3D sorround vehicle awareness and re-identification for autonomous driving applications

Vehicle tracking is a critical part in autonomous driving systems. Especially dense traffic patterns are particularly challenging when vehicles are being occluded or appear and disappear from the field of view. Humans have a situational awareness of what goes around their car, while extracting additional context information through the mirrors, they’re able to propagate local scene content from previous observations, and anticipate future traffic conditions. In this thesis we will focus on techniques of re-identifications and tracking across field of view of different cameras to create a consistent (3D) spatial and temporal representation of the traffic conditions. The work is also to be inspired and applied to the latest state-of-the-art approaches for pedestrian tracking. This research will be accomplished in collaboration with TRACE (Toyota Research for Autonomous Cars in Europe) Lab at ESAT, KU Leuven.