Level, downhill and uphill walking identification using neural networks

Body accelerations during human walking are recorded by a portable measuring device. A new method for parametrising body accelerations is introduced. The parameters are presented to a Kohonen neural network classifier and the feasibility of identification and dissociation of level and walking on a gradient is demonstrated. The most important and original aspect of this classification is its ability to identify the gradient of walking performed in free-living conditions from walking trained on a treadmill.<>