Independent component analysis of EMG for posture detection: Sensitivity to variation of posture properties

We applied Agglomerative Hierarchical Clustering (AHC) technique on independent components of low back surface electromyography (sEMG) signals, in order to differ sitting and standing postures. Preliminary results from small group of healthy subjects, suggested that presented method might be used to distinguish between two postures in different conditions. Clustering accuracy varied from 60% to 70% due to variation of position properties, even when the muscle activity was very low: from 11% to 23% of Maximal Voluntary Contraction (MVC).