Pose Estimation of an Industrial Pallet of Packaged Paper Rolls from a Depth Image

In automated warehouses knowing the pose of pallets is essential for accurate navigation of Autonomous Guided Vehicles (AGVs). This work presents an algorithm that determines the pose of a pallet from a depth image acquired by an on-board sensor. The problem is challenging because we consider pallets that contain layers of packaged paper rolls, that are plastic wrapped non-rigid objects. The method assumes that the nominal layout of the packages is known in advance. The algorithm extracts features of the pallet from the depth image, like faces and edges, and it applies a non linear optimizer to compute the pose of the pallet. Experiments were carried out on depth images of plastic-wrapped packages in several configurations.