System and method to recognize activities performed by an individual
A method implemented in a system which comprises a lightweight personal wearable monitoring device, supplied by a battery, comprising an accelerometer, a processing unit, a display, a remote server, said method comprising the steps: /a/ collecting, from a plurality of individuals caused to practice various physical activities from a set of predefined activities, acceleration data from sensing devices placed on each of said individuals, /b/ defining N small data-size specific metrics, computed from acceleration signals, which allow to define a global activity classifier, /c/ acquiring, at a first monitoring device worn by a first user, acceleration signals from the accelerometer of the first monitoring device, /d/ calculating, at the first monitoring device, over each lapsed time unit T 1 , specific metrics values from the sensed signals, to form a series of specific metrics values, /e/ send them to the processing unit, /f/ allocate an activity type for each time unit together with corresponding specific metrics values of the received series, to form a time chart of activity types presumably performed by the first user over a second period T 2, /g/ display the time chart of activity types presumably performed by the first user on the display and allow the first user to confirm or correct partly the type of activity performed over the second period, and allow correction by the first user.
1. A method implemented in a system which comprises:
at least a lightweight personal wearable monitoring device, supplied by a battery, comprising at least an accelerometer which senses acceleration signals,
a processing unit, a display,
at least a remote server,
said method comprising the steps:
/a/ collecting, from a plurality of individuals caused to practice various physical activities from a set of predefined activities, acceleration data from one or several sensing devices placed on each of said individuals, and at different body positions,
/b/ defining N small data-size specific metrics, to be computed from acceleration signals sensed by accelerometer(s) of monitoring devices, which allow to define a global activity classifier, said activity classifier being an algorithm designed to allocate a current activity among the a set of predefined activities,
/c/ acquiring, over time, at a first monitoring device worn by a first user, acceleration signals from the accelerometer of the first monitoring device,
/d/ calculating, at the first monitoring device, periodically at a predefined rate F 1 , and over each lapsed time unit (T 1 =1/F 1 ), specific metrics values from the sensed signals, to form a series of specific metrics values,
/e/ send this series of specific metrics values to the processing unit,
/f/ at the processing unit, allocate an activity type for each time unit together with corresponding specific metrics values of the received series, to form a time chart of activity types presumably performed by the first user over a second period (T 2 ),
/g/ display the time chart of activity types presumably performed by the first user on the display and allow the first user to confirm or correct partly the type of activity performed over the second period (T 2 ),
/h/ send a correction information back to the server, which allows to define or refine a user-specific activity classifier.
2. The method according to claim 1 , wherein the set of predefined activities comprises at least: normal walk, fast walk, run, dance, climbing stairs, rest, office sit, TV watch, sports like football, swimming, rowing, treadmill run, biking, skiing, skating.
3. The method according to claim 1 , wherein the number of specific metrics N is no greater than 8, wherein each metric has a size no greater than 8 Bits.
4. The method according to claim 1 , wherein the number of specific metrics N is equal to 5, wherein each metric has a size less than 5 Bits.
5. The method according to claim 1 , wherein each metric is computed at an end of every predefined time unit (T 1 ) with the acceleration data sensed over the lapsed predefined time unit, said time unit having a duration comprised between 10 seconds and 360 seconds.
6. The method according to claim 1 , in which the processing unit and the display form part of a smartphone.
7. The method according to claim 1 , in which at step /f/, at the processing unit, the allocation of type activity relies on a transfer matrix TM computation.
8. The method according to claim 1 , in which at step /f/, at the processing unit, the allocation of type activity relies on the classifier resulting from the global classifier and the user-specific activity classifier, incrementally amended through parameterization with the user feedbacks.
9. The method according to claim 1 , in which at step /f/, it further comprises:
Append time units in timed series of consistent activities, with elimination of unlikely transients.
10. The method according to claim 1 , in which the activity monitoring device has a low battery capacity, namely less than 100 mAh and a good battery autonomy namely no less than 2 weeks.
11. The method according to claim 1 , in which the activity monitoring device, the processing unit and the display form part of a single device such as a smartphone.
12. A system comprising:
at least a lightweight personal wearable monitoring device,
a processing unit, a display,
at least a remote server,
the system being configured to carry out the steps of:
/a/ collecting, from a plurality of individuals caused to practice various physical activities from a set of predefined activities, acceleration data from one or several sensing devices placed on each of said individuals, and at different body positions,
/b/ defining N small data-size specific metrics, to be computed from acceleration signals sensed by accelerometers of monitoring devices, which allow to define a global activity classifier,
/c/ acquiring, over time, at a first monitoring device worn by a first user, acceleration signals from the accelerometer of the first monitoring device,
/d/ calculating, at the first monitoring device, periodically at a predefined rate F 1 , and over each lapsed time unit, specific metrics values from the sensed signals, to form a series of specific metrics values,
/e/ send this series of specific metrics values to the processing unit,
/f/ at the processing unit, allocate an activity type for each time unit together with corresponding specific metrics values of the received series, to form a time chart of activity types presumably performed by the first user over a second period,
/g/ display the time chart of activity types presumably performed by the first user on the display and allow the first user to confirm or correct partly the type of activity performed over the second period,
/h/ send a correction information back to the server, which allows to define or refine a user-specific activity classifier.