Web mining driven semantic scene understanding and object localization

摘要

Knowledge acquisition from the Internet for robotic applications has received widespread attention recently. It has turned out to be an important supplementary or even a complete replacement to conventional robotic perception. In this paper, we investigate state-of-the-art online knowledge acquisition systems for robotic vision applications and present a framework for further fusion and tighter integration. Bootstrapped by an interconnected process wherein modules for object detection and supporting structure detection co-operate to extract cross-correlated information, a web text mining technique using sequential pattern retrieval is introduced for linking the search of objects with their potential localities. Experiments using an indoor mobile robot for an Active Visual Search (AVS) task demonstrate the benefits of our coherent framework for visual representation and knowledge acquisition from the Internet.