AGORASET: a dataset for crowd video analysis
Résumé
The ability of efficient computer vision tools (detec- tion of pedestrians, tracking, ...) as well as advanced rendering techniques have enabled both the analysis of crowd phenomena and the simulation of realistic sce- narios. A recurrent problem lies in the evaluation of those methods since few common benchmark are avail- able to compare and evaluate the techniques is avail- able. This paper proposes a dataset of crowd sequences with associated ground truths (individual trajectories of each pedestrians inside the crowd, related continuous quantities of the scene such as the density and the veloc- ity field). We chosed to rely on realistic image synthesis to achieve our goal. As contributions of this paper, a typology of sequences relevant to the computer vision analysis problem is proposed, along with images of se- quences available in the database.
| Origine | Fichiers produits par l'(les) auteur(s) |
|---|