Designing Tomorrow's Category-Level Object Recognition Systems: An International Workshop

点击翻译按钮获取中文摘要

Despite 40 years of research, however, today's recognition systems are still largely unable to handle the extraordinarily wide range of appearances assumed by common objects in typical images. Our tenet is that fundamental new advances in the design and implementation of automated object recognition systems can be achieved by integrating the sophisticated geometric and physical image formation models developed in the computer vision community with the effective models of data distribution and classification procedures developed in the statistical learning theory and theoretical computer science communities. We propose to hold a three-day workshop that will bring together prominent computer vision, machine learning, and computational geometry researchers interested in the fundamental and applicative aspects of object recognition, as well as representatives of industry and funding agencies. Its goals are (1) to promote the creation of an international object recognition community, with common datasets and evaluation procedures, (2) to map the state of the art and identify the main open problems, design issues, and interdisciplinary research opportunities, and (3) to articulate the industrial and societal needs for object recognition technology worldwide. The workshop will be held in Taormina, Sicily, in September 2003. NSF support is sought to partially fund the travel of 25 US academic researchers to the workshop. Additional funds have been solicited from industry and INRIA (France) to support non-US participants and the rest of the workshop expenses. Intellectual Merit and Broader Impacts: The proposed workshop is intended to set the stage for the development of (a) unified design principles for visual object recognition systems, leading to a completely new set of capabilities, (b) fundamental advances in machine learning and computational geometry in a setting---somewhat unusual for these disciplines---where the geometric and physical origins of the input data are well understood, but an appropriate catalog of features for representing them remains elusive, and (c) the successful deployment and evaluation of the developed technology in high-impact application domains. In particular, the workshop will address fundamental scientific issues that are directly relevant to (1) industrial needs in areas as diverse as biomedical imaging, human-computer interface, telecommunications, and transportation, and (2) applications of strategic national interest such as surveillance and security, target recognition, and monitoring. A concrete outcome of the workshop will be a report outlining its conclusions and available to the community at large via the World Wide Web.