Information fusion for USAR operations based on crowdsourcing

Abstract

In this paper, we introduce automatic information fusion methods for the urban search and rescue (USAR) operations that efficiently “crowdsource” victim detection tasks. We reduce the load on the operators requiring them to acknowledge only presence of the victim in an image (to annotate the image). The task of finding victim location is performed via automatic fusion of annotated images from the image queue. This multi-robot information fusion is conducted continuously; as robots collectively explore larger areas, the estimated victim coordinates converge to actual victim locations.