of the life pattern modeling information of each living alone generated through the life pattern modeling unit, the abnormality A database for processing storage of abnormality detection information of the detection unit and abnormality detection data transmitted to the provider terminal.
Life pattern modeling and abnormal detection device including a single.

Images