Human Activity Recognition of Exoskeleton Robot Based on Adaptive DTW Classifier

This paper proposes a method based on adaptive Dynamic Time Warping (DTW) classifier to recognize human activities of exoskeleton robots. First, the path constraint restriction of the traditional DTW algorithm is improved. Then a multi-voting weight distribution mechanism is designed for the DTW classifier when multi-sensor fusion. Finally, we propose a strategy to update the standard templates for classifier using the recognition result and current data. The experimental results show that the adaptive strategy can make the classifier gradually adapt to the user’s behavior habits, thereby improving the recognition accuracy.