ECAI 2004 Conference Paper

[full paper]

A Skeleton Based Method for Efficient 3D Object Localization: Application to teleoperation

Djamel Merad, Narjes Khezami, Samir Otmane, Malik Mallem

Our aim is to develop a vision system for teleoperation to localize an object. This system has to be used through Internet connection. The recognition problem addressed in this paper is to localize a 3D free-form object from a single 2D view of 3D scene. Using a skeletonization process allows to obtain two graphs, the first one representing an object in the scene (2D skeleton) and the second one representing a database object (3D homotopic skeleton). The method encodes geometric and topological information in the form of a skeletal graph and uses graph isomorphism techniques to match the skeletons and find the one-to-one correspondences of nodes in order to estimate the object’s pose. Knowing skeleton is a set of lines centred within the 3D/2D objects, our method transforms the problem of free form object localization into points and lines pose estimation. Some experimental results on real images demonstrate the robustness of the proposed method with regard to occlusion, cluster, shadows …

Keywords: 3D free form object localization, 2D & 3D skeletonization, graph & sub-graph matching, Teleoperation

Citation: Djamel Merad, Narjes Khezami, Samir Otmane, Malik Mallem: A Skeleton Based Method for Efficient 3D Object Localization: Application to teleoperation . In R.López de Mántaras and L.Saitta (eds.): ECAI2004, Proceedings of the 16th European Conference on Artificial Intelligence, IOS Press, Amsterdam, 2004, pp.1061-1062.