Vision Based Bin Picking Method Using Hierarchical Image Analysis

In this paper, we describe vision based bin-picking system for robot application using multiple local features, which are extracted from single camera. Multiple features are extracted from texture on object surface that has used to estimate surface rotation angle and distance to object to be picked. Challenging problem is to estimate accurate picking point and distance to object. It is difficult to solve aforementioned problem because the distorted image affected by illumination has caused reflection on the image surface or feature data loss of texture. In this paper, we proposed hierarchical analysis using multiple cues by multi-resolution images to estimate picking points of piled objects in the bin. The estimation of object location by coarse image and picking point by fine image have processed. We have tested to evaluate performance on ETRI database, which have captured under various lighting condition in the pilot system, which is constructed like industrial environment.