3D tracking based on possibilities rather than probabilities
Résumé
We propose in this paper a method to constraint the tracking to the physically possible space without knowledge of the scene. The idea is to use a naively made trajectory dataset as a base to track the targets into the possible space rather than the probable space. First, a trajectory database is computed under weak assumptions permissive classic Kalman Filter). Then, for the effective tracking, we replace the prediction phase of the KF to rely only on the dynamic stored in the database of historical trajectories. We show that performances can reach these obtained by other methods based on strong assumptions (e.g. the 3D model of the scene).
Origine | Fichiers produits par l'(les) auteur(s) |
---|
Loading...