DETECTION OF A PERSON FALLING
This invention relates to methods and systems for monitoring the condition and health of a person in a building, normally the person's home in which he or she lives alone. More particularly the invention concerns detecting reliably whether the person has fallen, and providing an appropriate alert.
People living alone, especially those with mobility problems, may have difficulty in summoning help if they fall, for example by using the telephone or a conventional emergency panic button. Conventional panic button systems involve the person wearing a device and pushing a button on the device in an emergency situation. The device then sends an alert by radio transmission to a dialler that connects to emergency services or to a remote monitoring centre, which can then take appropriate action. In many falls, elderly people in particular are unable to press the button, and may simply fall into an unconscious state or may become severely disorientated. Some devices activate a speaker phone, but again if the device is not activated, the speaker phone will not activate. The elderly incapacitated person may also not be able to shout to a distant room and have their voice picked up by a microphone.
Wearable devices with an acceleration sensor which can measure sudden changes in velocity, are also known. Such acceleration sensing systems are however unable to detect a gentle slide to the ground, which may happen for example when the person experiences a seizure. These systems are unable to distinguish between slow falls and normal motion around the house, so they either generate an overwhelming number of false alarms, or they fail to detect emergency situations. Such systems can also create false alarms if for example the wearable device, which may be a pendant, is dropped or thrown across the room. Even devices which detect tilt are also prone to false alarm, or to failure to activate in a genuine emergency. Accordingly, the present invention provides a wireless transmitter apparatus configured to be worn by a person to be monitored, comprising an accelerometer, a processor for comparing sensed values from the accelerometer with stored threshold values to provide and transmit an output signal indicative of whether the person may have fallen.
The invention also provides a method of indicating that a person normally wearing a wireless transmitter apparatus may have fallen, comprising using an accelerometer in the apparatus to sense acceleration, and comparing the sensed acceleration with stores threshold values to determine whether it is indicative of a fall.
In order that the invention may be better understood, a preferred embodiment will now be described with reference to the accompanying schematic drawings, in which:
Figure 1 shows a pendant embodying the invention;
Figure 2 shows a typical network within the home, embodying the invention;
Figure 3 is a block diagram of an entire system within the home and external to the home, embodying the invention;
Figure 4 is a flow diagram showing the detection of a fall in accordance with the invention;
Figure 5 illustrates parts of the tracking process using a tag, which can be used in the present invention; and
Figure 6 shows part of a database for use in the present invention.
The present invention arose from a realisation that the detection of a fall, using a wearable motion sensor, can be made more reliable by taking account of physiological data relating to the person being monitored and/or to the population in general and/or to the person's environment in the home and his pattern of activity. In other words, a form of bio-feedback is used in the detection of an event which should trigger an alarm. This appreciation led to the development of an entire system designed to provide the necessary bio-feedback and to control the entire process of detection and alerting.
As shown in Figure 1, a neck pendant device has a main body 10 attached to a strap 11. The main body 10 of the device is shown from its side in Figure Ia, from which can be seen a pair of panic push buttons 12, 13 and a pair of infra-red thermopiles 1, 2 on opposite sides, the thermopiles being heat sensors for detecting body temperature or ambient temperature. A front view of the device is shown in Figure Ib. It will be appreciated that the device can be worn with either of its main surfaces touching the body, so that the other surface detects ambient temperature in the room or on the floor or bed for example. Those skilled in the art will recognise that the pendant device could be incorporated into a watch, brooch, or other wearable item, and that other bio-metric sensors, for example, pulse and heart rate sensors, as well as e.g. carbon monoxide, smoke, gas, or other environmental sensors could also be incorporated into the same device.
The pendant includes a radio transceiver powered by an internal battery which is preferably rechargeable and/or replaceable. In one example, the battery is rechargeable through a solar cell provided on the housing of the pendant. The transceiver is connected to a data processor with its own solid state memory (not shown) for detecting a possible alarm condition.
A typical network of devices installed in a person's home is illustrated in Figure 2. There is one neck pendant 10 for the person living in the home; if there are two people then of course there would be two such pendants, separately identifiable. A number of tags Tl to T5 are installed at fixed locations distributed around the house. Each tag comprises a Passive Infra-Red (PIR) detection system for detecting the proximity of an infra red emitting body such as the person living in the house. In addition to the PIR, each tag also includes a radio transceiver, for communicating with the pendant 10 and also with a gateway device 14. For optimum resilience of the system, particularly to overcome the effect of obstacles to radio transmission, such as walls and electromagnetic interference from strong emitters such as refrigerators, each tag is also configured to communicate with any of the other tags, and is thus able to relay data. Such Communications networks are well known in the art, and will not be described further in this specification. The radio communications links are illustrated by the lines Ll to L12 in Figure 2. For example, data may be transmitted from the pendant 10 to the gateway device 14 by way of the nearest tag Tl on lines Ll and L2, or else using another tag as a relay, through tags Tl and T3 using lines Ll, L3 and L7. Data from the gateway device 14 may be returned in a similar manner.
The gateway device 14 comprises a radio transceiver for communicating with the tags, and also a data processor for storing and processing data from the tags and from the pendant 10. The gateway device would normally be powered from the mains electricity with battery back-up. In order to communicate with the system external to the home, the gateway device 14 is wired to a telephone dialler or cable modem or satellite modem 15, which has access to the Internet or telephone network 16. The purpose of each tag is to track the movement of the person around the house. The tags may for example be placed adjacent door thresholds or significant items of furniture such as the bath or the medicine cabinet.
Preferably, and as described below with reference to Figure 5, the tags include accelerometers for detecting the motion of movable objects to which they are fixed. They would then be attached to movable objects in the home, such as doors to the rooms, medicine cabinet doors and so on. This provides a more reliable indication of the passage of a human being, for tracking that person around the home. For example, with a PIR detector only it is difficult to tell whether a person entered or exited a room. With both the PIR and door sensors, the system senses if the door opens or closes, and measures the direction of the heat from the human body using the PIR. Taken together, this is a reliable indication of presence in a room.
A combination of such tags could of course be used. In the case of tags without accelerometers, confirmation that the tag has detected a person can be obtained by the simultaneous detection of motion of the person wearing the pendant 10: simultaneous detection of motion of the pendant and activation of the PIR in the tag is indicative of the presence of the person moving. Alternatively, as described below, the tag can provide confirmation by receiving personal ID from the pendant. Using conventional triangulation techniques, the location of the person in the home can also be verified by reading the radio signal strength of transmission between the pendant 10 and each of the tags Tl to T5: this processing is preferably performed in the gateway device 14.
A typical system for the overall detection and alerting of a person's fall is shown in Figure 3. It would be appreciated that the intelligence of the system is distributed between the pendant, the tags, the gateway device and external equipment including a remote platform 22, and this can be arranged in many different ways, with different functions carried out by different elements. In this example, a data analysis block 20 in the gateway device 14 receives detection data from each tag and pendant, and passes commands to additional sensors such as a video camera 25 within the home, to take a picture 26 which is then passed to an event confirmation handler 21. A variety of different sensors could be used to help confirm a specific event: for example smoke alarms, home alarms, carbon monoxide or other sensors, in addition to or instead of the video camera 25. Such events include fires and dangerous emissions, as well as falls.
Commands for data analysis can also be received directly from the remote platform 22. The remote platform 22 receives data from the gateway device 14 and transmits data to that device. The remote platform 22 also services other homes independently, and incorporates a database, such as that shown in Figure 6 and described below. The remote platform 22 communicates with an SMS server/SIP124, for sending text messages or voice messages relating to the alarm condition to people who may assist the person in the home, and also to the person in the home to provide reassurance or to provide interactive personal verification of an alarm condition. Those skilled in the art will readily realise that any form of messaging, such as MMS picture messaging, Instant Message, fax, email, could also be used. The remote platform 22 also communicates with the worldwide web 23 to provide communication with call centre operatives, family members or neighbours able to log in remotely. Such people can also receive alerts by telephone or mobile phone from the server 24, and they can request data from a particular sensor within the home, immediately through the system shown in Figure 3. So for example a family member can provide periodic checks, or can verify an alert condition, by logging on through the Internet 23 or by telephone using the server 24, and can access the picture 26 for example.
The use of the pendant 10 will now be described with reference to Figure 4. The accelerometer and the thermopile co-operate with the onboard processor and memory to provide a reliable indication of a fall, described as an "event". They also take account of historic data in the form of biofeedback.
When powered on, the pendant boots up and then fires or initiates a sample timer. The sample timer performs like a clock which controls the operation of the onboard sensors, i.e. the accelerometer, in a way that makes efficient use of battery energy, taking account of the likelihood of an event occurring. The duration set in the timer is variable, and is shown as one of the configurable values in the pendant: "sample timer". When the history of falls, or other environmental factors such as the presence of the person in the bathroom, or the person's recent discharge from hospital, suggest a greater likelihood of a fall event occurring, then the interval is reduced.. Controlling the sensor's sampling imposes an acceptable demand on the battery in the pendant.
At the end of the timed interval measured by the sample timer, the accelerometer sensors are powered up, and accelerometer data, for each of three orthogonal axes X, Y and Z, are taken and stored in a register in the onboard processor in the pendant. Each data value is compared with a preset low threshold value, and if they are all found to be below this value, then the sensors are powered down. The processor updates its memory with the detected data values, so that it can maintain a median value and make an appropriate adjustment, over time, to the basic threshold which it stores. The processor is configured to store accelerometer data obtained over a time window of typically 2 seconds, for averaging in order to improve signal to noise ratio. Varying battery voltage levels, and stray vibrations, lead to noise. In the process of powering down the sensors, this data storage window is cleared, and then the processor waits for the end of the next sample timer interval before powering up the sensors once again.
If the basic threshold check has a positive result, then the processor enters decision making mode. Accelerometer data are stored over the two second data window, typically at 500ms intervals, for each of the X5 Y and Z axes. These data are summed in a first cycle, and this is shown as an integration process in Figure 4. The outcome of the first cycle of processing is a median value for each axis, obtained by dividing the sum by the number of data values taken; a minimum value: X-Low, Y-Low, and Z-Low; and a maximum value: X-High, Y-High and Z-High. The decision maker in the processor also uses as an input the historic data of the minimum and maximum values for each axis, which are shown as configurable values in the pendant.
In this example, in a first cycle, a test is made separately for each axis to determine whether data from that axis should participate in a determination of a fall. For each axis, if the high value is more than 40% above the median value, or the low value is more than 40% below it, then the data values are used in the decision making process in the second cycle. Otherwise, they are ignored. The reason for this is that the data values are most likely to be typical of a fall.
In the second cycle, the filtered data from the first cycle, i.e. those which are to be used, from one or more axes, are summed in a further process shown as a definite integration. Thus the remaining values of X-Low, Y-Low and/or Z-Low are summed, and divided by their number, to provide an average low value, called XYZ-Low. Correspondingly, the remaining high values are summed and divided by their number to provide an average high value called XYZ-High. Each of these values may be an average of 1, 2 or 3 values taken from the first cycle. In this way, a large number of variables is processed statistically to produce just two significant values, hi this example, the value of XYZ- Low determined in the second cycle is compared with the stored historic averaged value XYX-Lo w, and correspondingly the measured value XYZ-High is compared with the stored historic averaged value of XYZ-High; if both are within a predetermined percentage of the corresponding stored value, i.e. they satisfy a threshold comparison, then this is determined as being indicative of a fall, in the process block "Is it a fall?" in Figure 4. In the affirmative, then the high and low values and median values used in the second cycle are stored by the processor and used to update the configurable values in an averaging process. In this way, the intelligence in the pendant itself can learn from falls that are determined to have occurred. Also, the processor causes the data to be output to the base station or gateway device 14, which communicates as shown with the remote platform 22. The database in the remote platform builds up the history of assumed falls, as well as confirmed falls. The remote platform 22 carries out processing as described below, taking account of the entire database, to provide updated configurable values for the pendant.
In the example shown in Figure 4, confirmation of a suspected fall is obtained with the use of one or both of the thermopiles 1, 2 in the pendant. Since the pendant may in practice be orientated with either of the thermopiles facing the body, the output of both thermopiles are usually used and compared. Maximum values of temperature constitute configurable values "Thermopile High", and minimum values constitute configurable values "Thermopile Low" in the pendant, as shown in Figure 4. These configurable values are used as threshold values of temperature. They are also adjusted using stored values in the database, taking account for example of physiological data personal to the wearer of the pendant, but also taking account of the expected ambient temperature and floor temperature and bed temperature in that person's home.
The processor in the pendant is capable of distinguishing conditions which may otherwise give rise to false alarms. If the pendant were not being worn any longer, then most likely the temperature detected by both thermopiles would be below Thermopile Low. In this case, the processor would decide to ignore the alert provided from processing the accelerometer data. This could also be verified by comparing the temperatures sensed by both thermopiles, since they would normally be expected to differ from one another. This however would depend upon the expected ambient temperature in the room. In the event that a person has fallen onto the floor, the pendant, if correctly positioned, would normally be between the person's body and the floor. In this case, one thermopile would register a higher temperature than the other; and also the lower temperature would be below the ambient room temperature, since floors are usually colder than the air temperature (homes with under floor heating would provide an exception to this, and this would be recorded in the database). Thus if one thermopile temperature measurement were substantially below Thermopile Low, and the other one were within a certain percentage of Thermopile High, then this would be indicative of a fall, and the processor would generate an event, i.e. provide an alarm message. The operation of each tag Tl to T5 will now be described with reference to Figure 5. In this preferred example, the tag has an accelerometer as well as a PIR.
The on board processor in the tag causes it to boot up when powered on, and, in a similar way to the pendant, a sample time is fired and the accelerometer is powered up at the end of the predetermined variable interval. This interval is preferably a configurable value for the tag, which can be updated using historic data generated by the tag itself, and also from the database. For example, if the person tends to slam a door, then threshold values of acceleration in one axis would be adjusted accordingly. The PlR sensor may also be powered up at this time, or else this may be delayed until the decision maker for accelerometer data provides a positive outcome.
As with the pendant, a preliminary threshold check is made, against basic thresholds, to determine whether there is any genuine activity. If there is, then the processor enters decision making mode. The decision making process involves comparing each sensed acceleration value, i.e. for each of the X, Y and Z axes, with a configurable threshold value. If these comparisons all agree that there is an event, then the processor proceeds to check the sensed output of the PIR, indicative of proximity of the person or another infrared emitter. If the PIR check yields a positive outcome, then the tag generates an event, i.e. it transmits a signal indicative of the person having passed that point in the house, hi this way, the tags provide reliable tracking of a person's movements within the house, and these data are recorded in the gateway device 14 and in the external database.
The gateway device includes intelligence for comparing the tracked movement of the person with normal patterns of activity, built up over a number of days or weeks, and can provide an alert to the remote platform 22 in the event of abnormal activity or inactivity. This would provide an alert unless, of course, the database had been informed that the person was away from home. Each tag may have its own internal power supply, and the drain on this power supply is then minimised by the sample timer, and the two-stage detection process whereby the PIR can be activated only upon a positive outcome of the accelerometer check.
As described above, the system includes a database, which may be distributed, but may be included in a single location in the remote platform 22 for example. A schematic structure for the database is shown in Figure 6.
The characteristics of falls, in particular the dynamic information such as acceleration ranges in each axis, provide valuable information for setting configurable threshold values, and for verifying whether a fall has taken place. These data tend to be typical of particular types of people, or particular medical conditions. They depend in part on the height of the person and his age and sex. They also depend on the full event itself, for example whether it is the result of a seizure, or an accidental impact or slipping on a carpet or rug for example. The person's body itself provides resilience in specific axes, which tends to slow the falling process, and causes accelerations below the acceleration due to gravity. Characteristic dynamic patterns of falls have been analysed, as a function of the person's speed in different horizontal and vertical directions.
Accordingly, the database records these fall values, which include the pendant configurable values for each axis and for the axes combined. The database includes fall values personal to individuals, but also statistically processed fall values for different populations.
In a similar way, thermopile values are accumulated for individuals and also for specific populations. These values of temperature depend on the person's body temperature which depends in part on their clothing. The values also depend on the domestic environment and the type of heating in the house. Since the person may come into contact with the mattress of a bed, or a sofa cushion, or, in the event of a fall, the floor, the characteristic ranges of temperatures of these items of furniture are also recorded.
For each person registered to use the system, the configurable values in the pendant are stored and are continually updated. Physiological data such as the height, weight, age, sex and medical history, including whether the person has recently been discharged from hospital, are stored in the database.
Data from the tags, organised by the gateway device, are stored as movement pattern characteristics. This may for example record the frequency of visits to the bathroom and bedroom and the duration of such visits. It may also record the frequency of the front door being opened.
Specific information as to the architecture of the house and the distribution of the tags is also recorded, for each subscribing individual, as shown in Figure 6. The position of auxiliary sensors such as the video camera are also recorded.
As mentioned above, triangulation techniques involving signal strength can be used to identify and track the location of the person within the home, and this data is also stored in the database.
Preferably, each neck pendant device includes a specific unique ID code which it transmits with its output data. This is preferably read by each tag, so that the tag then detects and logs the identity of the person. This provides confirmation and avoids false alarms for example due to cats and dogs.
Operation of the push buttons 12, 13 or either of them is also detected by the pendant processor, which verifies that there is a genuine "panic" actuation by any appropriate algorithm, before generating a "panic" event. For example, the processor may eliminate false alarms by requiring multiple operations of the same button.