Object recognition in the real world is a big challenge in the field of computer vision. Given the potentially enormous size of the search space it is essential to be able to make intelligent decisions about where in the visual field to obtain information from to reduce the computational resources needed. In this report a POMDP (Partially Observable Markov Decision Process) learning framework, using a policy gradient method and information rewards as a training signal, has been implemented and used to train fixation policies that aim to maximize the information gathered in each fixation. The purpose of such policies is to make object recognition faster by reducing the number of fixations needed. The trained policies are evaluated by simulation and comparing them with several fixed policies. Finally it is shown that it is possible to use the framework to train policies that outperform the fixed policies for certain observation models.

Tracking and understanding human gait is an important step towards improving elderly mobility and safety. This thesis presents a vision-based tracking system that estimates the 3D pose of a wheeled walker user's lower limbs with cameras mounted on the moving walker. The tracker estimates 3D poses from images of the lower limbs in the coronal plane in a dynamic, uncontrolled environment. It employs a probabilistic approach based on particle filtering with three different camera setups: a monocular RGB camera, binocular RGB cameras, and a depth camera. For the RGB cameras, observation likelihoods are designed to compare the colors and gradients of each frame with initial templates that are manually extracted. Two strategies are also investigated for handling appearance change of tracking target: increasing number of templates and using different representations of colors. For the depth camera, two observation likelihoods are developed: the first one works directly in the 3D space, while the second one works in the projected image space. Experiments are conducted to evaluate the performance of the tracking system with different users for all three camera setups. It is demonstrated that the trackers with the RGB cameras produce results with higher error as compared to the depth camera, and the strategies for handling appearance change improve tracking accuracy in general. On the other hand, the tracker with the depth sensor successfully tracks the 3D poses of users over the entire video sequence and is robust against unfavorable conditions such as partial occlusion, missing observations, and deformable tracking target.

Tracking and understanding human gait is an important step towards improving elderly mobility and safety. This thesis presents a vision-based tracking system that estimates the 3D pose of a wheeled walker user's lower limbs with cameras mounted on the moving walker. The tracker estimates 3D poses from images of the lower limbs in the coronal plane in a dynamic, uncontrolled environment. It employs a probabilistic approach based on particle filtering with three different camera setups: a monocular RGB camera, binocular RGB cameras, and a depth camera. For the RGB cameras, observation likelihoods are designed to compare the colors and gradients of each frame with initial templates that are manually extracted. Two strategies are also investigated for handling appearance change of tracking target: increasing number of templates and using different representations of colors. For the depth camera, two observation likelihoods are developed: the first one works directly in the 3D space, while the second one works in the projected image space. Experiments are conducted to evaluate the performance of the tracking system with different users for all three camera setups. It is demonstrated that the trackers with the RGB cameras produce results with higher error as compared to the depth camera, and the strategies for handling appearance change improve tracking accuracy in general. On the other hand, the tracker with the depth sensor successfully tracks the 3D poses of users over the entire video sequence and is robust against unfavorable conditions such as partial occlusion, missing observations, and deformable tracking target.

Les agents robotiques mobiles ou aériens sont confrontés au besoin de planifier des actions avec information incomplètesur l'état du monde. Dans ce contexte, cette thèse propose un cadre de modélisation et de résolution de problèmes deplanification de perception et de mission pour un drone hélicoptère qui évolue dans un environnement incertain etpartiellement observé afin de détecter et de reconnaître des cibles. Nous avons fondé notre travail sur les ProcessusDécisionnels Markoviens Partiellement Observables (POMDP), car ils proposent un schéma d'optimisation général pour lestâches de perception et de décision à long terme. Une attention particulière est donnée à la modélisation des sortiesincertaines de l'algorithme de traitement d'image en tant que fonction d'observation. Une analyse critique de la mise enoeuvre en pratique du modèle POMDP et du critère d'optimisation associé est proposée. Afin de respecter les contraintes desécurité et de sûreté de nos robots aériens, nous proposons ensuite une approche pour tenir compte des propriétés defaisabilité d'actions dans des domaines partiellement observables : le modèle AC-POMDP, qui sépare l'informationconcernant la vérification des propriétés du modèle, de celle qui renseigne sur la nature des cibles. Enfin, nous proposonsun cadre d'optimisation et d'exécution en parallèle de politiques POMDP en temps contraint. Ce cadre est basé sur uneoptimisation anticipée et probabilisée des états d'exécution futurs du système. Nous avons embarqué ce cadrealgorithmique sur les hélicoptères autonomes de l'Onera, et l'avons testé en vol et en environnement réel sur une missionde détection et reconnaissance de cibles. / Mobile and aerial robots are faced to the need of planning actions with incomplete information about the state of theworld. In this context, this thesis proposes a modeling and resolution framework for perception and mission planningproblems where an autonomous helicopter must detect and recognize targets in an uncertain and partially observableenvironment. We founded our work on Partially Observable Markov Decision Processes (POMDPs), because it proposes ageneral optimization framework for perception and decision tasks under long-term horizon. A special attention is given tothe outputs of the image processing algorithm in order to model its uncertain behavior as a probabilistic observationfunction. A critical study on the POMDP model and its optimization criterion is also conducted. In order to respect safetyconstraints of aerial robots, we then propose an approach to properly handle action feasibility constraints in partiallyobservable domains: the AC-POMDP model, which distinguishes between the verification of environmental properties andthe information about targets' nature. Furthermore, we propose a framework to optimize and execute POMDP policies inparallel under time constraints. This framework is based on anticipated and probabilistic optimization of future executionstates of the system. Finally, we embedded this algorithmic framework on-board Onera's autonomous helicopters, andperformed real flight experiments for multi-target detection and recognition missions.