Boosting with stereo features for building facade detection on mobile platforms

Abstract

Boosting has been widely used for discriminative modeling of objects in images. Conventionally, pixel- and patch-based features have been used, but recently, features defined on multilevel aggregate regions were incorporated into the boosting framework, and demonstrated significant improvement in object labeling tasks. In this paper, we further extend the boosting on multilevel aggregates method to incorporate features based on stereo images. Our underlying application is building facade detection on mobile stereo vision platforms. Example features we propose exploit the algebraic constraints of the planar building facades and depth gradient statistics. We've implemented the features and tested the framework on real stereo data.