3D matching by combining CAD model and computer vision for autonomous bin picking

Since today, most of the manufacturing companies have the operated CAD model, so CAD-based object autonomous bin picking which using 6DOF Manipulator may be a good option that can save time and increases productivity for an assembly line. This research aims to present an effectively autonomous method that can increases productivity as well as respond quickly of changing items based on customer demand for an assembly line which using 6DOF Manipulator by combining CAD data and computer vision system. Firstly, The 3D CAD model of grasped object is projected onto six different 2D planes, then combining six views to form the final pointcloud. Secondly, a voting scheme is used to estimated the 3D pose of object which is obtained by a 3D camera. For tuning a precision of an estimation such as surface normal, angle and location of an object, Iterative closest point (ICP) algorithm is applied. Before doing experiments, the recognition algorithm is verified through the simulation program. Through implement experiments, the system proved that it is stable, good precision and applicable in production line where mass product is produced. Moreover, the developed system allows non-expert users with basic knowledge about CAD drawing and image processing can generate a pose of an object from CAD model and transmit data to a manipulator for the bin picking task.