Fall detection using machine learning
A method and system for fall detection using machine learning are disclosed. The method comprises detecting at least one signal by a wireless sensor device and calculating a plurality of features from the at least one detected signal. The method includes training a machine learning unit of the wireless sensor device using the features to create a fall classification and a non-fall classification for the fall detection. The system includes a sensor to detect at least one signal, a processor coupled to the sensor, and a memory device coupled to the processor, wherein the memory device includes an application that, when executed by the processor, causes the processor to calculate a plurality of features from the at least one detected signal and to train a machine learning unit of the wireless sensor device using the features to create a fall classification and a non-fall classification for the fall detection.
1. A method for fall detection using machine learning, the method comprising:
detecting a first signal by a wireless sensor device, wherein the first signal includes a first acceleration sample detected by a tri-axial accelerometer of the wireless sensor device during a first time period of a user fall process;
detecting a second signal by a wireless sensor device, wherein the second signal includes a second acceleration sample detected by a tri-axial accelerometer of the wireless sensor device during a second time period of a user non-falling process;
obtaining a calibration vector from the first and second acceleration samples by receiving notification of posture information at the wireless sensor device;
calculating a first feature of the first time period using the detected first signal and the calibration vector;
calculating a second feature of the second time period using the detected second signal and the calibration vector;
initiating a support vector machine within an application of the wireless sensor device using a training data set that includes data of an intentional fall process and data of an intentional user non-falling process to determine a plurality of coefficients;
calculating a linear combination of features using the determined plurality of coefficients and the calculated first feature and the calculated second feature; and
detecting a fall of a user of the wireless sensor device using the linear combination of features.
2. The method of claim 1 , wherein the calculating is carried out during both periods of falls and activities of daily living.
3. The method of claim 1 , further comprising:
adapting the SVM using online user data submitted by a plurality of wireless sensor devices.
4. The method of claim 1 , further comprising:
calculating a severity metric of the detected fall using the linear combination of the features.
5. The method of claim 4 , further comprising:
determining whether the user is upright within a predetermined time period after the detected fall.
6. The method of claim 4 , further comprising:
reporting vital signs of the user before and after the detected fall.
7. The method of claim 1 , wherein the calibration vector is obtained once upon initialization.
8. A system for fall detection using machine learning, the system comprising:
a sensor to:
detect a first signal, wherein the first signal includes a first acceleration sample detected by a tri-axial accelerometer during a first time period of a user fall process;
detect a second signal, wherein the second signal includes a second acceleration sample detected by a tri-axial accelerometer during a second time period of a user non-falling process;
a processor coupled to the sensor;
a memory device coupled to the processor; and
an application coupled to the memory device, wherein the application, when executed by the processor, causes the processor to:
obtain a calibration vector from the first and second acceleration samples by receiving notification of posture information at the wireless sensor device;
calculate a first feature of the first time period using the detected first signal and the calibration vector;
calculate a second feature of the second time period using the detected second signal and the calibration vector;
initiate a support vector machine within the application using a training data set that includes data of an intentional fall process and data of an intentional user non-falling process to determine a plurality of coefficients;
calculate a linear combination of features using the determined plurality of coefficients and the calculated first feature and the calculated second feature; and
detect a fall of a user of the wireless sensor device using the linear combination of features.
9. The system of claim 8 , wherein the calculating is carried out during both periods of falls and activities of daily living.
10. The system of claim 8 , wherein the support vector machine (SVM) is a machine learning unit.
11. The system of claim 10 , wherein the application, when executed by the processor, further causes the processor to:
adapt the SVM using online user data submitted by a plurality of wireless sensor devices.
12. The system of claim 8 , wherein the application, when executed by the processor, further causes the processor to:
calculate a severity metric of the detected fall using the linear combination of the features.
13. The system of claim 12 , wherein the application, when executed by the processor, further causes the processor to:
determine whether the user is upright within a predetermined time period after the detected fall.
14. The system of claim 12 , wherein the application, when executed by the processor, further causes the processor to:
report vital signs of the user before and after the detected fall.