Attribute Based Affordance Detection from Human-Object Interaction Images

摘要

The detection of functional classification of an object, which is also referred as affordance is a prevalent researched topic in the domain of robotics and computer vision. Typically, the approaches regarding fine level affordance (affordance related to core traits of an object i.e. gras-pability, reliability etc.) detection are often disjoint from the techniques in higher level affordance detection (i.e. drinkability or pourability of a glass). In this paper, we have proposed an attribute based technique for higher level affordance detection which integrates methods from both fine level and high level affordance detection, and takes three prominent contexts (Human, Object and the ambience) into account. It further represents each of these contexts as a cluster of attributes rather than singular entities thus making the affordance detection process more semantic, efficient, dynamic and general.