Crowdsourcing for Search of Disaster Victims: A Preliminary Study for Search System Design

Abstract

Teams of unmanned aerial vehicles (UAV) have been suggested as sensor platforms for disaster victim search systems used shortly after natural disasters such as an earthquake or tsunami. Previous efforts have used UAVs equipped with video cameras for the disaster information gathering stage, with the information processing stage performed by either a single human searcher or a victim detection computer vision algorithm. We propose extending these efforts by investigating how a large and distributed "crowd" of volunteers may augment the information processing stage by helping search video feeds for disaster victims. An experiment is conducted comparing the victim detection accuracy between a single human searcher, a crowd of searchers, and a victim detection algorithm. Our preliminary results show that while victim search accuracy is sensitive to both UAV altitude and crowd size per video feed, crowdsourcing the search process can be more accurate than a single human or victim detection algorithm alone. These findings are a first step towards optimizing search system design with respect to both information collection and information processing augmented with crowdsourcing.