WO2016155789A1

WO2016155789A1 - Fall detection system and method

Info

Publication number: WO2016155789A1
Application number: PCT/EP2015/056991
Authority: WO
Inventors: Salvatore Longo
Original assignee: Nec Europe Ltd.
Priority date: 2015-03-31
Filing date: 2015-03-31
Publication date: 2016-10-06

Abstract

A fall detection system, comprises a first sensor installation (1), a second sensor installation (3), and a sensor signal analysis component (10) for analyzing measurements of said first and said second sensor installation (1; 3), wherein said first sensor installation (1) includes at least one presence sensor (3) that is configured to detect the presence of a person (5) within a predefined coverage area, wherein the detection of the presence of a person (5) by said first sensor installation (1), alone or in combination with sensor signal patterns with predefined characteristics measured by said first sensor installation (1), triggers measurements of said second sensor installation (3), wherein said second sensor installation (3) includes at least one distance sensor (4) that is configured to detect the height of a person (5) above a predefined ground level, and wherein said sensor signal analysis component (10) is configured to output a fall event in case the output of said at least one distance sensor (4) is below a predefined threshold (6). Furthermore, a corresponding method for detecting falls is disclosed.

Description

FALL DETECTION SYSTEM AND METHOD

The present invention generally relates to fall detection systems and to methods for detecting falls.

Basically, a fall can be defined as an event, which results in a person coming to rest inadvertently on the ground or floor or other lower level. Fall-related injuries may be fatal or non-fatal, though most are non-fatal. For example, children in the People's Republic of China, for every death due to a fall, there are 4 cases of permanent disability, 13 cases requiring hospitalization for more than 10 days, 24 cases requiring hospitalization for 1-9 days and 690 cases seeking medical care or missing work/school (for reference, see http://www.who.int/mediacentre/ factsheets/fs344/en/).

In particular, falls are one of the most common causes of injury among older adults. For example for the Canadian healthcare system, the falling injury costs in total over $6.2 billion in 2004 alone (for reference, see SMARTRISK (2009). The Economic Burden of Injury in Canada. SMARTRISK: Toronto, ON, ISBN 1 - 894828-50-X). Approximate 25-35% of elderly residents experienced fall-related injury more than one time per year. Nearly 30-40% of all falls visited an emergence room for treatment and needed to be hospitalized (for reference, see K. Brewer, C. Ciolek, M. F. Delaune: "Falls in community dwelling older adults: Introduction to the problem", APTA Continuing Education Series, pp. 38-46, July 2007). The demand for surveillance systems, especially for fall detection, has increased within the healthcare industry with the rapid growth of the population of the elderly in the world.

Detecting an unintentional fall is difficult due to the subtle and the complex nature of the body movement. Although today many mechanisms are available for detecting falls, the efficiency of fall detection and the reliability of posture recognition are always challenges.

Another important aspect that should be considered is the privacy. Many current solutions are based on camera installation. As well-known, however, cameras cannot be installed in various places due to certain people privacy laws. For these cases, there are several alternative solutions on the market like, e.g. wearable sensors that are trained to detect falls based on various accelerometer measurements.

In the state of the art fall detection methods can be divided roughly into three categories: wearable device based, ambience sensor based and camera (vision) based, as depicted in the classification overview shown in Fig. 1 , which is taken from M. Mubashir et al.: "A survey on fall detection: Principles and approaches", in Neurocomputing, vol. 100, January 2013, p. 144-152. In this document, the working principles of the existing methods of the different categories are described as follows:

Wearable device based approaches

"Wearable device based approaches rely on garments with embedded sensors to detect the motion and location of the body of the subject. [...] Technological developments have yielded devices that can measure activities using accelerometers. Accelerometry is composed of measure of acceleration of the body or parts of the body. It is one of the most extensively-used methods implemented for measuring physical activities to monitor activity patterns. [...] Physiological responses such as varying heart rate or blood pressure may result from physical activity and changes in body position. That makes the assessment of motion and posture a key factor in an ambulatory monitoring environment. [...] Accelerometry provides detailed information on behavior such as physical activity and inactivity. This information can be used to measure more comprehensive relationships among movement frequency, intensity and duration. [...] Tri-axial accelerometers are designed for simultaneous detection of acceleration in three axial directions."

Wearable devices have their advantages as well as disadvantages. The biggest advantage remains the cost efficiency of wearable devices. Installation and setup of the design is also not very complicated. Therefore, the devices are relatively easy to operate. The disadvantages include intrusion and fixed relative relations with the object, which could cause the device to be easily disconnected. Such disadvantages make wearable devices an unfavorable choice for the elderly.

Ambience based devices

"Ambience based devices attempt to fuse audio and visual data and event sensing through vibrational data. [...] The detection of events and changes using vibrational date can be useful in many ways such as monitoring, tracking, localization etc."

Most ambient device based approaches use pressure sensors for object detection and tracking. The pressure sensor is based on the principle of sensing high pressure of the object due to the object's weight for detection and tracking. It is very cost effective and less intrusive for the implementation of surveillance systems. However, it has a big disadvantage of sensing pressure of everything in and around the object and generating false alarms in the case of fall detection, which leads to a low detection accuracy.

Video Solutions

"Cameras are increasingly included, these days, in in-home assistive/care systems as they convey multiple advantages over other sensor-based systems. Cameras can be used to detect multiple events simultaneously with less intrusion. [...] Vision based systems tend to deal with intrusion better than other approaches. Recent research in computer vision on surveillance indeed provides a practical and complex framework."

The biggest advantage of camera solutions is the high accuracy of such technology. However, this solution has also the disadvantages to be expensive as solution and also privacy intrusive.

In view of the above it is an object of the present invention to improve and further develop a fall detection system and method in such a way that it is less cost intensive than existing camera solutions and, at the same time, privacy preserving as well as not bound to individual persons as the existing wearable solutions are. In accordance with the invention, the aforementioned object is accomplished by a system comprising the features of claim 1. According to this claim, such a system comprises a first sensor installation, a second sensor installation, and a sensor signal analysis component for analyzing measurements of said first and said second sensor installation, wherein said first sensor installation includes at least one presence sensor that is configured to detect the presence of a person within a predefined coverage area, wherein the detection of the presence of a person by said first sensor installation, alone or in combination with sensor signal patterns with predefined characteristics measured by said first sensor installation, triggers measurements of said second sensor installation, wherein said second sensor installation includes at least one distance sensor that is configured to detect the height of a person above a predefined ground level, and wherein said sensor signal analysis component is configured to output a fall event in case the output of said at least one distance sensor is below a predefined threshold.

Furthermore, the aforementioned object is accomplished by a method comprising the features of claim 6. According to this claim, such a method comprises, by means of a first sensor installation that includes at least one presence sensor, detecting the presence of a person within a predefined coverage area, wherein the detection of the presence of a person by said first sensor installation, alone or in combination with sensor signal patterns with predefined characteristics measured by said first sensor installation, triggers measurements of a second sensor installation, by means of said second sensor installation that includes at least one distance sensor, detecting the height of a person above a predefined ground level, and by means of a sensor signal analysis component, analyzing measurements of said first and said second sensor installation and outputting a fall event in case the output of said at least one distance sensor is below a predefined threshold.

The main idea is to use a sensor fusion system for detecting falls in an indoor scenario. The usage of distance sensors in combination with different sensors for triggering a distance sensor measurement helps to create a solution that can be used for detecting falls in various scenarios where cameras cannot be installed. Specifically, the presence sensor is used for triggering the measurement of the distance sensor. Generally, such a combination or fusion of data or information coming from the distance sensors with other sensors' data goes into the direction of "Virtual Sensing".

The present invention provides a solution based on sensor fusion that is relatively cheap compared to camera solutions and is privacy preserving compared to vision solutions. Therefore, the system according to the present invention can be used in many scenarios where cameras are not allowed, for instance in hospitals, airports, toilets or patient rooms, to name just a few possible application scenarios.

In addition to the above, the system and method according to the present invention are as generic as possible and not targeted for a certain type of person or related to a single subject like the wearable sensors.

According to an embodiment the presence sensors are infrared sensors detecting infrared radiation.

According to an embodiment the distance sensors are sensors using infrared ray beams, or ultrasonic sensors using transmitting and receiving of ultrasonic waves, or combinations of the two kinds of sensors.

For obtaining accurate distance measurements in terms of a person's height above a predefined ground level, typically the floor of an indoor location, and for obtaining a sufficiently large coverage area, according to an embodiment the presence and the distance sensors are mounted within an upper part of the indoor location, preferably at the ceiling of the indoor location.

According to an embodiment of the sensor signal analysis component includes a machine learning tool for being applied to measurements of the presence sensors and/or the distance sensors in order to reduce the number of false positives. In this context it may be provided that a classification of possible fall events is performed by way of processing measurements of the at least one presence sensor by means of the machine learning tool. This machine learning tool may be configured to apply supervised, unsupervised and/or semi supervised machine learning techniques. According to an embodiment it may be provided that the measurements of the distance sensors are employed for validating the classification results from the machine learning tool.

According to an embodiment it may be provided that the distance sensors record the temporal distribution of the height of a person located within the coverage area of the presence sensors.

There are several ways how to design and further develop the teaching of the present invention in an advantageous way. To this end it is to be referred to the patent claims subordinate to patent claim 1 and 6 on the one hand and to the following explanation of preferred embodiments of the invention by way of example, illustrated by the figure on the other hand. In connection with the explanation of the preferred embodiments of the invention by the aid of the figure, generally preferred embodiments and further developments of the teaching will we explained. In the drawing

Fig. 1 is a diagram showing a classification of prior art fall detection methods,

Fig. 2 is a schematic view of the sensor components of a fall detection system in accordance with an embodiment of the invention,

Fig. 3 is a schematic view illustrating the functional principle of a presence sensor installation in accordance with an embodiment of the invention,

Fig. 4 is a schematic view of a distance sensor installation in accordance with an embodiment of the invention,

Fig. 5 is a schematic view illustrating a stable distance measurement in accordance with an embodiment of the invention, Fig. 6 is a schematic view of the sensor components together with the sensor signal analysis component of a fall detection system in accordance with an embodiment of the invention, and

Fig. 7 is a diagram illustrating the measurement principle of a fall detection system in accordance with an embodiment of the invention.

Fig. 2 is a schematic view of the sensor components of a fall detection system in accordance with an embodiment of the invention. The example relates to an indoor scenario in which the relevant sensors of the system are mounted at the ceiling of the respective indoor location. Specifically, in the illustrated embodiment, the indoor fall detection system comprises a first sensor installation 1 (including a presence sensor 2) and a second sensor installation 3 (including two distance sensors 4).

The presence sensor 2 is an infrared presence sensor that detects the changes in infrared radiation, which occurred when a person moves into the detector's 2 coverage area, indicated by the cone in Fig. 2.

As the distance sensors 4, it is preferable to use either infrared distance sensors, or ultrasonic sensors that work by transmitting and receiving ultrasonic waves, or combinations of these two kinds of sensors.

According to an embodiment, a measurement of the distance sensors 4, for detecting the height of a person 5 and the changes of it over time, is triggered by the presence sensor 2. Activating the distance sensors 4 only in case of a person 5 being detected in the area by means of the presence sensor 2 enables a precise measurement of the person's 5 height in the indoor location. The trigger point may be chosen to be either the mere detection of a person's 5 presence in the area, or the detection of a person's 5 presence in the area together with a signal structure or pattern of the presence sensor 2 that fulfills certain criteria being indicative for a possible fall, as will be explained in detail below. In any case, a threshold 6 (indicated by the dotted line) is defined for the distance measurement, as will be explained in more detail in connection with Fig. 5 below.

The functional principle of a sensor that can be used as the presence sensor 2 in a fall detection system according to the present invention is illustrated in Fig. 3, which shows the Phidgets 1 1 1 1 _0 - Motion Sensor. Based on infrared radiation, this sensor detects temperature differences and is, thus, well suited to detect the presence and also the motion of people within a cone shaped coverage area by the persons' body temperature.

According to an embodiment of the invention, a measurement scenario could be implemented as follows:

In a first step, the presence sensors 2 are used for detecting the presence of a person 5 in a predefined surveillance area that is covered by the sensor installation 1 , 3. The sensor choice is not bounded to any specific sensor; however, one solution could be to use a motion sensor for detecting the person 5 walking inside an infrared cone. The detection will trigger a distance sensor 4 measurement. In this embodiment the distance sensor 4 measurement is only activated when a possible falling, indicated by the presence sensor 2 measurements, happens.

In a second step, after receiving the measurement from the presence detection sensor 2, the distance sensor 4 measurement is triggered. This measurement will calculate the height of the "object" inside the infrared cone, as will be described in more detail in connection with Fig. 5 below. For such solution a threshold 6 with respect to the measured object height is specified for deciding when a possible fall is detected.

In a third step, information from the presence sensor 2 plus the distance sensor 4 can be also sent to a machine learning module 7 that will use such information to reduce the number of false positives. Fig. 4 is a schematic view of a distance sensor installation 3 in accordance with an embodiment of the invention. Specifically, Fig. 4 depicts a distance sensor installation 3 in a corridor, where the distance sensors 4 are placed in a row for covering the entire area of the corridor. Generally, an important pre-step that has to be considered carefully is the study of the feasibility of the solution in the respective indoor environment. In particular, the questions where to place the distance sensors 4 and how to install the distance sensors 4 is very important for completely covering a desired surveillance area and for getting accurate solution estimation. For instance, for a correct measurement of the height of a person inside the sensor's (e.g. infrared or ultrasonic) measurement cone, most suitably the distance sensors 4 are placed on the ceiling 8. Another important aspect to be considered is the fact that the measurement cone should be, as far as possible, free of obstacles, like chairs, lamps, sofa etc. To some extent this might limit where such installation can be used; generally, however, the better the condition is fulfilled, the more accurate will be the fall detection results.

Fig. 5 is a schematic view that illustrates a distance measurement in accordance with an embodiment of the invention. In the illustrated scenario it is assumed that the distance sensor 4 measurement is triggered by the presence sensor (not shown), as described already above. The distance sensor 4 is configured to measure the shortest distance of an object within its coverage area, illustrated by the cone in Fig. 5, for instance, by emitting ultrasonic or microwave signals, which are reflected by objects, and detecting the corresponding reflected signals. From the time difference between the emission of a signal and the detection of the corresponding reflected signal, the distance of the reflecting object from the sensor 4 can be calculated, as well known in the art. In the scenario of Fig. 5, a person 5 moving within the detection cone in an upright position is detected by the distance sensor 4, since the person's 4 head is assumed to be the object in shortest distance to the distance sensor 4. Consequently, the distance sensor 4 outputs stable measurement results.

As also shown in Fig. 5, a threshold 6 (indicated by the dotted line) is defined for the distance measurement. This threshold 6 defines a configurable distance threshold, i.e. a distance from the distance sensor 4, wherein the occurrence of a fall event is assumed if the threshold 6 is exceeded by the distance measurements. The distance threshold 6 may be specifically adapted to different scenarios; however, typically this threshold 6 will be specified in such a way that it defines a level with a characteristic height above a ground level, i.e. the floor 9 of the respective indoor environment, as shown in Fig. 5. A characteristic height above the floor 9 that is typically not exceeded by a person's body that is lying on the floor after a fall has happened can be assumed to be in the range of e.g. 0.5 - 0.7 m.

According to a preferred embodiment, the sensor data analysis also considers how the height measurement changes over the time. For example, if one sees that the height measurement changes from a high level to a low level under the threshold 6 in a fast time frame, corresponding to a fast movement of a person from up to down, in this case there is high confidence that a possible falling is happening. Alternatively or additionally, the time interval between two subsequent distance sensor 4 measurements can be used to further advance the proposed solution. This could be calculated directly by a component on top of the sensor measurements and can be used for validate the fall detection.

Fig. 6 is a schematic view of the sensor components together with the sensor signal analysis component 10 of a fall detection system in accordance with an embodiment of the invention. The lower part of Fig. 6 corresponds to the scenario illustrated in and discussed in connection with Fig. 2. Basically, regarding the usage of this fall detection system it is possible to differentiate between two major usage modes:

Mode 1 : As shown in Fig. 6 the data coming from the presence sensors 2 will trigger the distance sensor 4 measurement. This will be reported to a result analysis tool 1 1 of the signal analysis component 10 that based on the measurement will detect the possible falls and will generate appropriate outputs, e.g. in form of alarms.

Mode2: For reducing the amount of false positives, the data from the presence sensors 2 (e.g. motion sensors) are sent in real time to a backend server 12 of the signal analysis component 10 that will process them and deternnine the type of activity in the area. The backend server 12 is the core component where the data are processed. The motion sensor data are used for detecting if there are more people passing by the sensor and if there is a falling case. A feature extraction tool 13 of the signal analysis component 10 receives the data from a front-end data collector, computes the features and creates the dataset that will be used by a classifier 14 for predicting possible falls.

As shown in the embodiment of Fig. 6, the classifier 14 can be a pre-trained machine learning classification module 15 that based on the input dataset produces a classified output. In this case, the choice of the machine learning (ML) algorithm will influence the classification accuracy. The usage of supervised machine learning could be associated to such solution for detecting the various activities in the monitored area and for reducing the number of false positives. For example, a supervised machine learning approach can be used for distinguishing false falling activities, e.g. people who just pick up objects from the ground. It is important to note that the present invention is independent from the type of machine learning technique used (e.g. Supervised, Unsupervised or Semi Supervised).

The result analysis tool 1 1 will analyze the classifier 14 results by validating the ML results with the measurements coming from the distance sensors 4, and it will generate an alarm in case of falls.

Fig. 7 is a diagram illustrating the measurement principle of a fall detection system in accordance with an embodiment of the invention. As illustrated at 701 , the measurement starts with a measurement from a presence sensor 2. If this sensor detects the presence of a person (illustrated in 702), an active distance sensor 4 measurement is triggered at step 703. The trigger point may be selected to be in the mere detection of the presence of a person by the presence sensor 2. Attentively, it may be provided that the data measured by the presence sensor 2 is preprocessed, e.g. by applying machine learning techniques, and the distance sensor 4 measurement is only triggered if the presence sensor 2 measurements yields a signal structure or pattern that is indicative for a possible fall. In step 704, the signal analysis component 10 checks whether the distance measured by the distance sensor 4 is above the predefined distance threshold 6. If this is the case, in step 705, this measurement is transferred to the result analysis tool 1 1 , which outputs a fall event (at step 706). On the other hand, if the measured distance is below the predefined distance threshold 6, at step 707, this is interpreted as being indicative for a "no fall" event, and the measurement data is transferred for further analysis to a machine learning module 15. Here, shown at 708, a machine learning classification model is applied in order to classify the event. In the context of this analysis, the human behavior that is normal or typical for the selected area may be taken into consideration. If the results of the machine learning module 15 are indicated for a fall event, a respective output is generated at step 706, otherwise the system returns to step 701 and continues to look out for presence as detected by the presence sensor 2.

In addition to the described embodiments, combinations of embodiments of the invention can be used with:

• A variety of interfaces including but not limited to Web Services, REST APIs, remote method execution and others

• In combinations of multiple deployments of the system that provides generalizable training sets and better accuracy

• In combinations of deployments originally intended for different purposes where sensor data can be reused

Many modifications and other embodiments of the invention set forth herein will come to mind the one skilled in the art to which the invention pertains having the benefit of the teachings presented in the foregoing description and the associated drawings. Therefore, it is to be understood that the invention is not to be limited to the specific embodiments disclosed and that modifications and other embodiments are intended to be included within the scope of the appended claims. Although specific terms are employed herein, they are used in a generic and descriptive sense only and not for purposes of limitation.

Claims

C l a i m s

1. Fall detection system, comprising:

a first sensor installation (1 ), a second sensor installation (3), and a sensor signal analysis component (10) for analyzing measurements of said first and said second sensor installation (1 ; 3),

wherein said first sensor installation (1 ) includes at least one presence sensor (3) that is configured to detect the presence of a person (5) within a predefined coverage area, wherein the detection of the presence of a person (5) by said first sensor installation (1 ), alone or in combination with sensor signal patterns with predefined characteristics measured by said first sensor installation (1 ), triggers measurements of said second sensor installation (3),

wherein said second sensor installation (3) includes at least one distance sensor (4) that is configured to detect the height of a person (5) above a predefined ground level, and

wherein said sensor signal analysis component (10) is configured to output a fall event in case the output of said at least one distance sensor (4) is below a predefined threshold (6).

2. System according to claim 1 , wherein said presence sensors (2) are infrared sensors detecting infrared radiation.

3. System according to claim 1 or 2, wherein said distance sensors (4) are sensors using infrared ray beams, or ultrasonic sensors using transmitting and receiving of ultrasonic waves, or combinations of said two kinds of sensors.

4. System according to any of claims 1 to 3, wherein said presence and said distance sensor (2; 4) are mounted at the ceiling (8) of an indoor environment.

5. System according to any of claims 1 to 4, wherein said sensor signal analysis component (10) includes a machine learning tool (15) for being applied to measurements of said presence sensors (2) and/or said distance sensors (4) in order to reduce the number of false positives.

6. Method for detecting falls, in particular by employing a system according to any of claims 1 to 5, comprising:

by means of a first sensor installation (1) that includes at least one presence sensor (2), detecting the presence of a person (5) within a predefined coverage area, wherein the detection of the presence of a person (5) by said first sensor installation (1 ), alone or in combination with sensor signal patterns with predefined characteristics measured by said first sensor installation (1), triggers measurements of a second sensor installation (3),

by means of said second sensor installation (3) that includes at least one distance sensor (4), detecting the height of a person (5) above a predefined ground level, and

by means of a sensor signal analysis component (10), analyzing measurements of said first and said second sensor installation (1 ; 3) and outputting a fall event in case the output of said at least one distance sensor (4) is below a predefined threshold (6).

7. Method according to claim 6, wherein a classification of possible fall events is performed by way of processing measurements of said at least one presence sensor (2) by means of a machine learning tool (15).

8. Method according to claim 7, wherein said machine learning tool (15) applies supervised, unsupervised and/or semi supervised machine learning techniques.

9. Method according to any of claims 6 to 8, wherein the measurements of said at least one distance sensor (4) are employed for validating said classification.

10. Method according to any of claims 6 to 9, wherein said at least one distance sensor (4) records the temporal distribution of the height of a person (5) located within the coverage area of said at least one presence sensor (4).