ビンピッキングのためのランダムフォレストによるバラ積み部品の3次元位置姿勢推定

Abstract

Listen

Translate article

This paper proposes a pose estimation method for random bin picking with random forest. Main contribution of the method consists of two parts. First, we provide novel 3D feature called cube based feature (CF) and pairs with segment based feature (PSF). To compute the proposed feature, it is not necessary to define the local coordinate system. Thus, the proposed 3D feature is strongly robust against measurement noise, which causes difficulty computing the local coordinate system with high repeatability. Second, we propose to utilize the random forest for correspondences finding between point cloud of scene and a model. Experimental results show that the proposed method can achieve high accuracy pose estimation of industrial parts.