Neural Information Processing

Rapid Object Recognition

While generic object detection is still a hard task for computers, humans are able to detect known objects very rapidly even in cluttered environments. In this recent project we use models of biological vision, knowledge about human search and detection strategies and inductive learning techniques to develop computer vision systems that allow the rapid detection of known, but arbitrary objects in known, but arbitrary environments. For this purpose we investigate techniques for the specification of both objects and environments, as well as attention-driven algorithms which allow rapid detection. We will combine algorithm development with experimental work on human perception.

Acknowledgements: This work is funded by BMBF and the Technische Universität Berlin.