Robust Multiview Registration of 3D Surfaces via `1-norm Minimization

Registration (or alignment) of surfaces is an important task in many areas such as robotics, computer aided modeling, virtual reality and surface reconstruction. Building a full model of a object requires registering multiple surfaces observed from different viewpoints. In this paper we present an approach robust to outliers, for registering multiple surfaces simultaneously when the transformation is rigid. Multiview registration simultaneously registers multiple scans. Given scans S1 . . .SN and a view-graph (i.e. a graph with edges between scans which have common points), G = ({1 . . .N},E). This problem can be generalized as