Shared Feature Object Detection

点击翻译按钮获取中文摘要

The goal of this research is to improve the ability of computers to recognize objects in the world. That ability is useful for a range of robotic, surveillance, and image interpretation tasks. Procedures will be developed that should allow for a dramatic scale-up in the numbers of different objects a computer can detect, as well as greatly expanding the range of clutter and adverse lighting conditions under which the computer can detect those objects. The key advance to be explored and developed is the sharing of visual features across different objects, as well as other imaging dimensions of viewpoint, lighting, scale, and position. This feature sharing should dramatically improve the speed and efficiency of detecting multiple objects, as well as improve generalization performance. A set of labeled image databases will be developed that will allow researchers to train object recognition algorithms. The techniques developed may be useful in such diverse areas as robotics, manufacturing, and surveillance.