Trajectories of Multiple People in Crowds Using Laser Range Scanner
ADIAVIAKOYE ; PLAINCHAULT ; BOURCERIE ; AUBERLET
Type de document
COMMUNICATION AVEC ACTES INTERNATIONAL (ACTI)
Langue
anglais
Auteur
ADIAVIAKOYE ; PLAINCHAULT ; BOURCERIE ; AUBERLET
Résumé / Abstract
In everyday life, we are seeing amazing choreography of movement's crowd of pedestrians. Pedestrians cross and avoid but do not appear voluntarily cooperate with each other. In this paper we track a crowd of pedestrians in a covered and crowded to understand social behaviour area. The observations from the data base provided by a network of laser rangefinder with a sampling frequency of 50 Hz. We introduce a stable feature extraction method based on accumulated distribution of successive laser frames. To isolate pedestrians, we propose a non-parametric method exploiting the Parzen windowing technique. We apply the new method of Rao-Blackwellized Monte Carlo data association for tracking a highly variable number of pedestrians. The algorithm is quantitatively evaluated through a social behaviour experiment taking place in the lobby of a school. During this experience nearly 300 students are tracked.
