Unsupervised Human Activity Segmentation Applying Smartphone Sensor for Healthcare

Activity-aware computing plays important role for pervasive healthcare such as health monitoring and assisted living. The collaboration of computation, telecommunication and sensing capabilities in smartphone helps the usage for user activity monitoring and recognition. However, it is still difficult to find the changing of action and retrieve the accurate activity information in human activity recognition work. This paper proposes the novel unsupervised human activity segmentation model which divides the continuous movement series into discrete activity sections. Minimized contrast segmentation algorithm with the correctness of sliding window based autocorrelation is implemented applying statistical model and time-series analysis to cover more useful signal properties including mean, variance, amplitude, and frequency. Experimental results on accelerometer embedded in smartphone show that the activity partition model achieves successful segmentation.