Projective ICP and Stabilizing Architectural Augmented Reality Overlays

Summary

The recently developed technique of Iterative Closest Point (ICP) matching has been used for a number of 3D-to-3D and 2D-to-2D point matching applications, and has been further developed in several useful ways, as described below. Central to these applications is the notion of rigid shape matching. This paper extends the concept to projective point matching, in which shapes are related by a projective transform rather than a Euclidean transform. With this extended technique, we show that directly registering 2D Augmented Reality (AR) overlays via a projective transform has greater registration stability than the more usual technique of estimating the 3D position of the overlay and then applying pinhole projection, which can produce noticeable frame-rate jitter of the graphical objects. Moreover, the technique does not rely on explicit feature point correspondence and tracking. We then further extend the technique to directly register 3D shapes projectively by using mutually constrained 2D projective mappings. These two new techniques enhance the repertoire of methods for producing high-detail, stable augmentation of built scenes.