Recognition of urban scene using silhouette of buildings and city map database

Abstract

This paper proposes an approach for understanding urban scenes by recognizing key buildings in silhouette using model-based object recognition scheme. A city map-database is used to build a world model, based on the information of location and orientation of the video camera measured by GPS. The general correspondence hypotheses between the line vectors extracted from the silhouette of ground objects and the world model are established by using dynamic programming technique. The general correspondence hypotheses are verified and modified according to the local position relations. The approach was successfully tested in recognition of the buildings in real scenes.