Localization of pallets on shelves in a warehouse using a wide-angle camera

This paper describes a method of detecting and localizing pallets on shelves using a wide-angle camera mounted on a vehicle several meters away along shelves. In previous work, we proposed the generation of reprojected images in which pallets appear to have a constant shape and size regardless of their location on the shelves and the use of the images both for detection and localization with simple edge template matching. Preliminary experiments conducted in the laboratory showed a high potential for practical usage. The present paper proposes three improvements. First, a similarity measure for binary patterns is newly added, aiming at robust pallet detection in real scenes. Second, a maximum similarity search based on the expected tangent of the ridge-shaped peak is proposed for the efficient refinement of localization. Third, effects of round corners (i.e., R-corners) of pallets are considered for higher accuracy in localization. Experiments were conducted using images taken in a real warehouse to examine the detection and localization ability of the proposed method. Despite having small areas in the original wide-angle view, pallets of interest were well detected in complex real scenes. Localization accuracy is discussed from the viewpoint of practical use.