The goal if this thesis is to develop cognitive and interaction capabilities for a mobile robot. To interact with its environment, a mobile robot needs to capable to interpret data from its sensors, amongst which the camera is one of the most important ones.
In this project, we focus on gesture and action recognition from videos taken from onboard cameras of small mobile robots. Several scientific problems are related to this task: (i) detection and segmentation of moving objects (humans, animals, and objects manipulated by them); (ii) the design of compact invariant and discriminant descriptors for this task; (iii) the design of a meta-model for the targeted objects and/or actions; (iv) the design of learning algorithms for this model; (v) the design of efficient detection and recognition algorithms.
The PhD has a strong constraint on the development of efficient algorithms allowing real time processing.