Multipurpose optimization of camera placement and application to random bin-picking

Due to increasing labor costs, the demand for factory automation (FA) continues to rise. Global demand also exists for industrial robots. Thus, autonomous robots that can perform advanced tasks are required in many scientific and technological fields. Although it is difficult for stand-alone robots to perform such random bin-picking tasks as grasping an arbitrary apple from a pile of them, we can solve this problem by installing external sensors in robots. However, the systems, which are built using expensive 3D cameras or laser range finders, are expensive. In this research, we construct a low-cost random bin-picking system using low-cost CCD cameras. Our experimental method focuses on camera calibration accuracy and determines the optimal camera placement of two cameras. First, we explore the effect of two camera placement on calibration accuracies. Next, we determine the optimal camera placement from an experimental technique based on camera calibration accuracy and a constraint condition. Finally, we verify the effectiveness of the camera placement by performing random bin-picking accompanied by searching and picking tasks. We determined optimal the camera placement that can distinguish overlapping objects. In addition, our low-cost picking system indicated the performance of random bin-picking operation for simple objects.