Abstract : In this paper, we address the analysis of activities from long range video sequences. We present a method to automatically extract spatial and temporal structure from a video sequence from low level motion features. The scene layout is first extracted, with a set of regions that have homogeneous activities called Motion Patterns. These regions are then analyzed and the recurrent temporal motifs are extracted for each Motion Patterns. Preliminary results show that our method can accurately extract important temporal motifs from video surveillance sequences.
https://hal-univ-bourgogne.archives-ouvertes.fr/hal-01018756
Contributeur : Yannick Benezeth
<>
Soumis le : vendredi 4 juillet 2014 - 22:13:00
Dernière modification le : mercredi 12 septembre 2018 - 01:27:31
Document(s) archivé(s) le : samedi 4 octobre 2014 - 13:06:10
Alberto Quintero Delgado, Yannick Benezeth, Désiré Sidibé. Automatic spatial and temporal organization of long range video sequences from low level motion features. IEEE CVPR Scene Understanding Workshop, Jun 2014, United States. 2 p. ⟨hal-01018756⟩
