A gesture expressive model based on Laban qualities

Today, gesture analysis lacks of global models able to characterize motion expressivity and its communicational character. In this paper, we propose a set of new gesture descriptors inspired from Laban Movement Analysis (LMA) and based on 3D body trajectories. We test our descriptors ability to characterize human actions in a machine learning framework (with SVM and different random forest techniques). The results obtained on Microsoft Research Cambridge-12 (MSRC-12) dataset and show very high recognition rates (more than 97%).