Localization of freely curved pipes for bin picking

A method for detecting the position and orientation of freely curved pipes is proposed for the purpose of bin picking. We assume that the geometrical model of the object of interest is given and the bin of objects can be observed using a widely available low-cost 3D sensor. As freely curved pipes do not have distinguishable feature points, typical feature-based matching methods cannot be applied directly in this case. Although the iterative closest point (ICP) algorithm [1] is a suitable strategy for registering a 3D model with free-form surfaces to its observed 3D data, two difficulties arise when applying the ICP algorithm in the case where the background is filled with objects with the same shape. One of these is the initial estimation of the position and orientation of an object from among similar shaped point clouds. The other difficulty is robust point matching between the 3D model and the observed data belonging to a single object. To solve these issues, we investigated two key ideas: 1) estimation of pipe orientation by checking the shape of the cross-section of the observed 3D data while changing cutting directions of the sections; and 2) point matching based on ridge lines with respect to the camera direction. Experiments using actual observed data show the effectiveness of the proposed method.