项目名称: 基于多尺度拓扑匹配流的宽基线基础矩阵估计方法研究
项目编号: No.61273279
项目类型: 面上项目
立项/批准年度: 2013
项目学科: 自动化技术、计算机技术
项目作者: 田金文
作者单位: 华中科技大学
项目金额: 80万元
中文摘要: 宽基线条件下基础矩阵估计在三维场景重建、视觉导航、智能交通和驾驶等计算机视觉相关领域具有广泛应用。针对宽基线基础矩阵估计中存在的两个难点,外点过多以及数据退化,本课题提出研究基于多尺度拓扑匹配流的宽基线基础矩阵估计方法。研究集成多种局部几何约束的自适应尺度选择的拓扑聚类方法,并提出匹配流思想,构建匹配流概率表达模型,研究有效的矢量域学习算法来估计图像对的全局匹配流。进而集成自适应尺度拓扑聚类的局部约束能力以及匹配流的全局泛化能力,研究拓扑匹配流去除外点算法。为了解决基础矩阵估计中的退化问题,研究匹配集合的多尺度拓扑匹配流数据表达模型,设计灵活的分层匹配流形结构,建立合适的退化判别准则,探索鲁棒的交叉采样策略,从而有效避免退化的出现。最后设计可靠的测试数据对基础矩阵估计算法进行实验评价,并探索其在三维重建中的应用研究,验证其实际应用性能。相关研究成果不仅理论意义重大,也具有潜在的应用前景。
中文关键词: 宽基线;图像匹配;基础矩阵;匹配流;拓扑聚类
英文摘要: Fundamental matrix estimation for wide baseline stero can be extensively applied to 3D reconstructure, robot vision navigation, intelligent traffic monitoring and driving. There are two difficulties in wide baseline fundamental matrix estimation. The one is due to too many outlines, and the other is due to data degeneration. In this project the wide baseline fundamental matrix estimation method is studied. Based on the previously developed mismatch eliminating algorithms based on topological clustering by us, multiple local geometrical constrains are integrated to remove the mismatches in the original topological clustering. And the scale of clustering can be adaptively selected. The match flow idea is presented in this project and the corresponding probability model is constructed. The effective vector field learning algorithm will be developed to estimate the global match flow to remove the outlines. To overcome the data degeneration when estimating fundamental matrix, we will develop multiple scale topological matching flow data representation model of match collection, design flexible hierarchical matching flow structure, construct suitable degeneration principle, and seek for rubost interaction sampling strategy. Therefore, the degeneration in fundamental matrix estimation can be avoided effectively. The re
英文关键词: wide baseline;image matching;fundamental matrix;match flow;topological clustering