A victim identification methodology for rescue robots operating in cluttered USAR environments

摘要

Our research focuses on developing an automated victim identification methodology for rescue robots in order to aid robot operators with the complex and stressful task of searching for victims in cluttered urban search and rescue (USAR) environments. In this paper, we present an approach that utilizes 2D and 3D sensory information from a real-time 3D sensory system for robust victim identification using both human geometric and skin region features. Our technique, uniquely, allows for the identification of partially occluded victims and single body parts that may be visible in cluttered USAR scenes using a Support Vector Machine-based classifier based on the aforementioned features. Unlike other approaches that focus on the recognition of one specific body part (such as the head) or the recognition of a small set of fixed body poses, we aim to identify multiple different body parts in a number of varying configurations to increase recognition rate. Experimental results illustrate the robustness of our methodology to find human victims in a variety of different poses in a rubble-filled USAR-like scene and its ability to potentially reduce operator workload.