Signal reproduction of acceleration sensor data for physical activity recognition

This paper addresses the challenges of flexible acceleration sensor sampling rates of mobile phones and describes the problem of signal reproduction of blurred or distorted signals for physical activity recognition. In this work, low and high order signal interpolation and approximation techniques were analyzed as well as splines to improve the recognition accuracy. The identified best algorithm was implemented to a single point activity recognition system. The signal reconstruction algorithm is used for real time recognition in a standard mobile phone with an integrated acceleration sensor. This phone and its application DiaTrace is able to recognize activity such as resting, walking, cycling, car driving or jumping just by wearing the mobile in the front pocket of a trouser. We could identify that spline algorithm is excellent for activity recognition of smooth and periodic movements, but lower polynomial interpolation has advantages at non periodic and discontinuous movements.