DE102008049194B4 - Method for detecting falls - Google Patents

Abstract

Method for detecting falls by means of a sensor (BS), characterized in that acceleration values (a) are measured by the sensor (BS) and from this a fall presumption alarm (f) is determined (1, 2) that is determined on the basis of the measured acceleration values (a) a ratio (R) between a first probability of an occurrence of the measured acceleration values (a) at a fall and a second probability for an occurrence of the measured acceleration values (a) without a fall is estimated (3) then the fall presumption alarm (f) and combining the ratio (R) with a known first rate (HF) of falls and a known second rate (HA) of fall victim alarms without falling (4), and then combining a fall presumption alarm (f), ratio (R) , known first frequency (HF) and known second frequency (HA) a third probability (pF) for an actual presence of a St estimated shortly (5).

Description

Technical area

The present invention relates to a method for detecting falls, in particular falls of elderly people, wherein a sensor is used for detecting.

State of the art

People in the western industrialized countries are getting older. In 1960, the European Union still had about 34 million over 65-year-olds, compared with about 61 million in 2000. For this reason, solutions and products are increasingly in demand today that enable a self-determined life as far as possible for the elderly or of physically impaired people in an efficient manner. One way to do so called home emergency calls represent, for example, usually consist of a remote-controlled speakerphone and a worn on the wrist or neck radio transmitter with alarm button. These systems can be called in an emergency by pressing the alarm button help, since in particular elderly and / or physically impaired persons of a number of dangers such. B. Circulatory weakness, falls and resulting injuries, etc. are exposed.

Especially with increasing age accumulate physical discomfort, for example, about 20 percent of the elderly suffer from dizziness, which can easily lead to falls. But falls in particular can pose a great danger to elderly or physically impaired persons if, for example, a fall is not detected and / or the alarm button of the house emergency call due to z. B. a loss of consciousness or an accident can not be operated.

Therefore, to detect falls, so-called fall detectors are used. The detection of falls is carried out via a sensor system from which accelerations and a position of the upper body can be measured in the body axes. Usually, such a fall detector worn on the belt or the upper body near the waist to reflect movements of the center of gravity and to reliably detect falls in this way. Such a method for fall detection is described, for example, in the document "Brown, Garrett: An Accelerometer Based Case Detector: Development, Experimentation and Analysis. University of Michigan, 2005. ", Brown notes in this document that detecting falls with a wrist mounted crash detector is relatively unreliable due to natural hand and arm movements.

However, worn on the belt crash detectors have the disadvantage that they z. B. worn in addition to the radio transmitter of the house emergency call and thus z. B. for an alarm in the event of a fall in addition to the house emergency call must be coupled. Belt-worn crash detectors also have low wear comfort and are limitedly suitable for night-time wear. In addition, for example, elderly people are reluctant devices such. As a fall detector attached directly to the body and thus rarely used.

In the work "Degen, T .; Jaeckel H .; Rufer, M .; Wyss, S .: SPEEDY: A Case Detector in a Wrist Watch. Swiss Federal Institute of Technology, 2003 "describes a method for detecting falls with an elderly fall detector integrated into a wristwatch. The fall detector described in this document can be worn comfortably and easily. However, the method described for fall detection has the disadvantage that, depending on the direction of a case falls with different reliability can be detected. Ie. in particular, when the fall is associated with a lateral fall or a fallback, errors in fall detection (eg falls are not detected) or false alarms due to arm movements from which the fall detector is impacted will occur.

Presentation of the invention

The invention is therefore based on the object of specifying a method of the type mentioned, by which falls easily detected with the help of a sensor with high reliability and false alarms can be avoided.

This object is solved by the subject matter of independent claim 1. Advantageous embodiments of the present invention are described in the dependent claims.

According to the method of the invention, it is provided that acceleration values are measured by the sensor and from this a fall presumption alarm is determined. On the basis of the measured acceleration values, a ratio between a first probability for occurrence of the measured acceleration values in the event of a fall and a second probability for occurrence of the measured acceleration values in the event that no fall is present is then estimated. Then, the crash presumption alert and the ratio are combined with a known first occurrence of falls and a known second occurrence of fall victim alerts without fall. From a combination of fall presumption alarm, ratio, known first frequency and known second frequency then becomes a third one Probability of actual fall is estimated.

The main aspect of the proposed solution according to the invention is that on the basis of the third probability of an actual occurrence of a fall with very high reliability, the actual presence of a fall can be determined and thus easily false alarms z. B. due to arm or hand movements can be avoided. In addition, the reliability of the method according to the invention or a reliable distinction between camber and non-camber is not influenced by whether the sensor is worn on the belt or on the wrist. A fall direction (forward, lateral or backward) also has no influence on the method according to the invention.

abgeschätzt wird. Dabei wird durch eine Variable R eine Abschätzung des Verhältnisses zwischen einer ersten Wahrscheinlichkeit für ein Auftreten der gemessenen Beschleunigungswerte bei einem Sturz und einer zweiten Wahrscheinlichkeit für ein Auftreten der gemessenen Beschleunigungswerte für den Fall, dass kein Sturz vorliegt, beschrieben. Von einer Variablen Q wird eine Abschätzung eines Quotienten der bekannten ersten Häufigkeit von Stürzen und der bekannten zweiten Häufigkeit von Sturzvermutungsalarmen für den Fall, dass kein Sturz vorliegt, angegeben und mit einer Variable f wird ein Sturzvermutungsalarm beschrieben, wobei von der Variablen f zwei Werte ~A, A angenommen werden. Der erste Wert ~A steht dabei für keine Sturzvermutung und der zweite Wert A steht für eine Sturzvermutung.It is favorable if the third probability for an actual occurrence of a fall on the basis of the formula

is estimated. In this case, a variable R is used to estimate the ratio between a first probability for an occurrence of the measured acceleration values in the event of a fall and a second probability for an occurrence of the measured acceleration values in the event of no fall. A variable Q is used to give an estimate of a quotient of the known first frequency of falls and the known second frequency of fall victim alarms in the event of no fall, and a variable f is used to describe a fall presumption alarm, where of the variable f two values ~ A, A are assumed. The first value ~ A stands for no fall presumption and the second value A stands for a fall presumption.

With the aid of the formula given above, a probability which is expressed by the variable p _F can be easily estimated with high reliability, which indicates whether there is actually a fall or not. A reliable distinction between fall and no fall can be made with this formula, although a prior knowledge of a fall or non-fall frequency is rather inaccurate. The prior knowledge is represented by the variable Q - the quotient of the known first frequency of falls and the known second frequency of fall presumption alarms in the event that there is no fall - and these known frequencies are, for example, from previous knowledge, experience about fall frequencies, in particular older people, etc. won. However, experiments in which the formula was tested to determine the third likelihood have shown, for example, that even if the variable Q is varied over four orders of magnitude, the determined third likelihood of actual lintfall is greater than about 99% of all observed falls at about 99 % and less than 1% in more than 90% of all non-falls observed.

An expedient development of the invention provides that on the basis of the third probability for the presence of a fall, an alarm message is generated and output. As a result, only an alarm message z. B. issued at an auxiliary service when there is actually a fall. This z. B. false alarms due to z. B. arm movements, etc. avoided. In addition, by generating and outputting the alarm even then help can be brought in a simple way or notified, for example, if a fallen person can no longer trigger an alarm, as they are z. B. is unconscious.

It is advantageous if the sensor measures an acceleration vector which is converted into acceleration values in scalar form, wherein in the case of a fall, the scalar form of the acceleration values is assigned the value 1, and otherwise, if there is no fall, the scalar form of the acceleration values (a) the value -1 is assigned. By converting the multidimensional acceleration vector into one-dimensional acceleration values, a probability density function (i.e., the distribution of falls and non-falls in the measured acceleration values) can be easily determined. Thus, the ratio between the first probability for the occurrence of the measured acceleration values in the event of a fall and the second probability for the occurrence of the measured acceleration values in the event of no fall is much easier to estimate than for acceleration values in the form of a multidimensional vector or at an acceleration vector. In the case of a fall, values of 1 or approximately 1, and in the case of a non-fall, values of -1 or approximately -1 should be assumed for the scalar-converted acceleration values, distributions of these values for falls and non-falls from so-called training data (eg, prior knowledge, experience about falls, test data, etc.) can be determined.

In a further embodiment of the method according to the invention, it is advantageous if a conversion of the acceleration vector into the acceleration values is carried out in scalar form with the aid of neural networks.

In a preferred embodiment of the invention, the sensor is designed as a three-axis acceleration sensor. An acceleration sensor is used to measure an acceleration such that an acting inertia force is determined and from this it can be deduced whether a speed increase or decrease is present. By means of a three-axis acceleration sensor, speed increases or decreases in the direction of the three coordinate axes of a coordinate system are measured, so that acceleration values or falls in each direction (eg forward, lateral, backward) can be measured or detected.

An expedient development of the invention also provides that the sensor is integrated into a wristwatch. Because of this, no additional components (eg sensor on the belt or on the wrist, etc.) are necessary for the fall detection. The detection of a fall and optionally the output of an alarm is done in a simple and convenient way by wearing the appropriate wristwatch, this z. B. can be worn at night.

Brief description of the drawing

The invention will now be described by way of example with reference to the accompanying figure. 1 shows an example of a sequence of the method according to the invention with reference to a schematic and exemplary basic structure of a fall detector.

Embodiment of the invention

In 1 is shown schematically and exemplarily a basic structure of a fall detector for carrying out the method according to the invention.

The fall detector comprises a sensor BS for measuring acceleration values a, wherein the sensor BS is designed, for example, as a three-axis acceleration sensor BS. Furthermore, the fall detector comprises by way of example a component FA for determining fall presumption alarms f and a component RA for estimating a ratio R from a first or second probability for an occurrence of the measured acceleration values a in the event of a fall or no fall. The fall detector also includes, by way of example, a component PA with which a third probability p _F for an actual occurrence of a fall is estimated.

The fall detector can z. B. integrated in a bracelet, a wristwatch or in a wrist-worn alarm device for home emergency calls.

To determine a fall are then in a first step 1 Acceleration values a detected by the acceleration sensor BS. When using a triaxial acceleration sensor BS z. For example, the acceleration values a are measured in the form of a multi-dimensional acceleration vector. For easier further processing and for easier estimation of the third probability p _F for an actual occurrence of a fall, the multidimensional acceleration vector is converted into scalar form, for example with the aid of neural networks, into acceleration values a. This conversion is carried out, for example, such that the acceleration values a in the case of a fall assume the value 1 or values which are very close to the value 1. In the case where there is no fall, the acceleration values a are assumed to be -1 or values which are very close to the value -1. A distribution of the acceleration values a for falls or for non-falls can be determined for example on the basis of training data.

The measured acceleration values a are then forwarded to the component FA for determining fall presumption alarms f. In parallel, the acceleration values a are also transmitted to the component RA for estimating a ratio R between a first and second probability for occurrence of the measured acceleration values a in a fall or no fall.

In a second process step 2 A fall presumption alarm f is then determined by the component FA on the basis of the acceleration values a, wherein two values A and ~ A can be assumed from the fall presumption alarm f. The value A stands for an alarm, which in case of a fall with z. B. subsequent immobility should occur. The value ~ A means that there is no fall at the measured acceleration values a. At the second process step 2 - So when determining the fall presumption alarm f - false alarms are allowed. Ie. From the fall presumption alarm f, the value A can be assumed even if there is no fall.

If the fall presumption alarm f has the value A, then in a third method step 3 estimated by the component RA a ratio R. The ratio R is a quotient of a first probability for an occurrence of the measured acceleration values a in the event of a fall and a second probability for an occurrence of the measured acceleration values a in the event that there is no fall. Decisive for a fall detection is that, for example, this ratio R in the case of a fall by a few orders of magnitude greater than the value 1 and in the case of a non-fall very much smaller than the value 1. This is achieved in particular by the conversion of the acceleration values a into scalar form.

In a fourth process step 4 become the second step 2 determined fall presumption alarm f and in the third step 2 estimated ratio R is passed to the component PA for estimating the third probability p _F for the actual presence of a fall. By means of this component PA are in the fourth step 4 the fall presumption alarm f and the ratio R are combined with a first known frequency HF of falls and with a second known frequency HA of fall victim alarms without fall (eg false alarm, etc.). First, a quotient Q ≈ HF / HA is formed from the first frequency HF and the second frequency HA. The known first frequency HF and the known second frequency HA are determined on the basis of prior knowledge of frequencies with which, for example, falls or possible fall presumption alarms without falling, in particular false alarms, can occur.

die dritte Wahrscheinlichkeit p_F für ein tatsächliches Vorliegen eines Sturzes abgeschätzt. Im Experiment wurde dabei beispielsweise festgestellt, dass das Produkt Q*R im Fall eines Sturzes sehr viel größer als der Wert 1 ist. Für den Fall, dass kein Sturz vorliegt, wurde vom Produkt Q*R ein Wert angenommen, welche viel kleiner als der Wert 1 war.Then in the fourth process step 4 using a formula

estimated the third probability p _F for an actual occurrence of a fall. In the experiment, it was found, for example, that the product Q * R is much larger than the value 1 in the event of a fall. In the case of no fall, the product Q * R was assumed to be much smaller than the value of 1.

In a fifth process step 5 is then derived from the estimated third probability p _F , whether there is actually a fall or the fall presumption alarm f represents a possible false alarm. In the event that a fall is actually determined based on the value of the estimated third probability p _F , for example, an alarm message can be generated and z. B. be routed via a house emergency call to an auxiliary device.

Claims (7)

Method for detecting falls by means of a sensor (BS), characterized in that acceleration values (a) are measured by the sensor (BS) and a fall presumption alarm (f) is determined therefrom ( 1 . 2 in that on the basis of the measured acceleration values (a) a ratio (R) between a first probability for an occurrence of the measured acceleration values (a) during a fall and a second probability for an occurrence of the measured acceleration values (a) without a fall is estimated ( 3 ) that then the fall presumption alarm (f) and the ratio (R) are combined with a known first frequency (HF) of falls and a known second frequency (HA) of fall victim alarms without fall ( 4 ) and then from a combination of fall presumption alarm (f), ratio (R), known first frequency (HF) and known second frequency (HA) a third probability (p _F ) for an actual occurrence of a fall is estimated ( 5 ). A method according to claim 1, characterized in that the third probability (p _F ) for an actual occurrence of a fall on the basis of the formula

is estimated ( 5 ), wherein a variable R is used to estimate the ratio between a first probability of an occurrence of the measured acceleration values (a) in the event of a fall and a second probability of occurrence of the measured acceleration values (a) without a fall, by a variable Q an estimate a quotient of the known first frequency (HF) of falls and the known second frequency (HA) of fall victim alarms (f) without fall and by a variable f a fall presumption alarm (f), of which two values ~ A, A are assumed where a first value ~ A stands for no fall presumption and a second value A for a fall presumption. Method according to one of claims 1 to 2, characterized in that on the basis of the third probability (p _F ) for the presence of a Fall an alarm message is generated and output ( 5 ). Method according to one of claims 1 to 3, characterized in that an acceleration vector is measured by the sensor, and that then the measured acceleration vector is converted into acceleration values (a) in scalar form, wherein in the case of a fall, the scalar shape of the Acceleration values (a) is assigned the value 1, and otherwise, if there is no fall, the scalar form of the acceleration values (a) is assigned the value -1. A method according to claim 4, characterized in that a conversion of the acceleration vector into the acceleration values (a) in scalar form by means of neural networks is performed. Method according to one of claims 1 to 5, characterized in that the sensor (BS) is designed as a triaxial acceleration sensor. Method according to one of claims 1 to 6, characterized in that the sensor (BS) is integrated into a wristwatch.

Images