Modeling and Recognizing Curved Surfaces from Image Outlines

点击翻译按钮获取中文摘要

Recognizing objects from a single view is hard, because their appearance depends crucially on the direction from which they are viewed. Curved surfaces are particularly subject to this effect. This research concentrates on recognizing curved surfaces from their outline in a single perspective image, using both theoretical studies and practical implementations, along two lines of investigation. The problem has already been solved for generic algebraic surfaces by the Principal Investigator. The result indicates that an image outline contains substantial amounts of information about an object's shape. The research will investigate extending this result to a wider range of popular surfaces, and will also investigate the practicalities of applying the resulting algorithms. Recent work by the PI and colleagues suggests that there is a large range of surface models which are intrinsically easy to recognize. These results suggest building a surface modeller oriented specifically towards computer vision applications, where the resulting models would be a fortiori easy to identify from any view. The research will investigate theoretical issues involved in constructing such a modeller, and will implement simple modelling system with these properties.