Bayesian Estimation of Snow-Avalanche Victim Pose: A Method to Assist Human and/or Robot First Responders to Quickly Locate a Buried Victim

Abstract

Finding a victim buried in a snow avalanche as quickly as possible can significantly increase the victim’s survival rate. A body-pose estimation algorithm is described that quickly and efficiently estimates the victim’s pose (3D location and orientation) underneath the snow. The algorithm exploits non-parametric Bayesian estimation and considers the uncertainty in an avalanche transceiver’s magnetic-field measurement. Simulation results compare the performances between three victim-search methods: (1) naive raster-scanning search, (2) traditional industry-standard search along the measured magnetic field lines, and (3) search by the Bayesian-based technique. The results show that the Bayesian-based technique accurately determines the victim’s pose within two minutes. In contrast, the raster-scanning and magnetic-field-line following methods yield search times more than three to four times longer.