摘要
本文阐述了一个以线条特征为匹配基元的双目立体视觉系统。匹配算法以环境相似性度量为核心,以多层表示为基础,采用由粗而精的控制策略,通过预测、归并、分解、传播四过程买现。匹配算法考虑到实际环境中的双目立体图象的非理想性及抽取出来的线条特征的非完整性,允许某种程度的纵向偏差,引入中心主轴概念,在分解过程中建立对应,计算视差。首先由前三个过程建立起可靠的粗层次的对应,称之为骨架对应,然后利用连续性约束假设将骨架对应的结果由传播过程逐级传播到整幅图象中去。
An line-segment-based bibocular stereo vision system is illustrated in this paper. The matching algorithm is based on the region similarity criterion by a coarse-to-fine strategy in multilevel representation. The algorithm is composed of four steps: predicting, appending, splitting, and spreading. Coping with the uncertainty of real stereo images and the incompleteness of extracted features, the algorithm accepts vertical diversions, to some extent. A concept of the principal axis is introduced and used for matching which takes place in the step of splitting. A reliable skeleton matching can be built at the end of the first three steps then it is spread over the whole image sequentially.
出处
《自动化学报》
EI
CSCD
北大核心
1990年第4期340-346,共7页
Acta Automatica Sinica
