Record the following sessions for this use case:
swipe_right.csv(5 min)swipe_left.csv(5 min)Rotation_Clockwise.csv(5 min)Rotation_Counter Clockwise.csv(5 min)Double_Knock.csv(5 min)Double_Thumb_Tap.csv(5 min)Idle.csv(5 min) — Samples where device stays stillUnknown_gestures.csv(5 min) — Additional gestures beyond the core 7 gestures to enhance dataset robustness
If your use-case is to detect 6 different gestures, the actual number of gestures you will need to collect is 8. The two additional classes must be:
idle— When the device is not being usedunknown— All other possible gestures and activities that a user might do when using the device
These classes are necessary to make sure the model does not produce false detections (prediction of one of the target classes when user is doing something else). It is advised to collect isolated sessions of gestures. If you have 8 classes, execute 8 sessions of data collection to make sure the activities do not get mixed up.