Information Sampling for Appearance based 3D Object Recognition and Pose Estimation
Niall Winters, José Santos-Victor · 2001
This paper is concerned with overcoming three problems associated with appearance based matching. The first is partial occlusion; the second background variation and the third is determining which image points from a set of images acquired a priori are the most discriminating. These data, either a single point or a number scattered throughout an image, are extracted by applying a statistical method we term Information Sampling. We show how to use the data yielded by Information Sampling to build Informative Local Appearance Spaces. Preliminary results indicate that our method achieves successful object recognition and pose estimation while overcoming the difficulties outlined above.