Skeletonization and 3D graph approach for thin objects recognition in pick and place tasks
Pierre Willaume, Pierre Parrend, Etienne Gancel, Aline Deruyver · 2017
Objects retrieval is an important task in pick and place processes, especially when the objects are provided in a disorganized set. However, the methods proposed in the literature are not suitable for thin objects because only key points or edges of the shapes are studied, whereas in real cases, the objects can be thin and without special features. In this paper, a method based on graph comparison between stereoscopic images and a model is proposed to retrieve thin featureless objects and without having a long pre-processing time. Firstly, a thinning algorithm is applied to extract the skeleton of both shapes. Secondly, the skeleton is translated into a graph to extract the significant information of each part of the skeleton. Finally, the distances between the nodes are computed and the topography of the graph is analyzed to retrieve the desired objects. A comparison has been realized with a key-points extraction algorithm.