Activity recognition using statistical gait parameters from a single accelerometer

Wearable sensor systems represent an increasingly viable approach to short and long term motion classification and gait estimation primarily due to decreased size, cost and broad applicability. Accurate human motion recognition by means of minimally intrusive and low-power systems has countless applications in healthcare, sports medicine, and military operational settings. In this work, human locomotion was characterized in terms of gait response using for two type of activities. A single sensor positioned on the chest was used for the recorded acceleration response. The proposed algorithm uses Gaussian mixture from the signal features and support vector machine classifier to characterize gait. Good discrimination between gaits at different phases of locomotion is observed for the selected activities. The proposed approach also shows low sensitivity to noise and exhibit robustness across sensor types.