Fine grained activity recognition using smart handheld
Jayita Saha, Chandreyee Chowdhury, Ishan Roy Chowdury, Priya Roy · 2018
Sensors embedded in smart handheld can be extremely useful in providing information on people's activities and behaviors, that can be extremely useful in smart home, smart healthcare applications. Existing work mostly uses one or more specific devices (with embedded sensors) for activity recognition and most of the time the detected activities are coarse grained like sit, walk etc. In this paper we propose a fine grained ubiquitous activity recognition system that uses only smartphone accelerometer that is available in any smartphone configuration. The system applies feature extraction and learning mechanism and selects minimal set of features that can precisely recognize twelve activities including static (sit on chair/floor, lying left/straight/right) and dynamic (slow/normal/brisk walk, running, climbing upstairs/downstairs, jogging, running) fine grained activities resulting in 92.7% accuracy with ensemble classifier.