立体视觉匹配技术

立体视觉匹配技术是计算机视觉领域中最为关键的研究分支。根据匹配基元的不同,立体视觉匹配算法分为区域匹配、特征匹配和相位匹配三大类。其中,相位匹配是近二十年才逐步发展起来的一类匹配算法。以往关于立体匹配算法的综述文章对相位匹配这类新型算法几乎没有系统的阐述,而且很少从算法设计的角度分析和比较现有的立体视觉匹配技术。该文将算法设计过程分成三个步骤,在各个步骤中采用由个性至共性的分析手段,对三类立体视觉匹配算法进行了详细的评述,包括它们的理论依据、基本特性和改进策略的分析和比较,表明各类匹配算法的设计具有自身的发展特性。另一方面,“不适定”视觉问题普遍存在于各类匹配算法中,因此它们对算法设计的优化又存在着许多共性。文章通过由点至面的分析过程,旨在为算法设计者从综合思考的角度去优化算法提供技术借鉴,包括匹配基元自身缺陷的克服和普遍存在的不适定视觉问题的解决。此外,算法的完善和更佳算法模型的推出还依赖于科学的算法评价手段,文中根据不同用途对算法评价方法进行了分类,使算法性能的评估有了科学的指导方法。

On Stereo Matching Technology

Stereo matching is the most important branch in computer vision.Due to various matching primitives,the stereo matching approaches are divided into three categories,namely,area -based matching method,feature -based matching method and phase-based matching method.Phase-based matching is developed in recent two decades,which is seldom systematically involved in published survey articles on stereo matching technology.In addition,these articles did not analyze and compare the existing stereo matching techniques with a view to the algorithm design.This paper divides the design process of stereo algorithm into four stages.In each stage,three categories are analyzed and compared on some issues,including the innate properties,theoretical argument as well as innovative approaches related.It is demonstrated that three categories have their individualities and there is also commonness existing among them with respect to the development in algorithm design.This paper aims to supply a useful reference for algorithm designers.In addition,the evaluation criteria are classified according to their application purposes to give an important access to reasonable performance analysis of the designed stereo methods.