gpmf-parser icon indicating copy to clipboard operation
gpmf-parser copied to clipboard

How to determine start and end of a ride

Open Fougl opened this issue 2 years ago • 5 comments

Hi, I would like some guidelines on how to make a model to determine start and end of a ride from gorpo accelerometer, gyroscope etc most robustly. Thanks

Fougl avatar Aug 24 '23 20:08 Fougl

That is will depend on you thresholds. Extract the all the data for a clip first, determine what trims a human editor would make, then look at the matching ACCL data, then design you filter threshold to match. Experiment a lot, with a lot of footage.

dnewman-gpsw avatar Aug 24 '23 20:08 dnewman-gpsw

I have different cameras for the rides, do you think i should develop model separately for each camera? What hitrate should I expect. Does combining ACCL, GYRO etc imrove the hitrate?

Fougl avatar Aug 24 '23 20:08 Fougl

Fusing ACCL and GYRO is generally a good idea, the code for this is complex and beyond me. All cameras filter these streams to ~200Hz, so they will be nothing custom need for each camera model. Hit rate will depend on how advanced you get, but I expect this is one of the simpler scenarios. You can also use GPS is that is available.

dnewman-gpsw avatar Aug 24 '23 20:08 dnewman-gpsw

And no way for some physicist to calculate exactly? because there are unwanted movements sometimes if one hits the camera with the hand. so its really tricky to get near 100% hitrate. but if someone can calculate the movement, like path or velocity. like smooth the data, use all sensors

Fougl avatar Aug 24 '23 22:08 Fougl

That is sensor fusion, yes it helps, if not required. Yes math with the help of a physicists. I'm not a physicist.

dnewman-gpsw avatar Aug 24 '23 23:08 dnewman-gpsw