How to implement a fall detection algorithm

[GoKu25]

Hi,

I have a Navigation system strapped to a shoe which has to detect a fall. At the moment, I don't have accelerometer values. I have just the Roll, Pitch and Yaw to work with. I made a rudimentary system to detect fall when the angle is above a certain threshold but it is prone to fall positives for instance, when the user is on his knees to tie his shoe laces. So I am thinking detectiion based on a rate change of angle value. 

From the data below, you can see that the pitch angle changes from 10 to 80 degrees in 527ms for a fall to the right. I tried using Peak Detection PtByPt. vi with a peak threshold of 60 and width of 10 but the detection was too slow. I ran the while loop at 20ms

Is there another way  to implement this?

Note: This is a system for worker in Industry so Activities of Daily Life does not need checking

01/03/2017  12:19:01,860   10,790822 

01/03/2017  12:19:01,984   15,046648
01/03/2017  12:19:02,083    26,481090
01/03/2017  12:19:02,201    40,054445
01/03/2017  12:19:02,300    61,342823
01/03/2017  12:19:02,387    82,453988

[Henrik_Volkers]

I would start to analyse real data.

If you curently don't have real data, use your smartphone and an app to get some data of the included 3axis accelerometer. I like to play with android 'Sensor Kinetics' or 'Physics Toolbox'

And never allow the operator to sit down 😄

[Alexander_Sobolev]

Peak detection does not work, because you need a peak (worker should stand up).

Take a threshold - determine moment when suspicious activity happened, then check what is happening (by slope of neighbouring points). 

You will need to record all situations that can happen (running, falling, getting under table to reach something, sitting with feet in different positions) and be able to identify all of them. May be add an emergency reset button for worker with a beeper to indicate alarm.