This paper addresses the issue of intelligent sensing for advanced robotic applications and is a continuation of our research in the area of innovative approaches for automatic selection of regions of observation for fixed and mobile sensors to collect only relevant measurements without human guidance. The growing neural gas network solution proposed here for adaptively selecting regions of interest for further sampling from a cloud of sparsely collected 3D measurements provides several advantages over the previously proposed neural gas solution in terms of user intervention, size of resulting scan and training time. Experimental results and comparative analysis are presented in the context of selective vision sampling.