RIA: Robust 3D Object Recognition Using Uncertain 2D Image Data

9309100个难民营今年是为期三年的持续奖励的第一年。该研究使用稳健统计和贝叶斯方法相结合的技术，从不确定的2D图像数据中建立用于稳健3D对象识别的数学模型，并开发匹配方案，该匹配方案使用这些模型来稳健地识别和计算对象的姿势，同时为所提出的假设提供置信度度量。这一结果应该能够使计算机视觉系统的开发能够处理实际输入数据中的缺陷，例如由于碎片或传感器噪声而导致的缺陷，而不需要求助于特别的试探法或试错程序。这样的系统应该提供更强大的性能并显著降低开发成本。

9309100 Camps This is the first year of a three-year continuing award. The research uses techniques from robust statistics coupled with a Bayesian approach, to develop mathematical models for robust 3D object recognition from uncertain 2D image data, and to develop matching schemes that use these models to robustly recognize and compute the pose of an object, while providing a measure of confidence level for the hypothesis made. The results should enable development of computer vision systems that can cope with imperfections in real input data, due for example to fragmentation or sensor noise, without resorting to use of ad hoc heuristics or trial-and-error procedures. Such systems should offer more robust performance and significant reductions in development cost.