Development of Structured Light Based Bin–Picking System Using Primitive Models

As a part of factory automation, bin-picking systems perform pick-and-place tasks for randomly oriented parts from bins or boxes. Conventional bin-picking systems can estimate the pose of an object only if the system has complete knowledge of the object (e.g., as a result of the geometric features of the object being provided by an image or a computer-aided design model). However, these systems require the features visible in an image to calculate the pose of an object, and they require additional setup time for an operator to register the reference model every time when the workpiece is changed. In this article, we propose a structured light based bin-picking system using primitive models with small amount of prior knowledge. To obtain a reliable 3D range image for comparison with conventional systems, we use a structured light sensor with gray-coded patterns. With the 3D range image, the pose of the object is estimated with the use of primitive segmentation, rotational symmetric object modeling, and recognition. Through experiments using an industrial robot, we validate that the proposed method can be employed for a practical bin-picking system.