Feature Space Exploration for Motion Classification Based on Multi-Modal Sensor Data for Lower Limb Exoskeletons

In this paper, we address the problem of finding a minimal multi-modal sensor setup for motion classification in lower limb exoskeleton applications while maintaining the classification performance. We present an approach for a systematic exploration of the feature space and feature space dimensionality reduction for motion recognition using Hidden Markov Models (HMMs). We evaluated our approach using IMU and force sensor data with 10 subjects performing 14 different daily activities. We perform a dimensionality reduction on sensor feature level with single- and multi-subjects and we explore the feature space using fine-grained features such as the force value of a single direction. Additionally, we investigate the influence of physical characteristics on the classification quality. Our results show that a subject specific and general reduction of the sensors is possible while still achieving the same classification performance.