Mixed signature descriptor with global invariants for 3D motion trajectory perception and recognition

Motion perception and recognition by trajectory are important for motion analysis. A flexible and adaptable descriptor of motion trajectory plays important role in the performance of perception and recognition. Motion trajectories usually use raw data for many tasks, which is not effective. Shape descriptors and signature of differential invariants are good for local features but global feature is lacking. In this paper, we present a mixed signature descriptor with global invariants which were not included in previous descriptors. The mixed signature is capable of capturing the global feature and is flexible in many tasks. The conducted experiments verified the capability of our method.