Extending Semantic Edge Labelling.

In earlier work [9, 10], we developed an approach to edge labelling and depth reconstruction based on fusion of registered depth and intensity images. This paper presents a new method and extensions to the first stage of that approach, that is the detection and semantic labelling of the edge data. Reflectance edges are extracted from an intensity image, and quadratic curves are fitted to range and intensity data along each row and column using a intensity guided splitting and merging process. Edge labels are determined by examination of the changes in depth, surface orientation and intensity at each edge site.