Active perception tasks driven by a cognitive simulation model

Our project of the cognitive simulation of a car driver aims at modelling precisely the activities of perception, decision and action of the human driver. We detail in this paper the functionalities of the perception module. Functioning alternatively on the modes top-down and bottom-up, the perceptive-cognitive human system alternates from the active seeking of pertinent information to the integration of unexpected events. While the first mode is mainly involved in action planning, the latter one is responsible for bringing into question the "driving frame" used by the driver. Our aim is to demonstrate here how these principles, inspired from the human model and applied to real data, constitute an original and integrative frame for representing, perceiving and analysing the road environment by the exteroceptive technologies. We also show that this cognitive architectures proposes circumstantiated and predictive means to drive perception, on the level of the physic sensors as well as on the algorithmic level.