working with accelerometer

I am trying to use tha accelerometer in an APP.
the simulator can offer some random data for that....
but how did you work to get some real data to check your algorithm ?
is there an app to just "record" real data on a watch without treatment ?
May I need to write/modify my app to write accelerator data ?

Top Replies

ekutter-dnu over 6 years ago +2

The .FIT file in the pitch counter is interesting. A couple of message records that I hadn't seen before, with ID 165 and 233. I'm guessing the record 165, which contains 3 values, is just the X, Y,…
vald70 over 6 years ago +1

I wrote an app that actually logs data and sends it to a server of mine via web requests. That let me play with the data in a windows application where debugging was much easier. I was then able to take…

All Replies

ekutter-dnu over 6 years ago

I smoothed out the curve with the following FIR values: 50,100,200,300,200,100,50

I then somewhat brute force broke the y wave into
flat - holding the club steady
rise - the back swing acceleration
fall - the back swing deceleration
rise - the actual swing acceleration,
fall - the swing deceleration

I came up with values for min/max duration and amplitude for each of those cycles by looking at the graphs. The FIR values I chose made sure to smooth out the wave in such a way that you didn't have little hic ups during any of the cycles where it briefly changed direction.

This is where having the program running on the computer really helped. Way quicker to code, test, and repeat regression tests from known samples with tools like Visual Studio and c#. (at least for me).

So I was sending the raw data to the computer so I could also play with the FIR formula.

Does this make sense? It didn't to me when I started as I had never done this kind of pattern matching.

vald70 over 6 years ago

I smoothed out the curve with the following FIR values: 50,100,200,300,200,100,50

I then somewhat brute force broke the y wave into
flat - holding the club steady
rise - the back swing acceleration
fall - the back swing deceleration
rise - the actual swing acceleration,
fall - the swing deceleration

I came up with values for min/max duration and amplitude for each of those cycles by looking at the graphs. The FIR values I chose made sure to smooth out the wave in such a way that you didn't have little hic ups during any of the cycles where it briefly changed direction.

This is where having the program running on the computer really helped. Way quicker to code, test, and repeat regression tests from known samples with tools like Visual Studio and c#. (at least for me).

So I was sending the raw data to the computer so I could also play with the FIR formula.

Does this make sense? It didn't to me when I started as I had never done this kind of pattern matching.

As you, I am starting in this type of algorithm, as for now, for me, FIR values seems a little black magic to tune ! as for detection, from the literature I read, there is a lot of statistical/AI and so on, which are way to complex IMHO to run on a watch. So I'a looking for simpler case a bit like your idea

ekutter-dnu over 6 years ago

Yea, it'd be fun to try to do it through neural-nets and effectively let it figure out the patterns. That goes way beyond my AI expertise. So I tried to come up with something simple, but it definitely isn't very flexible. I'd love to know what Garmin does for the detection. I'd hope they have some pretty advanced tools for analyzing this stuff. And they have a lot of uses for it. Swim stroke detection, running speed from wrist motion, cadence, golf swing detection. Sure hope they aren't doing it through special hard coded cases like mine.

vald70 over 6 years ago

I'd love to know what Garmin does for the detection. I'd hope they have some pretty advanced tools for analyzing this stuff. And they have a lot of uses for it. Swim stroke detection, running speed from wrist motion, cadence, golf swing detection. Sure hope they aren't doing it through special hard coded cases like mine.

Me too, as I've see that swim stroke/ muscle excercise and so on are detected/infer with reasonable guess ! and seem not using lot's of CPU !;)

vald70 over 6 years ago

Yea, it'd be fun to try to do it through neural-nets and effectively let it figure out the patterns. That goes way beyond my AI expertise. So I tried to come up with something simple, but it definitely isn't very flexible. I'd love to know what Garmin does for the detection. I'd hope they have some pretty advanced tools for analyzing this stuff. And they have a lot of uses for it. Swim stroke detection, running speed from wrist motion, cadence, golf swing detection. Sure hope they aren't doing it through special hard coded cases like mine.

may be take a look at this study