Motion Classification Based on Geometrical Features of Trajectories

This paper proposes a novel approach for motion classification based on geometrical features computed on trajectories. The method follows a machine learning approach trained and validated on synthetic datasets simulating several stochastic models. The resulting model enables, in particular, the recognition of different subdiffusive behaviors, offering a finer classification than the standard method based on mean square displacement. The method is assessed on a biological dataset containing trajectories of CCR5 cell receptors.