Signature Descriptor for Free Form Trajectory Modeling

Abstract

Motion trajectory modeling plays important role in characterizing human or robot action and behavior. However, effective and capable descriptors are lacking that can fully depict space trajectories. In this paper, we propose a novel signature mechanism for free form trajectory modeling in Euclidean space. The signature admits rich invariants due to the computational locality. By implementing the approximate signature, the noise-sensitive high order derivatives are avoided. The trajectory is recognized based on the customized signatures similarity metric. The conducted experiments verified the signature's effectiveness and robustness in 3-D trajectory representation and recognition.