Model-based 3D pose estimation for pick-and-place application

Abstract

Due to the recent development of industrial automation, some applications have been improved with computer vision techniques. One important task is to recognize and estimate the 3D pose of the object in the scene. In this work, we use a depth camera to capture the 3D information of a scene, and proposed a 3D pose estimation algorithm. A main difficulty of the 3D object recognition and pose estimation is the captured data may have noise from the environment light, shadow or sensors. In general, the reference model and target model are captured from the same depth camera, so they will have similar data structures. However, in our work, we consider the target model generated from Computer-Aided-Design, and the reference model is captured from the depth camera. The data from different sources will cause the estimation error. In this work, we have addressed this problem. Finally, we develop the simulation system for our proposed method, and also simulate a manipulator to accomplish the pick-and-place task.