I'm doing some gesture recognition from smartphone gyroscope sensor. As you can see other picture there are 2 gestures types A and B. Yellow lines just show which part of sensor data should be detected.
Detecting gesture B is pretty easy with just a simple threshold. However gesture A is more tricky because the slope of peaks are more gentle and there is also "bounce back" peaks. Besides there can be to A gestures close to each other (like shown on the end of above graph).
I'm quite new to pattern recognitions and wondering is somebody could give tips for ideas/keywords/algorithms/links I should explore in this case. I would prefer avoiding using pattern recognition based on training (neural networks etc.) and need to detect those realtime with minimum latency (so cannot use running average to smooth the signals).
You can have a look at some researches that were done by other colleagues (see below google link). https://scholar.google.co.jp/scholar?q=gesture+recognition++gyroscope+sensor&hl=fr&as_sdt=0&as_vis=1&oi=scholart&sa=X&ved=0ahUKEwjd8tvYt_3QAhWGy7wKHUzEDNAQgQMIHjAA