CNN 방식과 전통적 방식의 스테레오 매칭 알고리즘 비교

Abstract

CNN has been widely used in computer vision. In particular, CNN has been applied in stereo matching, which involves finding pixels corresponding to each other in a pair of images that capture a scene at the same time. This paper focuses on analyzing stereo matching methods based on CNN, and compares them with a traditional stereo matching method. CNN-based methods produced good matching results for ill-posed areas, including textureless areas, repeated patterns and highly reflective surfaces. By contrast, traditional stereo matching methods could not produce good matching results for such ill-posed regions. Therefore, in this paper, we demonstrate how good CNN-based methods could produce matching results for both indoor and outdoor images.