Abstract

The work concentrates on combining discrete and continuous data in an algorithm to detect complex activity patterns.With the InvenSense MotionFitTM Software Development Kit (SDK) accelerometer and gyrometer data are recorded with the MPU-9150 sensor. The raw data consisting of processed daily activities are preprocessed via a shifted window and different features are calculated. Afterwards activity recognition is done in MATLAB using the PMTK3 toolbox from Murphy et al. where the classification algorithms are continuous Hidden Markov Models (cHMM).