Development of a system to infer and visualize the behavior of search and rescue dogs based on inertial sensors

Abstract

We propose an advanced system that infers and visualizes the behavior of search and rescue (SAR) dogs while they are searching for victims in disaster sites where human cannot enter. Our system robustly infers SAR dogs’ behavior using machine learning methods with inertial data obtained from sensors attached to them. It automatically estimates the probability of each of the behavior candidates “running”, “walking”, “stopping”, “barking”, “sniffing some objects”, and “sniffing the air,” and presents the most likely one in real time. We also developed a system that visualizes the results of inference for handlers and rescue supervisors’ convenience.