Skip to Main content Skip to Navigation
New interface

Poursuite multi-cibles sur signal brut

Audrey Cuillery 1 
1 SIMSMART - SIMulation pARTiculaire de Modèles Stochastiques
Inria Rennes – Bretagne Atlantique , IRMAR - Institut de Recherche Mathématique de Rennes
Abstract : This thesis is about tracking several independent targets, using measurements collected by sensors or from a radar-image. The algorithms studied belong to the class of particle filters. In a first part, two algorithms proposed in the literature for multitarget tracking are compared. The ATRAPP algorithm proposed by Ubeda- Medina et al. makes it possible to generate shuffled multitarget particles (a multitarget vector where each target is possibly replicated from a different multitarget particle), under a posterior target independence assumption that is almost never met in practical situations. In a second part, an original algorithm is introduced, that uses an auxiliary Markov transition in its design and generalizes the auxiliary particle filter principle. In the context of multitarget tracking, this new algorithm makes it possible to generate shuffled multitarget particles, with no posterior target independence assumption. The optimal choice of the design parameters (auxiliary weights, auxiliary transition matrix or kernel) is discussed. Using an auxiliary transition kernel makes it possible to better shake the multitarget particles and to further decrease the variance. The optimal design reduces to the numerical solution of continuous optimization problems. In the special case where targets have a linear Gaussian dynamics, a special case of practical interest in applications, it is shown that these problems are strongly convex.
Document type :
Complete list of metadata
Contributor : ABES STAR :  Contact
Submitted on : Friday, November 18, 2022 - 2:51:12 PM
Last modification on : Monday, December 5, 2022 - 2:14:36 PM


Version validated by the jury (STAR)


  • HAL Id : tel-03850485, version 2


Audrey Cuillery. Poursuite multi-cibles sur signal brut. Traitement du signal et de l'image [eess.SP]. Université Rennes 1, 2022. Français. ⟨NNT : 2022REN1S027⟩. ⟨tel-03850485v2⟩



Record views


Files downloads