Map Registration with 3D Enhanced Terrain Classification

The ability to accurately and reliably register locally sensed data to a global map is generally considered a useful capability though this appears to be an under-researched topic except in the context of registering aerial imagery. With data from a ground vehicle the task of registration is made significantly more complex because the sensor viewpoint is more revealing of the scene structure; the task is essentially a three-dimensional one rather than a two-dimensional one. Here we consider a registration approach based on the use of locally derived terrain classification data, a novel element of this approach being the use of 3D information, obtained by structure-from-motion algorithms, to aid the terrain classification. In this paper we briefly review the benefits of map registration. We then present the results of a first proof-of-principal experiment demonstrating the use of 3D data in terrain classification. Finally we discuss some larger system aspects of the requirements of handling map data and the registration process