Modeling and Recognizing Three-Dimensional Curved Objects

这是该奖项为期三年的第一年。本研究探索计算机视觉对三维曲面物体的识别。将开发一种新的对象建模和识别技术，基于用离散的点网格表示对象，并将网格映射到球面表示。本项目开发的算法将匹配和姿态确定问题减少为在两个球面地图之间找到最佳旋转的问题之一。这在不使用任何组合搜索的情况下完成。这些算法为目标识别提供了一种鲁棒性和计算确定性的方法。通过消除传统方法施加的大多数形状限制，它们使物体识别在更广泛的应用中成为可能。最后，通过允许在存在遮挡的情况下进行形状匹配，它们消除了基于球面表示的现有方法的主要限制。

This is the first year of a three-year continuing award. The research explores computer vision for recognition of three- dimensional curved objects. A new technique for object modeling and recognition will be developed, based on representing objects by discrete meshes of points, and mapping of the meshes to a spherical representation. The algorithms developed in this project reduce the problem of matching and pose determination to one of finding the best rotation between two spherical maps. This is done without using any combinatorial search. These algorithms lead to a robust and computationally deterministic approach to object recognition. By removing most of the shape constraints imposed by conventional methods, they make object recognition possible in a wider range of applications. Finally, by allowing for shape matching in the presence of occlusions, they remove a major limitation of existing approaches based on spherical representations.