Planar Object Recognition For Bin Picking Application

This paper aims to present a vision-based bin picking system for assembly line in industry. The objects are flat, unique color and occluded each other inside a bin. The whole picking process is divided into two stages. At 3D localization stage, an estimation process using 3D data segment by Euclidean algorithm and calculating surface normal are proposed to estimate angle and position of an object. To reduce the burden time of 3D pose estimation, a voxel grid filter is implemented to reduce the number of points for the 3D cloud of the objects. As known the 3D image in bin often involves both heavy noise and edge distortions, so to prepare for the assembly a 5DOF robot will pick and place it in 2D table then an 2D camera is used to estimate the pose of the object correctly. To prove the efficiency of proposed system that can pick up all objects in the bin a series of experiments on a 6-axis robot are implemented.