Skip to Main content Skip to Navigation

Suivi et catégorisation multi-objets par vision artificielle. Applications au suivi de personnes et de véhicules

Abstract : This thesis addresses real-time automatic visual tracking and classification of a variable number of objects such as pedestrians or/and vehicles, under time-varying illumination conditions. The tracking system has been designed in LASMEA4 Laboratory during years 2006 to 2009, with Michel Naranjo as thesis director and Thierry Chateau as thesis manager. We first draw a landscape of applications where real-time multi-object tracking and classification is required. We then describe the main difficulties related to this functionnality, and the most frequently used methods. We choose to use a Particle Filter, operating in the joint multi-object joint configuration. This yields a high dimension state space. Getting efficient operation from a Particle Filter in such a state space is conditionned by the filter ability to " smartly " choose and propagate its particles, a process known as resampling. After reminding the principle of Particle Filter, we focus on major strategies for resampling particles : Partitionned Particle Filters and Markov Chain Monte-Carlo Particle Filters (MCMC PF). These strategies are benchmarked over synthetic data. Experiments show that we obtain best results with marginal move Markov Chain Monte-Carlo Particle Filters (MCMC PF). Based on this choice, a system is described for jointly tracking and classifying a variable number of objects under variable illumination. The measurement is provided by a static camera, and we deliberately chose to feed the filter with poor observation : a basic binary foreground / background segmentation. In order to comply with variable illumination, object shadows are modelled and the light source is jointly tracked with the multi-object configuration by the (MCMC PF). As a first contribution, we propose in this paper to jointly track the light source within the Particle Filter, considering it as an additionnal object. Illumination-dependant shadows cast by objects are modeled and treated as foreground, thus avoiding the difficult task of shadow segmentation. As a second contribution, we estimate object category as a random variable also tracked within the Particle Filter, thus unifying object tracking and classification into a single process. Real time tracking results are shown and discussed on sequences involving various categories of users such as pedestrians, cars, light trucks and heavy trucks.
Document type :
Complete list of metadatas

Cited literature [55 references]  Display  Hide  Download
Contributor : Camille Meyer <>
Submitted on : Thursday, August 23, 2012 - 2:53:11 PM
Last modification on : Thursday, January 11, 2018 - 6:17:21 AM
Long-term archiving on: : Friday, December 16, 2016 - 8:14:25 AM


  • HAL Id : tel-00724975, version 1


François Bardet. Suivi et catégorisation multi-objets par vision artificielle. Applications au suivi de personnes et de véhicules. Automatique / Robotique. Université Blaise Pascal - Clermont-Ferrand II, 2009. Français. ⟨NNT : 2009CLF21972⟩. ⟨tel-00724975⟩



Record views


Files downloads