Fall detection system using a combination of accelerometer, audio input and magnetometer
A wearable device for detecting a user state is disclosed. The wearable device includes an accelerometer for measuring an acceleration of a user, a magnetometer for measuring a magnetic field associated with the user's change of orientation, a microphone for receiving audio, a memory for storing the audio, and at least one processor communicatively connected to the accelerometer, the magnetometer, the microphone, and the memory. The processor is identified to declare a measured acceleration as a suspected user state, and to categorize the suspected user state based on the stored audio as one of an activity of daily life (ADL), a confirmed user state, or an inconclusive event.
1. A non-transitory computer readable storage medium including instructions that, when executed by a processor, cause the processor to
perform operations to:
receive a set of measurement data including accelerometer data, magnetometer data and audio data;
receive category information for the set of measurement data, the category information identifying a first set of measurement data categorized as indicative of a suspected fall, based on the accelerometer data, a second set of measurement data categorized as indicative of an activity of daily life (ADL) based on physical movement of users, and a third set of measurement data categorized as an inconclusive event;
train a fall model and train an ADL model using the set of measurement data and the category information; and
wherein the fall model and the ADL model are used to classify unclassified measurement data as indicative of a fall or an ADL, the ADL model classifying at least talking and walking as ADL events, and wherein the fall model is used to re-confirm whether the first set of measurement data categorized as indicative of the suspected fall indicates a fall.
2. The computer readable storage medium of claim 1 , wherein the fall model and the ADL model are Gaussian mixture model (GMM) models.
3. The computer readable storage medium of claim 2 , wherein the GMM models are trained using an expectation maximization algorithm.
4. The computer readable storage medium of claim 1 , wherein the set of measurement data includes prior events classified as fall and ADL events.
5. The computer readable storage medium of claim 1 , wherein the instructions further cause the processor to receive additional measurement data including in-use data and update the fall model and the ADL model using the additional measurement data.
6. The computer readable storage medium of claim 1 , wherein the instructions further cause the processor to implement Gaussian mixture model (GMM) classifiers on a subset of a selected feature space of events from the set of measurement data.
7. The computer readable storage medium of claim 6 , wherein the selected feature space is selected via a sequential forward floating selection (SFFS) algorithm to maximize classifier performance.
8. The computer readable storage medium of claim 1 , wherein the instructions further cause the processor to determine a threshold for classifying an event as a fall, and wherein the unclassified measurement data is classified based on a score for the unclassified measurement data and the threshold.
9. The computer readable storage medium of claim 1 , wherein the set of measurement data includes accelerometer data, magnetometer data, and audio data.
10. The computer readable storage medium of claim 1 , wherein to classify the unclassified measurement data as indicative of a fall or an ADL, the fall model and the ADL model are to determine a re-confirmation or change of classification from an initial classification received from a wearable device that generated the unclassified measurement data.
11. A method comprising:
receiving, at a processor, a set of measurement data including accelerometer data, magnetometer data and audio data;
receiving category information for the set of measurement data, the category information identifying a first set of measurement data categorized as indicative of a suspected fall, based on the accelerometer data, a second set of measurement data categorized as indicative of an activity of daily life (ADL) based on physical movement of users, and a third set of measurement data categorized as an inconclusive event;
training, using the processor, a fall model and training an ADL model using the set of measurement data and the category information; and
wherein the fall model and the ADL model are used to classify unclassified measurement data as indicative of a fall or an ADL, the ADL model classifying at least talking and walking as ADL events, and wherein the fall model is used to re-confirm whether the first set of measurement data categorized as indicative of the suspected fall indicates a fall.
12. The method of claim 11 , wherein the fall model and the ADL model are Gaussian mixture model (GMM) models.
13. The method of claim 12 , wherein the GMM models are trained using an expectation maximization algorithm.
14. The method of claim 11 , wherein the set of measurement data includes prior events classified as fall and ADL events.
15. The method of claim 11 , further comprising receiving additional measurement data including in-use data and updating the fall model and the ADL model using the additional measurement data.
16. The method of claim 11 , further comprising implementing Gaussian mixture model (GMM) classifiers on a subset of a selected feature space of events from the set of measurement data.
17. The method of claim 16 , wherein the selected feature space is selected via a sequential forward floating selection (SFFS) algorithm to maximize classifier performance.
18. The method of claim 11 , further comprising determining a threshold for classifying an event as a fall, and wherein the unclassified measurement data is classified based on a score for the unclassified measurement data and the threshold.
19. The method of claim 11 , wherein the set of measurement data includes accelerometer data, magnetometer data, and audio data.
20. The method of claim 11 , wherein to classify the unclassified measurement data as indicative of a fall or an ADL, the fall model and the ADL model are to determine a reconfirmation or change of classification from an initial classification received from a wearable device that generated the unclassified measurement data.