Personalized fall detector
A method and system for training a fall detection classifier using subject-specific movement data. A subject sets a preferred non-fall detection rate. Movement data responsive to a subject's movements during everyday activities are obtained over a predetermined data collection period. For each detected event in the movement data, values for one or more parameters that may (together or individually) indicate a fall are obtained. The obtained values are used to generate a subject-specific non-fall detection rate function. This non-fall detection rate function is used to derive a threshold value, in reference to the subject-set preferred non-fall detection rate, to distinguish fall events from non-fall events.
1. A computer-implemented method executed by a processing system that causes the processing system to perform operations comprising:
obtaining movement data responsive to a subject's movement during everyday activities over a predetermined data collection period;
detecting one or more events in the movement data;
for each particular detected event, obtaining a corresponding parameter values for one or more parameters from the movement data at the time of the particular detected event;
obtaining a fall event probability distribution that is predetermined or based on a combination of the corresponding parameter values;
determining a non-fall detection rate function based on the obtained fall event probability distribution, the non-fall detection rate function to determine a false alarm rate based on an input threshold; and
determining, using the non-fall detection rate function, a threshold value such that the false alarm rate is below a preset value.
2. The computer-implemented method of claim 1 , wherein determining the non-fall detection rate function is based on the fall event probability distribution and the corresponding parameter values.
3. The computer-implemented method of claim 1 , wherein determining the non-fall detection rate function is also based on a duration of the predetermined data collection period.
4. The computer-implemented method of claim 1 , wherein determining the threshold value comprises adapting the threshold value based on a difference or ratio between a set false alarm rate and an observed false alarm rate.
5. The computer-implemented method of claim 1 , wherein determining the threshold value comprises adapting the threshold value based on a relative difference between a current false alarm rate and a set false alarm rate.
6. A system comprising:
one or more processors; and
storage media storing instructions that, when executed by the one or more processors, cause the system to implement:
a user interface comprising an input for enabling a user to set a false alarm rate reference value;
a threshold determination subsystem is-configured to execute a non-fall detection rate function for determining a threshold set value corresponding to the false alarm rate reference value, the non-fall detection rate function generated based on a fall event probability distribution over movement data collected for the user; and
a fall detector configured to use the determined threshold set value for detecting a fall of a subject.
7. The system of claim 6 , wherein the false alarm rate reference value is relative to an existing value.
8. A computer-implemented method executed by a processing system that causes the processing system to perform operations comprising:
setting a preferred non-fall detection rate for a user;
obtaining subject-specific movement data by monitoring a subject's movements during everyday activities over a predetermined data collection period, wherein for each particular detected event in the subject-specific movement data, corresponding values for one or more parameters indicating a fall are obtained;
generating a subject-specific non-fall detection rate function based on the corresponding values, the subject-specific non-fall detection rate function to determine a false alarm rate based on an input threshold; and
determining a threshold value using the subject-specific non-fall detection rate function and the preferred non-fall detection rate set by the user to distinguish fall events from non-fall events.