Sleep determination system, sleep determination method, and learning model generating method

Abstract

To correctly grasp a sleeping time of an object person by properly determining a sleeping state.SOLUTION: A sleep determination system comprises: a plurality of sensors which is disposed at a plurality of places including a bed room and another place within a residence of an object person; a storage part which stores a learning model having been learned by collecting time-series detection data of the plurality of sensors and time-series sleep data on whether or not the object person is in a sleeping state; and a determination part which acquires the detection data of the plurality of sensors and determines the sleeping state of the object person using the acquired detection data and the learning model.SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] This specification discloses a sleep determination system, a sleep determination method, and a learning model generation method. [Background technology]

[0002] Conventionally, a system has been proposed in which a plurality of sensors are placed in various places in the subject's residence, including the bedroom, and the subject's sleep state is determined based on the detection results of the sensors. For example, in the system of Patent Document 1, if the subject remains in the bedroom until a predetermined sleep confirmation time has elapsed after it is detected that the subject has entered the bedroom, it is determined that the subject is in a sleep state, and if the subject does not return to the bedroom even after a predetermined wake-up confirmation time has elapsed after it is detected that the subject has left the bedroom, it is determined that the sleep state has ended and the subject has woken up. For this reason, if the time when the subject leaves the bedroom is less than the wake-up confirmation time, such as when the subject wakes up to go to the bathroom while sleeping, it is not determined that the subject has woken up. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2003-102704 A Summary of the Invention [Problem to be solved by the invention]

[0004] In the above-mentioned system, if the subject gets up to go to the toilet, the subject does not recognize the subject as awake and continues to sleep, but some subjects may go to the toilet frequently. In such cases, the subject continues to sleep even if the sleep state is frequently interrupted by going to the toilet, so the measured sleep time and the actual sleep time will differ, making it difficult to correctly grasp the sleep time.

[0005] A primary objective of the present disclosure is to correctly grasp a subject's sleeping time by appropriately determining the sleep state. [Means for solving the problem]

[0006] In order to achieve the above-mentioned main object, the present disclosure has adopted the following means.

[0007] The first sleep determination system of the present disclosure includes: A plurality of sensors are disposed at a plurality of locations including a bedroom and a non-bedroom in a residence of a subject; A memory unit that stores a learning model that is learned by collecting time-series detection data of the plurality of sensors and time-series sleep data indicating whether the subject is in a sleeping state; a determination unit that acquires detection data from the plurality of sensors and determines a sleep state of a subject using the acquired detection data and the learning model; The gist of the invention is to provide the following:

[0008] In the first sleep determination system of the present disclosure, detection data from a plurality of sensors is acquired, and the sleep state of a subject is determined based on the acquired detection data and a learning model. The learning model is learned by collecting time-series detection data from a plurality of sensors and time-series sleep data on whether the subject is in a sleeping state or not. By using such a learning model, the sleep state can be appropriately determined and the sleeping time of the subject can be correctly understood. [Brief description of the drawings]

[0009] [Figure 1] FIG. 2 is an explanatory diagram showing an example of sensors 30 arranged in a subject's residence 1. [Diagram 2] FIG. 1 is a diagram showing an outline of the configuration of a sleep determination system 10. [Diagram 3] 11 is a flowchart showing an example of a sleep state learning process. [Figure 4] FIG. 4 is an explanatory diagram showing an example of sleep data and detection data for each sensor. [Diagram 5] 5 is a flowchart showing an example of a sleep state determination process according to the first embodiment. [Figure 6] FIG. 4 is an explanatory diagram showing an example of detection data. [Figure 7] FIG. 4 is an explanatory diagram showing an example of a sleep determination result. [Figure 8] 10 is a flowchart showing an example of a sleep state determination process according to the second embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] [First embodiment] A first embodiment of the present disclosure will be described with reference to the drawings. FIG. 1 is an explanatory diagram showing an example of sensors 30 arranged in a residence 1 of a subject. FIG. 2 is a configuration diagram showing an outline of the configuration of a sleep determination system 10. The residence 1 is the residence of a subject to be watched or monitored, and is equipped with sensors 30 used to determine the subject's sleep state. The subject may be, for example, an elderly person living alone, but may also be a person receiving medical treatment or a person requiring care. The residence 1 may be a residence with a living room (dining room), kitchen, bedroom, etc., partitioned as shown in FIG. 1, or may be a one-room type residence.

[0011] The sleep determination system 10 includes a monitoring device 20, sensors 30, and a management server 40. The monitoring device 20 includes a control unit 22 and a communication unit 24, and receives detection data and the like from the sensors 30. The control unit 22 is configured as a microprocessor centered around a CPU, and includes a ROM, a RAM, and the like in addition to the CPU. The communication unit 24 is capable of communicating with the management server 40 via a network NW such as the Internet. When detection data and the like from the sensors 30 are input to the control unit 22, the control unit 22 transmits necessary data from the communication unit 24 to the management server 40 via the network NW.

[0012] As shown in Fig. 2, the sensors 30 include motion sensors 31, 32, and 33 arranged in each room, and a door sensor 35 provided at the front door. The motion sensors 31 to 33 are sensors that detect people within a detection area in a non-contact manner, and are configured, for example, as infrared sensors that detect infrared rays and convert them into an electrical signal. In this embodiment, the motion sensor 31 is arranged in the bedroom, the motion sensor 32 is arranged in the living room (dining room), and the motion sensor 33 is arranged in the toilet. The door sensor 35 detects whether the front door is opened or closed, and is configured, for example, as a magnetic opening / closing sensor having a permanent magnet fixed to the door side and a magnetic sensor fixed to the frame side.

[0013] The management server 40 includes a control unit 42, a storage unit 44, and a communication unit 46, and manages the entire system. The control unit 42 is configured as a microprocessor with a CPU as its core, and includes a ROM, a RAM, and the like in addition to the CPU. The storage unit 44 is configured with an HDD, an SSD, and the like, and stores data transmitted from the monitoring device 20. The storage unit 44 also stores a learning model 45 used to determine the sleep of the subject. The communication unit 46 is connected to the communication unit 24 of the monitoring device 20 via the network NW, and exchanges information with the communication unit 24 of the monitoring device 20. The management server 40 may be communicatively connected to the communication units 24 of multiple monitoring devices 20 arranged in multiple residences, and the sleep state of each subject in the multiple residences may be determined based on the data transmitted from each monitoring device 20. In addition, the administrator K of the subject can access the management server 40 via the network NW from his / her own mobile terminal M, personal computer, or the like to check the sleep state of the subject.

[0014] In the following, a process of generating the learning model 45 and storing it in the storage unit 44 and a process of performing sleep judgment of the subject will be described. FIG. 3 is a flowchart showing an example of the sleep state learning process. This process is executed by, for example, the control unit 42 of the management server 40 when a predetermined learning period for acquiring data necessary for the learning process is completed. The sleep state learning process may be executed by a computer other than the management server 40, and the generated learning model 45 may be stored in the storage unit 44. The predetermined learning period is a period of several months (for example, 3 months) or more. In this embodiment, during the learning period, data is acquired in a learning environment in which a specific sensor 50 having a higher detection accuracy than the human sensor 31 (32, 33) is temporarily placed in the bedroom in addition to the sensors 30. The specific sensor 50 is, for example, a millimeter wave radar sensor, but is not particularly limited, and may be a mat sensor or the like. In FIG. 2, the data of the specific sensor 50 is input to the monitoring device 20, but may be transmitted to the management server 40 via the network NW. The specific sensor 50 is removed from the bedroom when the learning period is completed. That is, in the subject's normal living environment, the sensors 30 are placed, but the specific sensor 50 is not placed.

[0015] In the sleep state learning process, time-series detection data from each human sensor (human sensors 31, 32, 33) during the learning period and time-series sleep data from the specific sensor 50 are acquired (S100). The specific sensor 50 has high detection accuracy, so it can acquire sleep data indicating the subject's bedtime and wake-up time with relatively high accuracy. The detection data (sleep data) from the specific sensor 50 is acquired as correct answer data when learning the sleep state.

[0016] Next, the control unit 22 exclusively extracts learning data to be used for learning and evaluation data to be used for evaluating the learning model from the detection data of each human sensor acquired in S100 (S110). For example, the control unit 22 randomly extracts half of the detection data of the sensors 30 as learning data regardless of the time series, and extracts the remaining half as evaluation data so as not to overlap with the learning data. Through the processing of S100 and S110, the control unit 22 acquires a plurality of data in which the time series sleep data (going to bed, getting up) of the specific sensor 50 is regarded as correct answer data and the time series detection data of each human sensor is regarded as learning data, as shown in FIG. 4. Each data includes the date (** / ** / **) and the hour, minute, and second (**:**:**), but this is only an example, and the data may include the hour, minute, and second instead of the hour, minute, and second. Next, the control unit 22 extracts the number of detections per unit time (e.g., per minute) for each human presence sensor from the learning data as features (S120), and generates a learning model by supervised learning (machine learning) using the extracted features and the time-series sleep data of a specific sensor 50 (S130).

[0017] After generating the learning model in this way, the control unit 22 evaluates the learning model using the evaluation data extracted in S110 (S140). In S140, the sleep state determination as a result of inputting the evaluation data into the learning model is compared with the detection data of the specific sensor 50 in the same time period, and the correctness of the determination can be confirmed to evaluate the reliability of the learning model. The determination of the sleep state will be described later, but if the desired reliability is not obtained, the control unit 22 may generate the learning model again. When the desired reliability is obtained, the control unit 22 stores the generated learning model in the storage unit 44 (S150) and ends this process. As a result, the learning model 45 is stored in the storage unit 44.

[0018] Next, a process for determining whether a subject is asleep will be described. Fig. 5 is a flow chart showing an example of the sleep state determination process of the first embodiment, which is executed at regular intervals, for example, once a day. In the sleep state determination process, the control unit 22 acquires detection data of the sensors 30 for a predetermined determination period (for example, 24 hours) after the previous sleep state determination process is executed (S200). Next, the control unit 22 sets the first predetermined time of the acquired detection data as the section to be determined (S210), and extracts the number of detections for each sensor in the section to be determined and the sections before and after it as a feature (S220).

[0019] Then, the control unit 22 performs model-based determination to determine the sleeping state using the extracted feature amount of each section and the learning model 45 (S230). In this embodiment, for example, the section to be determined is 10 minutes, and the sections before and after it are each 30 minutes, and the number of detections for each sensor in each section is counted per unit time (for example, per minute) and extracted as a feature amount. Note that 10 minutes and 30 minutes are examples, and the section to be determined may be an integer multiple of the unit time. Then, the control unit 22 compares the pattern of the extracted feature amount with the pattern of the feature amount of the sleeping state in the learning model to determine whether or not the user is in a sleeping state. Note that, when one of the sections before and after the section to be determined does not exist, such as when a predetermined time at the beginning of the detection data is the section to be determined, the determination may be made based on the pattern of the feature amount in the section to be determined and the other section before and after the section.

[0020] When the model-based determination is performed in this manner, the control unit 22 determines whether the determination result is sleep or not (S240). When the control unit 22 determines that the subject is asleep, it performs a rule-based determination to determine the sleep state based on the detection data of the sensors 30 in the section to be determined (S250), and determines whether the determination result is sleep or not (S260). In the rule-based determination of this embodiment, for example, when the subject is at home based on the detection response of the door sensor 35, there is no detection response of the human presence sensors 32 and 33 arranged other than the bedroom, and there is a continuous detection response of the human presence sensor 31 arranged in the bedroom, the subject is determined to be asleep. Therefore, the control unit 22 determines that the subject is not asleep when there is a detection response of either of the human presence sensors 32 and 33, or when there is no continuous detection response of the human presence sensor 31.

[0021] When the control unit 22 determines in S260 that the judgment result is sleep, that is, when both the model-based judgment and the rule-based judgment are sleep, it estimates that the judgment target section is sleep (sleep section) (S270). On the other hand, when the control unit 22 determines in S240 that the judgment result of the model-based judgment is not sleep, or when the control unit 22 determines in S260 that the judgment result of the rule-based judgment is not sleep, it estimates that the judgment target section is not sleep but non-sleep (non-sleep section) (S280). Then, the control unit 22 determines whether there is unprocessed data (S290), and when it determines that there is unprocessed data, it sets the next predetermined time to the judgment target section (S295), proceeds to S220, and repeats the process. On the other hand, when the control unit 22 determines in S290 that there is no unprocessed data, it ends this process.

[0022] Here, FIG. 6 is an explanatory diagram showing an example of the detection data. FIG. 7 is an explanatory diagram showing an example of the sleep determination result, and the range estimated as sleep in the sleep state determination process is shown in a dotted frame. Also, FIG. 6 and FIG. 7 exemplify one day's worth of detection data of the sensors 30. In the above-mentioned sleep state determination process, model-based determination and rule-based determination are performed, so even if the model-based determination determines that the person is asleep, the rule-based determination may not determine that the person is asleep. For example, in the example of FIG. 6 and FIG. 7, since there is a detection reaction of the toilet (human sensor 33) at time t1 in the 3 o'clock hour, even if the model-based determination determines that the person is asleep, the rule-based determination does not determine that the person is asleep. On the other hand, before and after time t1, there is no detection reaction of the living room or toilet (human sensors 32, 33) and there is a continuous detection reaction of the bedroom (human sensor 31), so the rule-based determination also determines that the person is asleep. For this reason, the time period before and after time t1, except for time t1, is estimated as sleep. Similarly, since there is a detection response in the living room at time t2 in the 8 o'clock hour and a detection response in the toilet at time t5 in the 17 o'clock hour, the range excluding times t2 and t5 is presumed to be sleeping. Also, since there is a detection response at the front door (door sensor 35) at times t3 and t4, for example, it is determined that the subject is not at home but absent between times t3 and t4. Therefore, even if there are consecutive detection responses in the bedroom due to erroneous detection by the human sensor 31 between times t3 and t4, the subject will not be erroneously presumed to be asleep.

[0023] Here, the correspondence between the components of this embodiment and those of the present disclosure will be clarified. In this embodiment, the sensors 30 (human sensors 31-33, door sensor 35) correspond to a plurality of sensors, the memory unit 44 of the management server 40 corresponds to the memory unit, and the control unit 42 that executes the sleep state determination process (at least S200-S240, S270-S295) corresponds to the determination unit. In addition, by explaining the operation of the sleep determination system 10, an example of the sleep determination method and learning model generation method of the present disclosure will also be clarified.

[0024] As described above, the sleep determination system 10 (first sleep determination system) of the first embodiment acquires detection data from the sensors 30, and determines the sleep state of the subject based on the acquired detection data and the learning model 45. The learning model 45 is learned by collecting time-series detection data from the sensors 30 and time-series sleep data on whether the subject is in a sleeping state or not. By using such a learning model 45, the sleep state can be appropriately determined, and the sleeping time of the subject can be correctly understood.

[0025] In the sleep determination system 10 of the first embodiment, the learning model 45 is stored in the storage unit 44. The learning model 45 is learned by collecting data detected by a specific sensor 50, which is temporarily placed in a bedroom (sleeping place) during the learning period and has a higher detection accuracy than the sensors 30, as sleep data. This improves the reliability of the learning model 45, and the sleep state can be more appropriately determined. In addition, since the subject or the manager K does not need to collect sleep data and the burden of collecting sleep data can be reduced, the learning period can be set to a relatively long period such as several months, and the reliability of the learning model 45 can be further improved. In addition, the specific sensor 50 is temporarily placed only during the learning period, and the detection data of a relatively inexpensive infrared sensor (human sensor 31, 32, 33) or a magnetic sensor (door sensor 35) is used in the sleep state determination process. This allows the sleep determination system 10 to have a simple configuration with reduced costs.

[0026] Furthermore, in the sleep determination system 10, a learning model 45 that is trained from time-series detection data using the number of detections per unit time of each human presence sensor as a feature is stored in the storage unit 44. The control unit 42 then extracts the number of detections of each human presence sensor in a predetermined time period to be determined and in sections before and after the period as feature amounts from the acquired detection data, and determines the sleep state in the target section using the extracted feature amounts and the learning model 45. This allows for more appropriate sleep state determination by considering the feature amounts of not only the section to be determined but also the sections before and after the section.

[0027] Furthermore, the sleep determination system 10 can perform model-based determination that determines the sleep state using the detection data and the learning model 45, and rule-based determination that determines the sleep state based on the detection data that the human presence sensors 32 and 33 located outside the bedroom do not detect the sleep state and the human presence sensor 31 located in the bedroom continues to detect the sleep state. This makes it possible to prevent erroneous determination of the sleep state and to determine the sleep state with high accuracy.

[0028] [Second embodiment] Next, a second embodiment of the present disclosure will be described. Although illustration is omitted, the configuration of the sleep determination system of the second embodiment is the same as that of the first embodiment except that the learning model 45 is not stored, so the description of the configuration will be omitted and the sleep state determination process will be described. In the second embodiment, the learning model 45 is unnecessary, and the sleep state learning process is not performed. In addition, the sleep state determination process of the second embodiment omits model-based determination for determining the sleep state using the detection data and the learning model 45, and performs rule-based determination. For this reason, as shown in FIG. 8, the processes of S220 to S240 are omitted. In addition, when the control unit 42 sets the target section in S210, it performs the process of S250. That is, the control unit 42 determines that the target person is asleep when there is no detection response from the human sensors 32 and 33 arranged outside the bedroom and there is a continuous detection response from the human sensor 31 arranged in the bedroom, in a state where it is confirmed that the target person is at home based on the detection response of the door sensor 35.

[0029] In this way, the sleep determination system of the second embodiment (second sleep determination system) acquires detection data from the human sensors 31, 32, 33, and determines the sleep state based on the absence of detection from the human sensors 32, 33 located outside the bedroom and continuous detection from the human sensor 31 located in the bedroom. Therefore, the sleep state can be appropriately determined with relatively simple processing, and the subject's sleeping time can be correctly understood.

[0030] It goes without saying that the present disclosure is in no way limited to the above-described embodiments, and can be embodied in various forms as long as they fall within the technical scope of the present disclosure.

[0031] In the first embodiment, both the model-based and rule-based determinations are performed, but the rule-based determination may be omitted, i.e., S250 and S260 of the sleep state determination process in FIG.

[0032] In the first embodiment, the learning model 45 is created using the number of detections per unit time for each human sensor as a feature, but the learning model 45 may be created using another value extractable from the detection data as a feature. In the model-based determination, the sleep state is determined using the feature values ​​in the section to be determined and the sections before and after the section and the learning model 45, but this is not limited to this, and the sleep state may be determined using the feature values ​​only in the section to be determined and the learning model 45. In the first and second embodiments, the sleep state is determined by executing the sleep state determination process at regular intervals, for example, once a day, but this is not limited to this, and the sleep state may be determined every time detection data required for determination, such as detection data in the section to be determined, is obtained.

[0033] In the first embodiment, the sleep data is collected using a specific sensor 50 that is temporarily placed during the learning period and has higher detection accuracy than the sensors 30 (human sensor), but the use of the specific sensor 50 is not limited to this. For example, data on at least one of going to bed and getting up may be acquired by manual operation of the subject or manager K. Alternatively, the sleep data may be collected by attaching a wearable sensor to the subject only during the learning period.

[0034] In the first and second embodiments, the sensors 30 include the motion sensor 31 arranged in the bedroom, the motion sensors 32 and 33 arranged outside the bedroom, and the door sensor 35, but the present invention is not limited to this and may include only a motion sensor and not the door sensor 35. Furthermore, it is sufficient to include the motion sensor 31 arranged in the bedroom and one or more motion sensors arranged outside the bedroom, and the locations of the motion sensors 32 and 33 outside the bedroom are just examples and may be located in the kitchen, washroom, bathroom, balcony, etc. Furthermore, in the case of a one-room type residence, the motion sensor 31 may be located near the bed (sleeping area) and the other motion sensors may be located in places other than near the bed.

[0035] In the first and second embodiments, the control unit 42 of the management server 40 executes the sleep state determination process, but this is not limited to this. For example, the monitoring device 20 of each residence may be capable of executing the sleep state determination process. Furthermore, the result of the determination process of the monitoring device 20 may be transmitted to the management server 40. In the first embodiment, when the monitoring device 20 executes the sleep state determination process, the monitoring device 20 may be provided with a storage unit that stores a learning model.

[0036] Although the first and second embodiments have been described as being in the form of a sleep determination system, they may be in the form of a sleep determination method by a computer. Also, the first embodiment may be in the form of a learning model generation method by a computer.

[0037] This specification also discloses the technical idea of ​​changing the "sleep determination system according to claim 1 or 2" in claim 4 originally filed to "sleep determination system according to any one of claims 1 to 3." [Industrial Applicability]

[0038] The present disclosure is applicable to the manufacturing industry of sleep determination systems, etc. [Explanation of symbols]

[0039] 1 residence, 10 sleep determination system, 20 monitoring device, 22, 42 control unit, 24, 46 communication unit, 30 sensors, 31, 32, 33 human presence sensor, 35 door sensor, 40 management server, 44 memory unit, 45 learning model, K administrator, M mobile terminal, NW network.

Claims

1. A plurality of sensors are disposed at a plurality of locations including a bedroom and a non-bedroom in a residence of a subject; A memory unit that stores a learning model that is learned by collecting time-series detection data of the plurality of sensors and time-series sleep data indicating whether the subject is in a sleeping state; a determination unit that acquires detection data from the plurality of sensors and determines a sleep state of a subject using the acquired detection data and the learning model; A sleep determination system comprising:

2. The memory unit stores the learning model learned by collecting data detected by a specific sensor that is temporarily placed in the bed during a learning period and has a detection accuracy higher than that of the sensor as the time-series sleep data. The sleep determination system according to claim 1 .

3. the storage unit stores the learning model that is learned from the time-series detection data using the number of detections per unit time for each sensor as a feature amount, The determination unit extracts, from the acquired detection data, a number of detections for each sensor in a predetermined time period to be determined and in periods before and after the period as features, and determines the sleep state in the target period using the extracted features and the learning model. The sleep determination system according to claim 1 .

4. The determination unit is capable of performing a model-based determination for determining a sleep state using the detection data and the learning model, and a rule-based determination for determining a sleep state based on the detection data that there is no detection of a sensor placed other than the sleeping place and the detection of a sensor placed in the sleeping place is continuous. The sleep determination system according to claim 1 .

5. A plurality of sensors are disposed at a plurality of locations including a bedroom and a non-bedroom in a residence of a subject; a determination unit that acquires detection data from the plurality of sensors and determines a sleep state of the subject based on the fact that, among the acquired detection data, there is no detection from a sensor disposed other than the sleeping area and detection from a sensor disposed in the sleeping area is continuous; A sleep determination system comprising:

6. (a) collecting time-series detection data from a plurality of sensors disposed at a plurality of locations, including a bedroom and a non-bedroom location, in a residence of a subject, and time-series sleep data indicating whether the subject is asleep or not, and generating a learning model; (b) acquiring detection data from the plurality of sensors, and determining a sleep state of the subject using the acquired detection data and the learning model; A sleep determination method comprising:

7. acquiring detection data from a plurality of sensors disposed in a plurality of locations including the bedroom and locations other than the bedroom in the residence of the subject, and judging the sleeping state of the subject based on the fact that, in the acquired detection data, there is no detection from the sensors disposed in locations other than the bedroom and detection from the sensor disposed in the bedroom is continuous; A sleep determination method comprising:

8. A step of determining the sleeping state of the subject when the detection data of the multiple sensors is input, using a learning model generated by collecting time-series detection data of multiple sensors arranged in multiple locations, including the bedroom and non-bedroom locations, in the subject's residence, and time-series sleep data indicating whether the subject is in a sleeping state or not; A sleep determination method comprising:

9. (a) collecting time-series detection data from a plurality of sensors disposed at a plurality of locations, including a bedroom and a non-bedroom, in a residence of a subject; (b) collecting time-series sleep data indicating whether the subject is in a sleep state; (c) generating a learning model for determining a sleep state of a subject using the collected detection data and the collected sleep data; A learning model generation method comprising:

Images