A novel stereoscopic cue for figure-ground segregation of semi-transparent objects

Computer Vision Workshops(2011)

引用 30|浏览5
暂无评分
摘要
The visual perception of semi-transparent objects, such as drinking glasses, is an open challenging problem. Unlike opaque objects, semi-transparent objects violate many of the standard vision assumptions, among them that figure-ground segmentation contains salient boundaries. More specifically, reliable motion and stereo cues for segmenting semi-transparent objects are not present because of the infeasibility of establishing correspondence. This paper describes a new discovery that semi-transparent objects are salient on the plane-parallax image generated by the inverse perspective mapping. A novel cue is introduced that reveals objects extruding from a planar support surface. Points on the support plane are consistent with a planar homography transformation, whereas extruding points from textured surfaces violate this mapping. Furthermore, extruding semi-transparent objects violate the mapping due to the refraction of light and strong specularities. The utility of this new cue is demonstrated in a novel detection and localization approach, where the cue is matched to a database of 3D models of semi-transparent objects. Preliminary empirical results suggest that the presented approach produces a small set of candidate locations for semi-transparent objects and yields accurate localization.
更多
查看译文
关键词
image motion analysis,object detection,stereo image processing,figure-ground segregation,inverse perspective mapping,motion cue,object detection approach,object localization approach,planar homography transformation,planar support surface,plane-parallax image,semi-transparent object segregation,stereoscopic cue,vision assumption,visual perception
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要