Object Recognition in Multisensory Scene Analysis

Abstract

A perception strategy for recognizing an object on the basis of color and range data is illustrated by finding a desk in an office scene. Geometrical models of objects (such as a desk, a chair, and a glass) are described by transformation (using scaling, rotation, and translation matrices) of primitive surfaces (unit cube, sphere, and cylinder). An image-bounding window for the acquisition element of an object is based on range data. Image bounding windows are derived for four cases in which single and multiple acquisition elements are oriented vertically and horizontally.