Integration method of local evidence for part-affordance estimation of everyday objects

Shapes of everyday objects are designed to achieve the predefined purpose of use. In this research, we call various kind of inherent functions as "part-affordance", and we have developed the method to perceive it from point cloud captured by a depth sensor. Difficulty of the estimation of it is that same local surfaces do not always have same affordance. In order to deal with this issue, we propose a method which integrates the evidence generated by local feature while considering the continuity of surface structure. Our experiments using a publicly available datasets confirmed that the proposed method have increased the recognition rate from 57% to 73% in comparison with the previous method. Moreover, we demonstrated that the robot arm can perform the task according to estimated part-affordances.