Abstract
We propose an automatic object modeling framework composed of a network of cameras distributed in different places of the house. Our system can automatically construct multi-appearance object models by simply observing when humans move objects from one place to another as part of their daily activities. Our approach consists in placing vision nodes in places where objects are commonly located, i.e. table, desk, kitchen, etc. When an object is moved from one place to another and vice versa, the system automatically discovers the object and gradually acquires different appearances as more visual information of the object is presented. We propose an aspect selection algorithm that avoids collecting unnecessary appearances by selecting only the ones that contain relevant visual information.
