Vision-based spatial awareness for autonomous cars

Mirrors provide human drivers an awareness of what is going on around their car. This work focuses on creating a similar, 360 degree awareness for autonomous cars based on camera input. In order to achieve such spatial awareness, the information from multiple cameras (including large field-of-view fisheye cameras) will be fused, and properties of other traffic agents such as distance and speed relative to the autonomous vehicle will be estimated. Another part of this work will be aimed at improving the generalization of deep neural networks to previously unseen data.