3D measurement and perception are getting actively participating in industrial applications, include but not limited to 3D modelling, 3D parts inspection, and bin picking. This work focuses on the design of a multi-modality structured light system used for 3D modelling and workpiece pose estimation. The proposed system is mounted on the end effector of an industrial robot arm as a hand-eye configuration. It is capable of scanning the target objects of various materials and sizes at sub-pixel accuracy. When it’s at resting mode, with only one camera on, the 3D pose of in-coming workpieces, which can be applied in the followup tasks, such as object grasping, picking, or tracking, will be estimated in a near real-time fashion based on the known CAD model.

For the 3D scanning part,...[ Read more ]

3D measurement and perception are getting actively participating in industrial applications, include but not limited to 3D modelling, 3D parts inspection, and bin picking. This work focuses on the design of a multi-modality structured light system used for 3D modelling and workpiece pose estimation. The proposed system is mounted on the end effector of an industrial robot arm as a hand-eye configuration. It is capable of scanning the target objects of various materials and sizes at sub-pixel accuracy. When it’s at resting mode, with only one camera on, the 3D pose of in-coming workpieces, which can be applied in the followup tasks, such as object grasping, picking, or tracking, will be estimated in a near real-time fashion based on the known CAD model.

For the 3D scanning part, a compact and highly customizable structured light scanner along with its software solution was made. It has built-in functions of camera-projector calibration, hand-eye calibration, multiple pattern codec strategies aim at different scenarios, classic point cloud registration, and 3D modelling result visualization. As for 6 DoF pose estimation, a statistical region-based segmentation model is used to mask the target object on the image, and together with given CAD, a level-set embedded cost function is defined. And a Gaussian-Newton-like pose optimization strategy is utilized to solve the best-estimated pose. This framework also shows its robustness when dealing with some challenging conditions, like textureless targets, cluttered scenes, and object movement.