Recognition of Artiicial 3-d Landmarks from Depth and Color | a Rst Prototype of Electronic Glasses for Blind People

In this paper we present a robust method for the recognition of 3D landmarks. Firstly the stereo and color information are extracted and combined together to acquire candidate 3D primitives of landmarks as super-segments, and then structured indexing is incorporated with pyramid and fuzzy methods. These approaches prove to be successful in the oo-line experiment for locating and recognizing many artiicial objects from a single view in indoor or outdoor scenes. The algorithms to the approaches have been implemented on a parallel computer system real-timely. A prototype navigation system upon these algorithms and the parallel system for blind people is sketched, its performance in the on-line recognition of landmarks is given nally.