In this paper, we evaluate experimentally several methods for recognizing walking activities using on-body wireless nodes equipped with inertial and orientation sensors. The walking activities (walking on flat surfaces, uphill and downhill, upstairs and downstairs) are selected by healthcare experts as being relevant for elderly patients suffering from Chronic Obstructive Pulmonary Disease (COPD). We target specifically recognition methods that operate with time-domain features only, due to the limitations of sensor nodes in terms of computational power, memory and energy. The results show that among all sensors the compass sensor performs best regardless of the classification method. Both the compass and gyroscope sensors outperform significantly the accelerometer sensor, which is typically used in previous related work. The best trade-off in performance vs. time complexity is obtained by fusing the gyroscope and compass information, with an overall accuracy of 91%.
|Name||Ambient Intelligence and Smart Environments|
|Workshop||7th International Conference on Intelligent Environments, IE 2011|
|Period||25/07/11 → 28/07/11|
|Other||25-28 July 2011|