A method for estimating the volume of clusters built by Growing Neural Gas

Abstract

3D space perception is playing an important role in autonomous robots completing a task adaptively in the form of detecting target objects and estimating the 3D pose of target objects. This paper utilizes a growing neural gas (GNG) based method called GNG with different topologies (GNG-DT) for reconstructing unstructured point clouds. Next, for extracting a feature from clustering results, we propose a GNG based volume estimation method. Finally, we display a sequence of experimental results of the proposed method using simulation data sets and 3D point cloud datasets to evaluate the proposed method and discuss its effectiveness.