Real-time Qualitative Detection of Multi-colored Objects for Object Search

摘要

This paper considers the task of using a mobile camera to search for a specified object in a room. We call this the object search task. Introspection reveals that humans perform such tasks countless times every day. However, despite its pervasive nature, object search has been the target of little research effort. The paper begins with a general discussion of the object search problem. Given that the goal of object search is to reliably find the desired object with a minimum of effort, we identify three main abilities that any object search system must possess. These are the abilities to apply multi-stage object recognition strategies, to reason about occlusion, and to use high-level knowledge about spatial contexts to predict likely locations of objects. A preliminary implementation of object search is presented that illustrates a method for realtime detection of the presence of known multicolored objects in a scene. The method is based on the assumption that the color histogram of an image can contain object "signatures" which are invariant over a wide range of scenes and object poses. The resulting'algorithm has been easily implemented and used to build a robot that can direct its gaze over a room searching for an object.