Fusion de données hétérogènes pour la perception de l'homme par un robot mobile
Thierry Germa · HAL (Le Centre pour la Communication Scientifique Directe) · 2010
This work has been realized under the CommRob european project involving several academic and industrial partners. The goal of this project is to build a robot companion able to act in structured and dynamic environments cluttered by other agents (robots and humans). In this context, our contribution is related to multimodal perception of humans from the robot (users and passers-by). The multimodal perception induce the development and integration of perceptual functions able to detect, to identify the people and to track the motions in order to communicate with the robot. Proximal detection of the robot's users uses a multimodal perception framework based on heterogeneous data fusion from different sensors. The detected and identified users are then tracked in the video stream extracted from the embedded camera in order to interpret the human motions. The first contribution is related to the definition of perceptual functions for detecting and identifying humans from a mobile robot. The second contribution concerns the spatio-temporal analysis of these percepts for user tracking. Then, this work is extended to multi-target tracking dedicated to the passers by. Finally, as it is frequently done in robotics, our work contains two main topics : on one hand the approaches are formalized; on the other hand, these approaches are integrated and validated through live experiments. All the developments done during this thesis has been integrated on our platform Rackham and on the CommRob platform too.