Unsupervised learning for improving efficiency of dense three-dimensional scene recovery in corridor mapping

Abstract

In this work, we perform three-dimensional scene recovery from image data capturing railway transportation corridors. Typical three-dimensional scene recovery methods initialise recovered feature positions by searching for correspondences between image frames. We intend to take advantage of a relationship between image data and recovered scene data to reduce the search space traversed when performing such correspondence matching. We build multi-dimensional Gaussian models of recurrent visual features associated with distributions representing recovery results from our own dense planar recovery method. Results show that such a scheme decreases the number of checks made per feature to 6% of a comparable exhaustive method, whilst unaffecting accuracy. Further, the proposed method performs competitively when compared with other methods presented in literature. © 2011 Springer-Verlag.