Selective Vision Sensing with Neural Gas Networks

Vision sensing systems are experiencing an unprecedented growth in numerous applications. The collection of such a rich flow of information has brought a new challenge in the selection of only relevant features out of the avalanche of data generated by the sensors. This paper presents some aspects of our research work on intelligent sensing for advanced robotic applications. The main objective of the research is to design innovative approaches for automatic selection of regions of observation for fixed and mobile sensors to collect only relevant measurements without human guidance. A solution using neural gas networks has been investigated to adaptively select regions of interest that require further sampling from a cloud of 3D measurements sparsely collected. The technique automatically determines bounded areas where sensing is required at high resolution to accurately map 3D surfaces. It provides significant benefits over brute force strategies as scanning time is reduced and datasets size is kept manageable. Experimental evaluation of this technology is presented for 3D surface sampling/sensing.