Intelligent Vehicles

(semantic scene interpretation in a driving context)


In the context of the DrivSco project, we applied the Early Cognitive Vision framework to the problem of interpreting visual scenes in a driving context. This involved several components:

First, making use of the semantics attached to the visual primitives and the relations thereof in the Early Cognitive Vision framework, we designed a Bayesian driving scene interpretation approach [2,3,4]. This was concerned with the extraction of the road's lanes and driving relevant signs.

Second, one important information to be processed by a driver assistance system is other vehicles on the road. We approached this problem by combining an optic flow based independent motion detection technique with a sparse feature-based tracking and modelling of independently moving objects [1].

This is illustrated by the following video of the system's analysis of a typical driving scene.

This video of the system running was produced by Lars Baunegaard With Jensen at the University of Southern Denmark

This research was conducted at the Cognitive Vision Lab, of the University of Southern Denmark, as part of the European project DrivSco. Acknowledgements: Lars Baunegaard With Jensen, Emre Başeski, Bart Boesman, Younes Peter Touati, Karl Pauwels, Marc Van Hulle, Florian Pilz, Sinan Kalkan, Florentin Wörgötter and Norbert Krüger (see references for details).

References