System for automatic fall detection for elderly people

1. Apparatus for detection of human falls, comprising:

an acceleration detector, for detecting vibration events, the acceleration detector attached to a floor, and configured to sense low amplitude vibrations with accelerations up to 2 g;

a microphone, for placing in association with said acceleration detector for detection of sound events, and

a classification unit comprising an acceleration input for input of data relating to said vibration events from said acceleration detector and a sound input for input of data relating to said sound events from said microphone, the unit configured to classify separate events from said microphone and said acceleration detector respectively as concurrent events from said microphone and said acceleration detector, thereby to determine whether a human fall is indicated, wherein said microphone has a frequency range of substantially 20-16000 and an S/N ratio in excess of 58 dB.

2. The apparatus of claim 1 , further comprising an alarm unit, associated with said classification unit for providing an alarm output when said human fall is indicated.

3. The apparatus of claim 1 , wherein said acceleration detector has a sensitivity to vibration of substantially 1V/g.

4. Apparatus for detection of human falls, comprising:

an acceleration detector, for detecting vibration events, the acceleration detector attached to a floor, and configured to sense low amplitude vibrations with accelerations up to 2 g;

a microphone, for placing in association with said acceleration detector for detection of sound events, and

a classification unit comprising an acceleration input for input of data relating to said vibration events from said acceleration detector and a sound input for input of data relating to said sound events from said microphone, the unit configured to classify separate events from said microphone and said acceleration detector respectively as concurrent events from said microphone and said acceleration detector, thereby to determine whether a human fall is indicatedwherein said concurrent events comprise vibration events, and sound events, said classification unit being configured to extract features from said events, and to compare a distribution of said features with prestored features of learned events, thereby to classify said events as belonging to a human fall or not belonging to a human fall, and wherein said extracted features comprise sound event energy and sound event length and Mel frequency cepstral coefficients, and vibration event energy, and vibration event length.

5. The apparatus of claim 4 , wherein said classification unit is configured to extract from said sound events a sound event length feature and a sound event energy feature.

6. The apparatus of claim 4 , wherein said classification unit is configured to extract from said vibration events a vibration event length and a vibration event energy.

7. The apparatus of claim 4 , wherein said extracted features comprise seventeen features.

8. Method for detection of human falls, comprising:

detecting low amplitude vibration events in a floor with accelerations of up to 2 g,

detecting sound events,

classifying vibration and sound events detected at the same time as concurrent vibration and sound events, and

using said classified concurrent events to determine whether a human fall is indicated wherein said classifying comprises extracting features from said events, and comparing a distribution of said features with prestored features of learned events, thereby to classify said events as belonging to a human fall or not belonging to a human fall; and

wherein said extracted features comprise sound event energy and sound event length and at least one member of the group consisting of shock response spectra, Mel frequency cepstral coefficients, vibration event energy, and vibration event length;

the method further comprising detecting sound events using a microphone, wherein said microphone has a frequency range of substantially 20-16000 Hz and an S/N ratio in excess of 58 dB.

9. The method of claim 8 , further comprising providing an alarm output when said human fall is indicated.

10. The method of claim 8 , comprising detecting said vibration events using a sensitivity of substantially 1V/g.

11. The method of claim 8 , wherein said classifying comprises extracting from said sound events a sound event length feature and a sound event energy feature.

12. The method of claim 8 , wherein said classifying comprises extracting from said vibration events a vibration event length and a vibration event energy.

13. The method of claim 8 , wherein said extracted features comprise shock response spectra, Mel frequency cepstral coefficients, vibration event energy, vibration event length, sound event energy and sound event length.

14. The method of claim 11 , wherein said extracted features comprise seventeen features.