PhD subject: Models and sequential Monte-Carlo methods for tracking in high dimensional spaces
Supervisors / contacts: Christelle Garnier & François Septier (email@example.com, firstname.lastname@example.org)
Institutions: IMT Lille Douai & CRIStAL (Centre de Recherche en Informatique, Signal et Automatique de Lille, UMR CNRS 9189), Lille, France
Duration: 3 years (from Sept./Oct. 2018), expected funding from the University of Lille and IMT Lille Douai
Multiple object tracking is an important area of research. The range of applications is broad including surveillance, transports, biomedicine, meteorology, search rescue operations… Sequential Monte-Carlo methods , also known as particle filters, are currently widely used to estimate the object states (position, velocity) from the noisy data provided by sensors. However, they become ineffective for high dimensional problems, especially when the number of objects increases.
Recently, particle filters based on the partition of the large state space into subspaces of smaller dimension have been proposed [2-3]. They are promising both in terms of performance and computational cost for general estimation problems. The PhD proposal aims to explore these partitioning techniques and to develop new algorithms to push the limits of particle filtering in high dimension and to improve the tracking of a large number of objects in a scene. To implement this type of techniques, it is necessary to identify a factorization of the state space, which means to identify object groups and to model intra- and inter-groups interactions . Algorithms should be developed to jointly estimate the number and the structure of groups and the states of the objects.
A more detailed description is given here (in french).
Data processing, high dimension, Monte-Carlo methods, object and group tracking, interaction models, partitioning.
To apply for this position, please send us a CV and a motivation letter.
 A. Doucet, S. Godsill, and C. Andrieu, “On sequential Monte Carlo sampling methods for Bayesian filtering,” Statistics and Computing, vol. 10, pp. 197–208, March 2000.
 F. Septier and G.W. Peters, “An overview of recent advances in Monte-Carlo methods for Bayesian filtering in high-dimensional spaces”, Theoretical Aspects of Spatio-Temporal Modeling, SpringerBriefs – JSS Research Series in Statistics, 2015.
 P. Rebeschini and R. van Handel, “Can Local Particle Filters Beat the Curse of Dimensionality ?”, The Annals of Applied Probability, 2015.
 M. Oulad Ameziane, C. Garnier, F. Septier and E. Duflos, “Visual tracking of multiple objects using a local particle filter”, GRETSI 2017, Juan-les-Pins, France, Sept. 2017.
(c) GdR 720 ISIS - CNRS - 2011-2018.