JP6611871B1 - Monitoring device - Google Patents

Abstract

A monitoring device for detecting an operation for each person to be monitored is provided. A monitoring device for monitoring a specific subject extracts a feature amount of the subject's motion based on image data of the subject's motion and generates a learning model of the subject's motion. 411, an operation model storage unit 221 that stores a learning model of the operation generated by the operation learning unit 411, and an operation for determining the operation of the subject based on the operation learning model stored by the operation model storage unit 221 And a determination unit 211. [Selection] Figure 4

Description

  The present invention relates to a monitoring device that monitors a specific target person in a home or facility.

  2. Description of the Related Art Conventionally, there has been known an apparatus that determines whether or not a subject's operation is abnormal based on image data captured by a monitoring camera and reports the abnormality (see, for example, Patent Document 1). The apparatus of Patent Document 1 learns a series of standard operations when using an automatic transaction apparatus, and determines that an abnormality occurs when the subject's movement deviates from the standard movement.

JP 2018-10332 A

  In the device described in Patent Document 1, in order to determine whether or not the subject's motion is abnormal by comparing it with the standard motion, the characteristic motion corresponding to the subject's body condition, etc. It cannot be determined accurately. However, it is desirable to determine an abnormal action (abnormality) for each subject who has a different body condition or the like, especially when performing 24-hour posture monitoring for the purpose of nursing care.

One aspect of the present invention, generates a particular a monitoring device for monitoring a subject, extracting a feature amount of operation of the subject based on the image data of the operation subject, learning model of the operation of the subject a learning model generating unit to a learning model storage unit for storing a learning model of the operation generated by the learning model generating unit, based on the stored learned model by learning model storage unit, the operation of the subject An operation determining unit for determining , a biological information acquiring unit for acquiring the biological information of the subject, and an abnormality determining unit for determining whether or not an abnormality has occurred in the subject based on the biological information acquired by the biological information acquiring unit And comprising. The learning model generation unit extracts the feature amount of the target person's movement based on the image data of the target person's movement when the change determination unit determines that the target person has changed, and learns the target person's abnormal movement Generate a model.

  According to the present invention, an operation can be detected for each person to be monitored.

The figure which shows roughly an example of installation of the monitoring apparatus which concerns on embodiment of this invention. The perspective view of the monitoring apparatus which concerns on embodiment of this invention. Sectional drawing of the monitoring apparatus which concerns on embodiment of this invention. The block diagram which shows roughly the structure of the monitoring apparatus which concerns on embodiment of this invention. 5 is a flowchart illustrating an example of an operation determination process based on image data, which is executed by the controller group in FIG. 4. 5 is a flowchart illustrating an example of posture determination processing based on image data, which is executed by the controller group in FIG. 4. The flowchart which shows an example of the abnormality determination process based on attitude | position data performed with the controller group of FIG. The flowchart which shows an example of the abnormality determination process based on the measured value performed with the controller group of FIG. The flowchart which shows an example of the learning process of the operation | movement performed with the controller group of FIG. The figure for demonstrating operation | movement of the monitoring apparatus which concerns on embodiment of this invention.

  Hereinafter, embodiments of the present invention will be described with reference to FIGS. The monitoring apparatus according to the embodiment of the present invention is installed in a room of a nursing facility, for example, in order to constantly monitor a specific target person, and a normal operation is a standard operation depending on a physical state such as a bent waist. It learns and determines various actions including an abnormal action to be notified for a different subject.

  FIG. 1 is a diagram schematically illustrating an example of installation of monitoring apparatuses 1A and 1B (hereinafter collectively referred to as 1) according to an embodiment of the present invention. The monitoring device 1 is configured by a monitoring camera or the like, and is installed in a place where a monitoring target person spends alone, such as a living room of a nursing facility or the home of an elderly person living alone, and monitors the monitoring of the target person. The image data output by the monitoring device 1 is transmitted to a terminal installed in a nurse station of a care facility, a portable terminal of a family of a person to be watched over, or the like. FIG. 1 is an example when the monitoring device 1 is installed in a living room of a care facility. For example, as shown from the left side in FIG. 1, the living room has an entrance / exit of the living room, a partitioned area AR1 (toilet, etc.), an area AR2 (washing area, etc.) in which the subject walks, and the subject is mainly lying Area AR3 (area where beds are placed).

  As shown in FIG. 1, the monitoring device 1 </ b> A is installed on a wall or the like opposite to the entrance / exit. The monitoring device 1A monitors the entry of the subject into or out of the room, monitors the movement of the subject in the area AR2, and monitors the state of the subject sleeping in the area AR3 from above. . The monitoring device 1B is installed at a position where the subject is not concerned, such as the ceiling of the partitioned area AR1. Although FIG. 1 shows an example in which the monitoring devices 1A and 1B are installed, the number of monitoring devices 1 may be one or three or more depending on the size and layout of the living room.

  FIG. 2 is a perspective view of the monitoring device 1. The monitoring device 1 includes a substantially bulb-shaped housing 11. The housing 11 has a conductive base portion 12 in which a screw portion 12a is formed at one end portion. The screw portion 12a is screwed into a socket at the tip of a stand that can be attached to a wall surface, for example, and power is supplied to the main body via the base portion 12. You may make it screw on the socket of the duct rail attached to the wall surface. A cover 13 made of a translucent resin material is attached to the other end of the housing 11. By adopting such a light bulb shape, the monitoring device 1 can be easily installed like a commercially available light bulb, and can reduce the consciousness that a general surveillance camera gives to the subject. it can.

  FIG. 3 is a cross-sectional view of the monitoring device 1. As shown in the figure, in the housing 11, a camera 14 that captures the person being watched over, a plurality of (for example, eight) LEDs 15 that are arranged around the camera 14 at equal intervals in the circumferential direction, and around the camera 14. Three controllers 20, 30, and 40 (controller group 50) arranged at equal intervals in the circumferential direction are arranged, and electric power is supplied to them from the outside through the base 12 and a cable (not shown). The camera 14 is connected to each of the controllers 20, 30, and 40 via a cable (not shown). The casing 11 is provided with a slit 11s for exhaust heat (FIG. 1). The controller group 50 includes a notification unit such as a nurse call parent device (nurse call board) installed in a nurse station, and a management terminal for creating and managing care records (for example, a laptop computer for a care staff). Are electrically connected by wire or wirelessly. Although not shown, each controller 20, 30, 40 is provided with an external input port such as a USB (Universal Serial Bus).

  The camera 14 always images the person being watched over and transmits the image data to the controller group 50. As the camera 14, for example, a normal RGB camera having an optical zoom function of about 5 to 10 times is used. As shown in FIG. 3, the camera 14 is arranged at the center in the housing 11 so that the lens portion is parallel to the surface of the cover 13. The camera 14 has a lens unit fixed to the cover 13, and a main body unit is movable in the length direction of the housing 11 indicated by an arrow along a rail 11 r provided on the housing 11, thereby adjusting the magnification appropriately. can do.

  LED15 is comprised by high-intensity white LED, for example. The LED 15 is arranged so that the irradiation surface is parallel to the surface of the cover 13. When monitoring is performed in a dark room such as when the subject is sleeping, an infrared camera is used as the camera 14 instead of a normal RGB camera, and an infrared LED is used as the LED 15 instead of a high-intensity white LED. May be. While the RGB camera from which the infrared cut filter has been removed is used as the camera 14, a high-brightness white LED and an infrared LED may be alternately arranged around the camera 14 at equal intervals in the circumferential direction.

  Here, the problems of conventional watching systems such as nurse calls and surveillance cameras will be described. In the conventional nurse call monitoring system, when the subject himself feels anomaly, the subject himself / herself operates the nurse call cordless handset in the room to notify the nursing staff. For this reason, when the subject person loses consciousness or cannot move, the notification cannot be performed. Further, in a conventional monitoring system using a monitoring camera, a nursing staff or the like looks at the video of the monitoring camera to determine whether there is any change in the subject. For this reason, when one nursing staff oversees multiple subjects, or when a single family member constantly monitors the subjects who are living alone, it may be delayed to notice any changes in the subjects due to oversight or oversight. There is. For example, if the subject loses consciousness or becomes unable to move due to a fall or stagnation, the subject is left until the care staff or the like notices.

  Therefore, in the embodiment of the present invention, various actions of the watching target person including an abnormal action (abnormal action) when there is an abnormality are learned as an action model, and the watching target imaged by the surveillance camera (camera 14) is acquired. By determining whether or not the person's action is an abnormal action, it is configured to automatically detect an abnormality of the subject person.

  The motion of a watched person who needs care, such as an elderly person, may differ from the standard motion of a standard adult, depending on the physical condition such as bending of the waist. In such a case, it is difficult to learn a standard operation in advance and detect an anomaly as compared with the operation of the person being watched imaged by the surveillance camera. For example, there is a possibility that an operation that does not cause a problem for the target person, such as an operation that moves while the waist is bent, may be determined as an abnormal operation.

  In order to prevent such erroneous detection, the monitoring device 1 learns an operation model for each watching target person, determines an action for each target person, and detects an anomaly. Specifically, based on the care record of the person being watched over, various actions including an abnormal action to be notified for each target person are demonstrated in advance by a care staff and learned as an action model. For example, in the case of the monitoring device 1 that watches an object person whose waist is bent, an action model that learns an action that moves while the waist is bent, an action that falls, and the like is learned as an action model. Further, by further learning the actual motion of the target person as a motion model during watching and monitoring of the target person, the motion model is updated to match the actual situation of the target person.

  A configuration of the monitoring apparatus 1 that can realize the learning of the behavior model as described above and the determination of the behavior and the abnormality will be described.

  FIG. 4 is a block diagram schematically showing the configuration of the monitoring device 1. The monitoring device 1 includes a camera 14, a controller group 50 that is electrically connected to the camera 14, a notification unit 16 that is electrically connected to the controller group 50 in a wired or wireless manner, and a management terminal 17. The controller group 50 includes controllers 20, 30, and 40 that are configured to be able to communicate with each other.

  The notification unit 16 is, for example, a nurse call parent device installed in a nurse station of a care facility, and notifies a care staff or the like according to a command from the controller group 50. The alerting | reporting part 16 is good also as portable terminals, such as a smart phone not only in what is installed in the same premise as the monitoring apparatus 1, but the watching target person's family. When the monitoring device 1 is installed in the home of a person who is living alone, the notification unit 16 may be a terminal of a medical institution or the like. Moreover, it is good also as the alerting | reporting part 16 combining said terminal.

  The management terminal 17 is, for example, a notebook computer used by a nursing staff at a nursing facility to create and manage nursing care records. Text data such as a target person's action schedule included in the nursing care record and a judgment result of the behavior of the subject person Are transmitted to and received from the controller group 50. The management terminal 17 is not limited to being installed on the same premises as the monitoring device 1 and may be a mobile terminal such as a smartphone possessed by the family of the person being watched over. When the monitoring device 1 is applied to watch over a subject who is hospitalized in a medical facility, the care record may be created using the management terminal 17 as a terminal of a medical institution or the like. Also, the management terminal 17 may be a combination of a plurality of terminals.

  The controller 20 includes a computer having a GPU (Graphics Processing Unit) 21, a memory 22 such as a ROM and a RAM, and other peripheral circuits.

  As a functional configuration, the GPU 21 includes a motion determination unit 211 that determines the current motion based on the motion model of the watching target person, a posture determination unit 212 that determines the current posture based on the posture model of the target person, And an output unit 213 that outputs the posture determined by the determination unit 212 to the controller 30. The motion determination unit 211 and the posture determination unit 212 may be configured by different GPUs.

  The memory 22 has, as a functional configuration, an operation model storage unit 221 that stores a model of an operation of a subject including normal operations such as movement, meal, and sleep and abnormal operations such as falls and aspiration, A posture model storage unit 222 that stores a model of the posture of the subject such as a sitting position and a supine position.

  The controller 30 includes a CPU (Central Processing Unit) 31, a memory 32 such as ROM and RAM, and a computer having other peripheral circuits.

  As a functional configuration, the CPU 31 determines whether or not the change in the posture of the target person output by the output unit 213 is abnormal based on the abnormal posture change determination condition and identifies the type of the abnormal posture. A change determination unit 311; a vital measurement unit 312 that measures vital signs such as heart rate, blood pressure, respiration rate, blood oxygen saturation (SPO2), body temperature, etc. of the subject based on image data from the camera 14; Determination of whether or not an abnormality has occurred in the subject based on a current measurement value or the like by the measurement unit 312, and determination of an abnormality determination unit 313 that identifies the type of the abnormality, determination of the abnormal posture change determination unit 311 and the abnormality determination unit 313 An anomaly information output unit 314 that outputs anomaly information to the notification unit 16 and the management terminal 17 according to the result, an abnormal posture change determination unit 311, and an anomaly determination unit 3 A learning time determination unit 315 that determines times tps to tpe and tvs to tve to be learned as an abnormal operation of the subject in the image data from the camera 14 according to the type of anomaly specified by 3, and a learning time determination unit Image data at times tps to tpe and tvs to tve determined by 315 is extracted, and learning data for the motion (abnormal motion) of the subject along with the type of anomaly identified by the abnormal posture change determination unit 311 and the anomaly determination unit 313 As a learning data output unit 316 that outputs to the controller 40.

  The vital measurement unit 312 measures various vital signs of the subject based on the image data from the camera 14. For example, the heart rate can be measured based on fluctuations in the intensity of light received by the camera 14. Of the three component light received by the RGB camera, the blue (B) component light with a short wavelength is reflected by the subject's skin surface, and the green (G) component light with a longer wavelength is present in the presence of capillaries. It penetrates into the dermis layer and is absorbed by hemoglobin contained in red blood cells in the blood. That is, the light reception intensity of the blue (B) component varies according to the brightness of the room (environment light), and the light reception intensity of the green (G) component varies according to the pulsation of the subject. Therefore, it is possible to measure the pulsation cycle (heart rate) without being affected by the ambient light based on the fluctuation of the received light intensity.

  The memory 32 has, as a functional configuration, an action schedule storage unit 321 that stores an action schedule of the target person such as a meal or entrance / exit that is input in advance based on a care record of the target person, and when the target person has an abnormality. Abnormal posture change determination condition storage unit 322 for storing the abnormal posture change (posture change) determination condition for each type of anomaly, and the abnormal change determination condition based on the measured value of the vital sign of the subject. Each of which has an abnormality determination condition storage unit 323 and a vital measurement value storage unit 324 that stores a measured value of the vital sign of the subject measured by the vital measurement unit 312. As an example of a condition for determining an abnormal posture change, for example, in the case of determining a fall, the condition is that the posture of the subject changes from standing to lying within a certain time (for example, within 3 seconds). To do. Moreover, when determining aspiration, it is presumed that it is estimated that it is during a meal based on an action plan, and that a heart rate exceeds a predetermined value (for example, 100 times / minute).

  The controller 40 includes a computer having a GPU (Graphics Processing Unit) 41, a memory 42 such as a ROM and a RAM, and other peripheral circuits.

  The GPU 41 has, as a functional configuration, an action learning unit 411 that extracts and learns a feature amount corresponding to a feature of an action based on learning data of the action of the subject, and generates a learning model of the action of the subject, A posture learning unit 412 that extracts and learns a posture feature amount based on learning data of the posture of the person, and generates a learning model of the posture of the subject, and the behavior of the subject and the posture learning model of the controller 20 An output unit 413 that outputs to the model storage unit 221 and the posture model storage unit 222.

  The memory 42 has, as a functional configuration, a motion learning data storage unit 421 that stores, for each type of motion, motion learning data of the subject output by the learning data output unit 316, and a subject captured in advance by the camera 14. A posture learning data storage unit 422 that stores learning data (image data) of each posture for each type of posture. The motion learning data storage unit 421 is configured to store, in addition to the learning data output from the learning data output unit 316, learning data of the motion of the subject previously captured by the camera 14 for each type of motion. Also good.

  The motion learning unit 411 and the posture learning unit 412 perform supervised learning of image data by a known machine learning program suitable for learning of image data, for example, deep learning which is a machine learning method using a multilayer neural network. . In other words, the motion learning unit 411 learns image data (image data at times tps to tpe, tvs to tve, etc.) stored and accumulated by the motion learning data storage unit 421 for each type of motion, and for each type of motion. Generate a behavior model. In addition, the posture learning unit 412 learns the image data stored and accumulated for each posture type by the posture learning data storage unit 422, and generates a posture model for each posture type. The learned behavior model and posture model are output to the controller 20 by the output unit 413, stored in the behavior model storage unit 221 and the posture model storage unit 222, respectively, and updated.

  The motion determination unit 211 of the controller 20 determines the current motion of the subject based on the latest motion model of the subject stored in the motion model storage unit 221, and determines whether or not it is an abnormal motion. judge. In addition, the posture determination unit 212 determines the current posture of the subject based on the latest posture model of the subject stored in the posture model storage unit 222.

  More specifically, the motion determination unit 211 extracts the feature amount of the current motion of the subject in the image data from the camera 14 and compares the target motion model with the motion model of the target to determine the current motion of the subject. At the same time, it is determined whether or not this is an abnormal operation. In addition, the posture determination unit 212 extracts a feature amount of the current posture of the subject in the image data, and compares the feature amount with the feature amount of the posture model of the subject stored in the posture model storage unit 222. Determine the posture. The motion or posture having the highest degree of correlation with the learning model (abnormal motion model, posture model) is determined as the current motion or posture. Note that the motion determination unit 211 can also calculate an absolute amount of motion (activity amount) based on the current motion of the subject in the image data. By recording changes in the amount of activity along with vital signs, it is possible to determine not only accidents and illnesses but also early symptoms of dementia.

  FIG. 5A is a flowchart illustrating an example of an operation determination process based on image data that is executed by the GPUs 21 and 41 and the CPU 31 of the controller group 50 in accordance with a program stored in the memories 22, 32, and 42 of the controller group 50 in advance. The process shown in this flowchart is started, for example, when the monitoring apparatus 1 is turned on, and is repeated at a predetermined cycle. First, in step S <b> 1, current image data captured by the camera 14 is read by the processing in the operation determination unit 211. Next, in step S2, the motion of the subject in the current image data is determined based on the motion model stored in the motion model storage unit 221, and in step S3, it is determined whether the motion is an abnormal motion. If the result is affirmative in step S3, the process proceeds to step S4. If the result is negative, the process is terminated. In step S <b> 4, the abnormality information output unit 314 outputs the abnormality information to the notification unit 16 and the management terminal 17.

  FIG. 5B is a flowchart illustrating an example of posture determination processing based on image data executed by the controller group 50. First, in step S <b> 10, current image data captured by the camera 14 is read by processing in the posture determination unit 212. Next, in step S <b> 11, the posture of the subject in the current image data is determined based on the posture model stored in the posture model storage unit 222. Next, in step S <b> 12, data on the current posture of the subject determined by the posture determination unit 212 is output to the abnormal posture change determination unit 311 by the processing in the output unit 213.

  FIG. 6A is a flowchart illustrating an example of an anomaly determination process based on posture data, which is executed by the controller group 50. First, in step S <b> 20, the current posture data of the target person output by the output unit 213 is read by the processing in the abnormal posture change determination unit 311. Next, in step S21, based on the abnormal posture change determination condition stored in the abnormal posture change determination condition storage unit 322, the posture change between the posture read before and the posture read this time is abnormal. And determine the type of anomaly. If the result is affirmative in step S21, the process proceeds to step S22. If the result is negative, the process is terminated. In step S <b> 22, the abnormality information output unit 314 outputs the abnormality information to the notification unit 16 and the management terminal 17.

  Next, in step S <b> 23, times tps to tpe to be learned as the abnormal motion of the subject are determined based on the type of anomaly identified by the abnormal posture change determination unit 311 by the processing in the learning time determination unit 315. Next, in step S24, the learning data output unit 316 extracts the image data from the time tps to tpe determined by the learning time determination unit 315 from the image data from the camera 14, and learns the motion of the subject person. Output as data.

  FIG. 6B is a flowchart illustrating an example of an anomaly determination process based on measurement values, which is executed by the controller group 50. First, in step S <b> 30, current image data captured by the camera 14 is read by processing in the vital measurement unit 312. Next, in step S31, the current vital sign of the subject is measured based on the image data. Next, in step S <b> 32, the measurement value by the vital measurement unit 312, the target person's action schedule stored in the action schedule storage unit 321, and the abnormality determination condition storage unit 323 are stored by the process in the abnormality determination unit 313. Based on the change determination condition, it is determined whether or not a change has occurred in the subject, and the type of change is specified. If the determination in step S32 is affirmative, the process proceeds to step S33, and if the determination is negative, the process is terminated. In step S33, the abnormal information output unit 314 outputs the abnormal information to the notification unit 16 and the management terminal 17.

  Next, in step S34, the process at the learning time determination unit 315 determines times tvs to tve to be learned as the abnormal motion of the subject based on the type of change specified by the change determination unit 313. Next, in step S35, the learning data output unit 316 extracts the image data of times tvs to tve determined by the learning time determination unit 315 from the image data from the camera 14, and learns the motion of the subject person. Output as data.

  FIG. 7 is a flowchart illustrating an example of an operation learning process executed by the controller group 50. Unlike the processes of the flowcharts of FIGS. 5A to 6B, the process illustrated in this flowchart is started at any time in response to an instruction input by a user who manages the monitoring device 1 via an appropriate switch or the like. First, in step S <b> 40, the learning data of the subject's motion output by the learning data output unit 316 and stored in the motion learning data storage unit 421 is read by the processing in the motion learning unit 411. Next, in step S41, the motion of the subject is learned, and in step S42, a motion model is generated. Next, in step S <b> 43, the behavior model generated by the behavior learning unit 411 is output to the behavior model storage unit 221 of the controller 20 by the processing in the output unit 413.

  The posture learning process is executed in the same manner as the action learning process. That is, the learning data of the subject's posture stored in the posture learning data storage unit 422 by the processing in the posture learning unit 412 is read (S40), the posture of the subject is learned (S41), and a posture model is generated ( (S42) The posture model is output to the posture model storage unit 222 of the controller 20 by the processing in the output unit 413 (S43).

  The main operation of the monitoring device 1 according to the present embodiment will be described more specifically. FIG. 8 is a diagram for explaining the operation of the monitoring device 1. In FIG. 8, from the top of the figure, the posture of the subject determined by the posture determination unit 212, the behavior plan of the subject stored by the behavior plan storage unit 321, and the abnormal posture change by the abnormal posture change determination unit 311. The determination result, the heart rate as an example of the vital sign measured by the vital measurement unit 312, and the abnormal determination result by the abnormal determination unit 313 are respectively shown in time series.

  It is assumed that the monitoring device 1 monitors, for example, an elderly subject who is bent in a nursing facility. When the subject person alone in the living room of the care facility falls from time tps to tpe, the monitoring device 1 installed in the living room determines that the posture change is abnormal and specifies that the fall has occurred ( S10 to S12, S20 to S21). The change of the subject is notified to the care staff via the nurse call board (notification unit 16) installed in the nurse station (S22). For this reason, even when the subject person cannot use the nurse call by himself because he or she cannot move due to falling, the care staff can be notified of the change. Even when the care staff is keeping an eye on the video of the monitoring camera 14, it is possible to notice the change in the subject by the notification by the monitoring device 1. Moreover, the information on the change of the target person is also transmitted to the notebook personal computer (management terminal 17) of the care staff (S22). As a result, information such as the type and time of occurrence of the anomaly that has occurred in the subject is automatically recorded in the format of the nursing care record and the burden on the nursing staff can be reduced.

  Image data of times tps to tpe estimated that the subject has fallen is extracted and accumulated as learning data of the fall motion (S23, S24). The user of the monitoring device 1 connects the monitoring device 1 with the cover 13 removed from the socket to a user terminal such as a personal computer via a USB cable or the like, and learns the operations stored and accumulated in the memory 42 of the controller 40 It is possible to check the data accumulation status. When the user confirms the accumulation of learning data of an amount necessary for learning of the overturning operation and instructs the start of learning of the overturning operation, learning by the controller 40 is started, a learning model is generated, and the learning of the controller 20 is performed. The model is updated (S40 to S43). The learning model used to determine the movement is updated from the movement model generated based on the demonstrations of nursing staff, etc., to the learning model of the movement by the actual target person, thereby determining the movement of the target person. Accuracy is improved.

  If the subject's aspiration occurs during the meal and the heart rate of the subject rises above a predetermined value (for example, 100 times / minute) at the time ta due to the concealment, the monitoring device 1 causes aspiration (abnormality). It is determined that it has occurred (S31 to S33). Image data at times tvs to tve, which are estimated to have caused aspiration and aspiration, are extracted and accumulated as learning data for aspiration operations, and learned according to user instructions (S34, S35, S40). To S43). Since the time extracted as action learning data is determined according to the type of anomaly, it is possible to learn including the action before the anomaly occurs. For example, it is possible to learn including an inappropriate posture at the time of eating and drinking, which is an operation before aspiration occurs.

  After the aspiration behavior model is updated, if the subject at the time of eating or drinking takes an inappropriate posture that leads to aspiration, the monitoring device 1 determines that the behavior is aspiration (operation that leads to aspiration), A change in the subject is notified (S1 to S3). The care staff who received the notification can be careful of the posture of the subject when eating or drinking by rushing to the subject's room or by a nurse call. Even if the subject eats and drinks alone, the occurrence of aspiration can be prevented in advance. Moreover, since the posture that leads to aspiration is learned for each subject, it can be applied to a subject whose proper posture during eating and drinking is different from the standard adult posture due to bending of the spine or hemiplegia. In addition, when nursing staff assist the target person at the time of eating and drinking, according to the state of the target person, judging the posture that leads to aspiration while looking at the state, the posture at that time as the action of aspiration You can also learn. That is, an abnormal operation can be detected for each watching target person.

  In the monitoring device 1, the action schedule of the subject person and the measured value of the vital sign are stored in time series in the memory 32 of the controller 30. When the person being watched over visits a medical institution, these data can be presented to the doctor. By comparing the schedule of actions such as meals and waking up with changes in vital signs, it can be used for medical treatment and diagnosis of the subject.

According to this embodiment, the following effects can be obtained.
(1) The monitoring device 1 that monitors a specific watching target person extracts a feature amount of the watching target person's movement based on the image data of the watching target person's action, and generates a learning model of the watching target person's action. Based on the behavior learning unit 411, the behavior model storage unit 221 that stores the behavior learning model generated by the behavior learning unit 411, and the behavior learning model stored in the behavior model storage unit 221, And an operation determination unit 211 that determines whether or not (FIG. 4). By learning the action for each subject, it is possible to determine an action including an abnormal action (abnormal change) for each subject who has a different body state or the like.

(2) The monitoring apparatus 1 determines whether or not an abnormality has occurred in the watching target based on the vital measurement unit 312 that measures the vital sign of the watching target and the vital sign measured by the vital measuring unit 312. An anomaly determination unit 313 is further provided (FIG. 4). The motion learning unit 411 extracts the feature amount of the watching target person based on the image data of the watching target person's motion when it is determined by the change determination part 313 that the watching target person has changed, and the watching target person A learning model of abnormal behavior is generated. Thereby, based on the change of the vital sign, it is possible to automatically extract the learning data of the operation when the abnormality occurs and the operation before and after the occurrence of the abnormality.

(3) The monitoring device 1 further includes a management terminal 17 that acquires an action schedule of the watching target person (FIG. 4). The anomaly determination unit 313 further determines whether an anomaly has occurred in the person being watched based on the action schedule acquired by the management terminal 17. By comparing the action schedule of the subject and the change in vital signs, it can be used for medical treatment and diagnosis of the subject.

  In addition, the said embodiment can be deform | transformed into various forms. Hereinafter, modified examples will be described. In the above-described embodiment, the monitoring device 1 learns the operation of a specific target person and determines the operation of the target person based on the learned operation. It is not limited to such a thing. For example, apply a monitoring device to a subject with a specific physical condition, learn a motion model according to that physical condition, and use the learned motion model to determine the motion of another subject with a similar physical condition May be.

  In the above embodiment, the vital measurement unit 312 measures the vital sign of the subject based on the image data from the camera 14, but the configuration of the biological information acquisition unit that obtains the subject's biological information is as described above. Not limited to things. You may acquire the measured value of the vital sensor of the type with which a subject wears by a cable or radio | wireless.

  In the above embodiment, the text data of the action schedule of the target person included in the care record created by the management terminal 17 is transmitted to the controller group 50, but the action schedule acquisition unit for acquiring the action schedule of the target person The configuration is not limited to this. For example, as the determination of the operation performed on the controller group 50 side, the behavior of the subject person such as a meal or sleep is automatically determined based on the image data from the camera 14, and the determined behavior is text data for nursing records. May be output to the management terminal 17.

  In the above embodiment, the monitoring device 1 is configured to include the camera 14 in the main body, but the monitoring device that monitors a specific target person is not limited to this. For example, the image data of the subject person may be acquired from a monitoring camera or the like different from the monitoring device by wire or wireless, and the learning and determination of the operation may be performed based on the acquired image data. In this case, for example, image data of surveillance cameras from different manufacturers can be used as learning data by supporting network camera standardization standards such as ONVIF (Open Network Video Interface Forum).

  The above description is merely an example, and the present invention is not limited to the above-described embodiments and modifications unless the features of the present invention are impaired. It is also possible to arbitrarily combine one or more of the above-described embodiments and modifications, and it is also possible to combine modifications.

1 (1A, 1B) monitoring device, 11 housing, 11r rail, 11s slit, 12 base part, 13 cover, 14 camera, 15 LED, 16 notification part, 17 management terminal, 20, 30, 40 controller, 21, 41 GPU, 22, 32, 42 memory, 31 CPU, 50 controller group, 211 motion determination unit, 212 posture determination unit, 213 output unit, 221 motion model storage unit, 222 posture model storage unit, 311 abnormal posture change determination unit, 312 Vital measurement unit, 313 Change determination unit, 314 Change information output unit, 315 Learning time determination unit, 316 Learning data output unit, 321 Action schedule storage unit, 322 Abnormal posture change determination condition storage unit, 323 Change determination condition storage unit, 324 vital measurement value storage unit, 411 motion learning unit, 412 posture learning unit, 413 Output unit, 421 Motion learning data storage unit, 422 Posture learning data storage unit

Claims (2)

A monitoring device for monitoring a specific target person,
A learning model generating unit on the basis of the image data of the subject's operation to extract the feature amount of operation of the subject, to generate a learning model of the operation of the subject,
A learning model storage unit for storing a learning model of the operation generated by the learning model generating unit,
Based on the stored learned model by the learning model storage unit, and the operation determination unit determines the operation of the subject,
A biological information acquisition unit that acquires biological information of the subject;
An abnormality determination unit that determines whether or not an abnormality has occurred in the subject based on the biological information acquired by the biological information acquisition unit ;
The learning model generation unit extracts a feature amount of the subject's motion based on image data of the subject's motion when the anomaly determination unit determines that an abnormality has occurred in the subject, and the target the supervision device and generating to said Rukoto learning model of abnormal operation.   The monitoring device according to claim 1,
  An action plan acquisition unit for acquiring the target person's action plan;
  The monitoring apparatus according to claim 1, wherein the abnormality determination unit further determines whether or not an abnormality has occurred in the target person based on the action schedule acquired by the action schedule acquisition unit.

Images