Groupe de Travail Piétons et Foules Séminaire Piétons et Foules

Séminaire Piétons et Foules - Seminar Pedestrians and Crowds

Le séminaire du groupe de travail "Piétons et Foules" a pour but de réunir des chercheurs de diverses disciplines qui s'intéressent aux foules et aux piétons.
Ce séminaire abordera des thématiques variées: détection automatique de piétons et tracking, modélisation, simulation, expériences, etc.

The seminar of the Working Group "Pedestrians and Crowds" aims at gathering researchers from various domains interested in crowds and pedestrians.
The range of topics that will be discussed will be very large: Automatic detection and tracking, modeling, simulation, experiments, etc.

Organisateurs - Organizers :
Cécile Appert-Rolland (LPT) et Emanuel ALDEA (SATIE)

S'inscrire à la liste de diffusion - Subscribe to our mailing list

Se désinscrire de la liste de diffusion - Unsubscribe

Vendredi 21 Decembre 2018: Alexandre ALAHI (EPFL)

      Titre - Title:   Socially-aware Transportation
      Lieu - Location:     Auditorium, Bat. 660 DIGITEO, rue Noetzlin 91190 Gif-sur-Yvette
      Heure - Time:     10h00
      Abstract   Humanity is at the dawn of a digital revolution where Artificial Intelligence (AI) is poised to reshape the future of transportation with self-driving cars, delivery robots, and intelligent machines more broadly. To this end, a fundamental challenge is to develop machines that can not only perform intelligent tasks, but do so while co-existing with humans in the open world. Machines need to learn unwritten common sense rules, ethics, and comply with social conventions. Delivery robots should respect personal space, yield right-of-way, and ultimately ``read'' the behavior of others to effectively navigate crowded spaces. While AI has made great progress in classifying images or playing games driven by well-defined set of rules, intelligent machines still lack common sense and the ability to make seamless, safe, moral and efficient decisions in crowded social scenes. To reach this ambitious goal, I propose empowering machines with a type of cognition I call socially-aware AI, i.e., systems equipped with perception and social intelligence. In other words, I aim to develop systems that have the capacity to i) understand human behavior and ii) effectively navigate and negotiate complex social interactions and environments. In this talk, I will present our latest works towards socially-aware transportation.

Vendredi 21 Decembre 2018: Soutenance de thèse N. PELLICANO

      Titre - Title:   Tackling pedestrian detection in large scenes with multiple views and representations
      Lieu - Location:     Auditorium, Bat. 660 DIGITEO, rue Noetzlin 91190 Gif-sur-Yvette
      Heure - Time:     14h00

Mardi 25 Septembre 2018: Alexandre NICOLAS (LPTMS)

      Titre - Title:   Specificities of individual and collective pedestrian dynamics
      Lieu - Location:     Auditorium, Bat. 660 DIGITEO, rue Noetzlin 91190 Gif-sur-Yvette
      Heure - Time:     14h00

Lundi 12 Mars 2018: E. Monari, T. Pollok, S. Voth (Fraunhofer IOSB)

      14h00   E. Monari       Short Introduction of Fraunhofer IOSB and Research Group on Image-based Real-Time Systems
14h15   T. Pollok   Extrinsic Camera Calibration of Non-Overlapping Cameras based on Scene 3D-Reconstructions from a Single Video
  Abstract   Camera pose estimation describes the process of estimating the six degrees of freedom (DOF) of a camera - namely the three translation parameters (3D position) and three angles for orientation in 3D space (rotation) with respect to a reference 3D coordinate system. Estimation of the 6DOF is a highly important task in many computer vision applications (e.g. stereo-vision, camera arrays for depth estimation and 3D scene reconstruction, video surveillance, etc.). Pose estimation (also called extrinsic orientation or extrinsic multi-camera calibration) provides information about geometric relationships between camera coordinate systems and (together with available intrinsic camera parameters) allows in many cases to determine positions of objects in the scene in a common 3D coordinate system. In video surveillance, tracking of objects across multiple cameras is a typical and challenging task which requires such information for handing over the position information of tracked objects between cameras. Though there are many existing solutions for targeting the camera pose estimation problem, there are still missing approaches for practical and robust estimation of extrinsic camera orientations. In particular for large camera networks with a high number of distributed sensors over a large site with potentially nonoverlapping fields of views. In this talk an approach is presented which tackles the pose estimation problem for distributed, non-overlapping cameras. The basic idea is to record a video from the scene in which all cameras are located. State of the art SLAM or structure-from-motion (SfM) algorithms are used to create a 3D reconstruction from the video, which results in a 3d pointcloud that is represented using multiple keyframes that is used as a common reference coordinate space. Each cameras pose can be estimated by finding 2D-3D correspondences between the image points of the camera and 3D points in the point cloud. 
15h00   E. Monari, S. Voth   Current Research Activities on Video-based Crowd Monitoring

Lieu - Location:   Auditorium, Bat. 660 DIGITEO, rue Noetzlin 91190 Gif-sur-Yvette

Lundi 29 Janvier 2018: Armin SEYFRIED (Research Center Juelich - Germany)

      Titre - Title:   Bottleneck flow - where pedestrian dynamics need social psychology
      Résumé - Abstract:   An experiment in which a large group of people enters a hall through two different spatial barrier structures is analysed. Physical measurements show the influence of the spatial structure on the dynamics. Density, waiting time and speed of progress show large variations. A questionnaire study reveals how people perceive and evaluate these entrance situations.
      Lieu - Location:     Auditorium, Bat. 660 DIGITEO, rue Noetzlin 91190 Gif-sur-Yvette
      Heure - Time:     10h30

Lundi 4 Décembre 2017: Julien PETTRE (INRIA - Rennes)

      Titre - Title:   Velocity-based algorithms for crowd simulation
      Résumé - Abstract:   A crowd simulator is a computer program which aims at computing the motion of numerous people gathered in a same place (a crowd). Microscopic algorithms follow a bottom up approach to this problem: as in real crowds, the global motion of the crowd results from all the combination of local interactions between people while they move indendently. In this presentation, we will detail the velocity-based approaches, which enabled significant progresses for microscopic crowd simulators in terms of realism. Velocity-based approaches simulate the motion of agents so that they always follow an admissible velocity, i.e., allowing a collision-free trajectory in the near future. We will explain how to compute admissible velocities for agents, and how these techniques recently evolved to produce always better simulation results. We will also present a specific category of velocity-based approaches, called vision-based algorithms. The specificity of vision-based approaches is that, in contrast with other categories of approaches, agents retrieve the information they need to compute admissible velocities through a "virtual retina", which simulates a human visual perception. We will conclude the presentation by presenting our vision for the future of this topic.
      Lieu - Location:     Room 2014, Lab. SATIE, Bat. 660 DIGITEO, rue Noetzlin 91190 Gif-sur-Yvette
      Heure - Time:     10h
          Une pièce d'identité vous sera demandée à l'entrée

Lundi 6 Novembre 2017: Présentations courtes

      Titre - Title:   Tour de table des sujets en cours / Questions ouvertes
      Lieu - Location:     Salle 114, Laboratoire de Physique Theorique, Bat 210 (1er etage).
      Heure - Time:     10h30

Mardi 18 Juillet 2017: Alexandre NICOLAS (LPTMS - Orsay)

      Titre - Title:   Microscopic dynamics of pedestrian flows through a narrow exit: Experimental findings about the influence of behaviours and modelling
      Résumé - Abstract:   Recent experimental works about pedestrian flows through an exit have clarified details of the evacuation process and the influence of the behaviours [Pastor et al., PRE, 2017; Nicolas et al., TRB, 2017] . Statistical laws have been brought to light for competitive evacuations, which are strongly reminiscent of granular flows through an orifice, with in particular heavy-tailed (power-law) distributions p(Δ t) of intervals Δ t between successive escapes. Although a variety of cellular-automaton models for pedestrian dynamics are available, as far as I know, none of them was proven to reproduce these empirical microscopic statistics of escape times.
I will discuss the difficulty of obtaining a power-law distribution p(Δ t) with a lattice-based model and the ingredients that need to be incorporated to do so, the first of which is some kind of disorder. I will then introduce a minimal cellular-automaton which reproduces the experimental microscopic statistics in a semi-quantitative way, by accounting for heterogeneity in the pedestrians' behaviours [Nicolas et al., PRE, 2016].
      Lieu - Location:     Salle 114, Laboratoire de Physique Theorique, Bat 210 (1er etage).
      Heure - Time:     14h

Lundi 12 Juin 2017: Andrea Cavallaro (Centre for Intelligent Sensing - Queen Mary University of London, UK)

      Titre - Title:   Multiple target tracking for wearable and robotic cameras
      Résumé - Abstract:   Vision offers a powerful sensing modality to understand and interact with the physical world. The rapid progress in hardware, models and algorithms is supporting the emergence of applications for the recognition of events from wearable smart cameras and camera-equipped robots, such as unmanned land and aerial vehicles (i.e. self-driving cars and mini-drones). In this context I will present an online multi-target tracker that exploits both high- and low-confidence target detections in a Probability Hypothesis Density Particle Filter framework to continuously localise people from moving cameras. High-confidence detections are used for label propagation and target initialization, whereas low-confidence detections only support the propagation of labels. Data association is performed after prediction to avoid computationally expensive labelling procedures such as clustering. I will discuss results on the Multiple Object Tracking benchmark dataset and present several application scenarios.
      Lieu - Location:     Auditorium du Batiment 660 DIGITEO, rue Noetzlin, Gif-sur-Yvette
      Heure - Time:     10h00

Lundi 22 Mai 2017: Jennifer Vandoni (Laboratoire SATIE)

      Titre - Title:   An Evidential Framework for Pedestrian Detection in High-Density Crowds
      Lieu - Location:     Salle 114, Laboratoire de Physique Theorique, Bat 210 (1er etage).
      Heure - Time:     10h30

Jeudi 27 Avril 2017 : Sylvain Faure, Fatima Al Reda, Bertrand Maury (LMO - Laboratoire de Mathématiques d'Orsay)

      Titre - Title:   Approche granulaire de phénomènes d'évacuation de foules: modélisation mathématique et confrontation aux expériences
      Lieu - Location:     Salle 114, Laboratoire de Physique Theorique, Bat 210 (1er etage).
      Heure - Time:     11h

Le séminaire est financé par le projet PERCEFOULE (PALM)