Privacy-preserving radar-based fall monitoring

1. A smart-home device with integrated fall detection, the smart-home device comprising:

a housing;

a speaker within the housing;

a microphone within the housing;

a monolithic radar integrated circuit that comprises an antenna array for receiving reflected radar, wherein:

the monolithic radar integrated circuit is housed by the housing; and

a processing system, comprising one or more processors, that is in communication with the monolithic radar integrated circuit, the processing system being configured to:

receive raw waveform data from the radar integrated circuit;

filter, from the raw waveform data, waveform data indicative of static objects to obtain motion-indicative waveform data;

perform center-of-mass tracking on a moving object present within the motion-indicative waveform data;

classify the moving object based on the center-of-mass tracking using a pre-trained machine learning model;

determine that a fall by a person has occurred; and

output an indication that the fall has been detected.

2. The smart-home device with integrated fall detection of claim 1 , wherein the monolithic radar integrated circuit has dimensions of less than 8 mm by 6 mm by 2 mm.

3. The smart-home device with integrated fall detection of claim 2 , wherein the processing system being configured to output the indication comprises the processing system being configured to cause the speaker of the smart-home device to output speech announcing that the fall has been detected and request that an auditory response be provided.

4. The smart-home device with integrated fall detection of claim 3 , wherein the processing system is further configured to receive an auditory response via the microphone and process the auditory response.

5. The smart-home device with integrated fall detection of claim 4 , wherein the processing system is further configured to perform an action in response to the auditory response, wherein the action is selected from the group consisting of:

transmit a communication via a wireless network to an emergency contact; and

output a spoken message indicating that no further action is to be taken.

6. The smart-home device with integrated fall detection of claim 5 , wherein the monolithic radar integrated circuit emits frequency-modulated continuation wave (FMCW) radio waves.

7. The smart-home device with integrated fall detection of claim 1 , wherein the smart-home device is a home assistant device.

8. The smart-home device with integrated fall detection of claim 1 , wherein the processing system being configured to output an indication that the fall has been detected is based on the processing system classifying the moving object based on the center-of-mass tracking using the pre-trained machine learning model.

9. The smart-home device with integrated fall detection of claim 8 , wherein the processing system being configured to classify the moving object based on the center-of-mass tracking using the pre-trained machine learning model comprises applying a random forest pre-trained machine learning model.

10. The smart-home device with integrated fall detection of claim 9 , wherein the pre-trained machine learning model as installed on the smart-home device is static and was created based on a training set of data indicative of situations in which a person has fallen and in which a person has not fallen.

11. The smart-home device with integrated fall detection of claim 8 , wherein the processing system being configured to perform center-of-mass tracking comprises the processing system being configured to perform center-of-mass tracking over a rolling historic time period window.

12. The smart-home device with integrated fall detection of claim 11 , wherein the rolling historic time period window is between one and fifteen seconds.

13. The smart-home device with integrated fall detection of claim 8 , wherein the processing system is further configured to create a plurality of heat maps based on the motion-indicative waveform data, wherein each heat map of the plurality of heat maps is indicative of detected motion at a plurality of distances in a plurality of directions along a particular axis.

14. The smart-home device with integrated fall detection of claim 8 , wherein the processing system being configured to receive the raw waveform data comprises the processing system being configured to receive a plurality of individual sets of raw waveform data from a plurality of antennas of the antenna array of the monolithic radar integrated circuit.

15. The smart-home device with integrated fall detection of claim 8 , wherein the processing system being configured to classify the moving object is performed by the processing system based on at least four physical features of the center-of-mass tracking.

16. The smart-home device with integrated fall detection of claim 1 , wherein the monolithic radar integrated circuit emits radar at greater than 40 GHz.

17. A method for performing fall detection, the method comprising:

positioning a home assistant device with a room such that the home assistant device is intended to remain stationary, wherein the home assistant device comprises:

a speaker;

a microphone; and

a monolithic radar integrated circuit that comprises an antenna array for receiving reflected radar;

transmitting, by the home assistant device, radar waves within the room;

based on reflected radar waves within the room, creating raw waveform data;

filtering, from the raw waveform data, waveform data indicative of static objects to obtain motion-indicative waveform data;

performing center-of-mass tracking on a moving object present within the motion-indicative waveform data;

classifying the moving object based on the center-of-mass tracking using a pre-trained machine learning model;

determining that a fall by a person has occurred based on classifying the moving object; and

outputting, by the home assistant device, speech announcing that the fall has been detected via the speaker of the home assistant device.

18. The method for performing fall detection of claim 17 , the method further comprising:

receiving an auditory response via the microphone of the home assistant device;

performing an action in response to the auditory response, wherein the action is selected from the group consisting of:

transmitting a communication via a wireless network to an emergency contact; and

outputting a spoken message indicating that no further action is to be taken.

19. The method for performing fall detection of claim 17 , wherein classifying the moving object based on the center-of-mass tracking using the pre-trained machine learning model comprises applying a random forest pre-trained machine learning model.