The Factorization Method for Image Sequence Analysis

这是为期三年的连续奖项的第一年。这项研究的长期目标是在视频速率下从电视图像中恢复准确的摄像机运动和密集的三维形状信息。在以前的工作中，首席调查者开发了一种基于矩阵的因式分解方法，基于密集图像序列的一帧到一帧跟踪的一组特征点。实验表明，与现有的形状和运动恢复系统相比，该方法的性能有了显著的提高。这项研究现在将发展一种图像序列分析的噪声敏感性的表征，探索高效和增量的因式分解的数值方法，并重新制定透视投影、多运动和密集形状结果的方法。作为一个新研究领域的探索性进展，还将对运动分析和多帧目标识别之间的联系进行调查。这项研究应该有助于更好地从理论上理解视觉运动分析，并产生一种视觉模块，使机器人能够在环境中定位自己，绘制自己周围环境的地图进行导航和避障，并感知物体的形状，以便识别或操作它们。

This is the first year of a three-year continuing award. The long-term goal of this research is the recovery of accurate camera motion and dense three-dimensional shape information from television images at video rate. In prior work, the Principal Investigator developed a matrix-based factorization method for this task, based on a set of feature points tracked from frame to frame of a dense image sequence. Experiments with the method demonstrated a dramatic performance improvement over existing shape and motion recovery systems. This research will now develop a characterization of the noise sensitivity of image sequence analysis, explore efficient and incremental numerical methods for factorization, and reformulate the method for perspective projection, multiple motions, and dense shape results. As an exploratory advance into a new area of research an investigation of the link between motion analysis and multi- frame of object recognition will also be conducted. This research should contribute to a better theoretical understanding of visual motion analysis, and produce a vision module that would let a robot localize itself in the environment, draw a map of its own surroundings for navigation and obstacle avoidance, and perceive the shape of objects in order to recognize or manipulate them.