测绘学报

• 学术论文 • 上一篇    下一篇

基于向量匹配的稀疏深度图生成算法

孙敏1,胡争2   

  1. 1. 北京大学遥感所
    2. 中南大学信息物理工程学院,北京大学遥感所
  • 收稿日期:2009-09-14 修回日期:2010-03-29 出版日期:2011-02-25 发布日期:2011-02-25
  • 通讯作者: 孙敏

Vector Based Sparse Depth Map Algorithm

  • Received:2009-09-14 Revised:2010-03-29 Online:2011-02-25 Published:2011-02-25

摘要: 在基于序列影像获取三维空间数据的方法中,人们需要提取影像中每个像素的三维值,但由此得到的三维数据用在三维地理信息系统(3DGIS)建模时需要进行大量简化,同时当影像中存在非特征区域时,易产生大量的错误匹配。本文提出一种基于向量的匹配算法,将图像看成是一个连续的三维曲面,主要以特征区域匹配为核心,通过计算每个曲面点处的法向量,再利用该向量进行匹配。匹配的代价因子使用匹配窗口中对应向量的余弦值之总和。向量的变化反映了影像纹理的特征,从而可以利用向量回避纹理缺乏区域的错误匹配,同时可加快计算速度。论文最后给出的实验部分表明本算法能有效地生成稀疏深度图,可以满足3DGIS中的建模需求。

Abstract: In the method of 3D spatial data capturing based on the image sequences, people want to extract 3D data for each pixel, but the data extracted from this way need to be greatly simplified if used for 3D modeling in 3D Geographical Information System (3DGIS). On the other hand, if sparse texture areas exist in the images, it’s difficult to overcome the error matches. This paper presents a vector based matching algorithm. The algorithm takes one image as a continuous 3D surface, and mainly matches feature areas by computing vectors in each normal vector on the surface and matching corresponding vectors. The summary of cosine values of corresponding vectors and feature description value of the corresponding areas in two match windows are adopted as the cost factors. The changes of vectors reflect the image texture information, which can be used to evaluate image feature areas, filter sparse texture areas to avoid mismatch and save computer time. The results from experiment in the last part of the paper indicate that the proposed algorithm is effective for sparse depth map producing and can satisfy 3D modeling in 3DGIS.