RIA: Detailed Three-Dimensional Model-Building from Image Sequences

9308415海豹：这是为期三年的持续研究启动奖的第一年资助。这项研究的长期目标是开发和实现从视频序列中建立完整的三维对象模型的新策略。高效的3D建模能力是移动机器人视觉任务的基础，例如学习新形状、操作复杂对象、绕过障碍物导航以及识别已知形状。该项目的一个中心焦点是分析使用由三维形状的平滑和非平滑分量形成的图像曲线在最小观看条件下从图像序列中恢复全局三维形状信息的问题。查看情况包括两个主要场景：主动反馈策略，其中使用图像序列中的信息来控制相机系统；以及未校准相机装置，其中相机参数未知。这项工作是基于对图像序列中投影曲线的行为及其与三维形状的关系的分析。这项研究将提供更深入的理论和经验理解，将导致开发有效的、实施的和测试的策略，以在最小观看条件下构建三维对象的详细模型。

9308415 Seales This is the first-year funding of a three-year continuing Research Initiation Award. The long-term goal of this research is to develop and implement new strategies for building complete models of three-dimensional objects from video sequences. Efficient 3D model-building capabilities are fundamental for mobile robot vision tasks such as learning new shapes, manipulating intricate objects, navigating around obstacles, and recognizing a known shape. A central focus of this project is the analysis of the problem of recovering global three-dimensional shape information from an image sequence under minimal viewing conditions using image curves that are formed from both smooth and non-smooth components of three- dimensional shapes. The viewing situations include two main scenarios: active feedback strategies where information in the image sequence is used to control the camera system; and uncalibrated camera rigs where camera parameters are unknown. This work is based upon an analysis of the behavior of projected curves in image sequences and their relationship to three-dimensional shape. This research will provide a deeper theoretical and empirical understanding that will lead to the development of efficient, implemented and tested strategies for building detailed models of three-dimensional objects under minimal viewing conditions.