SAR<em>Bayes</em>: Bayesian Models for Search & Rescue
Home Blog Projects Methods Downloads Credits Links

Introduction

SARBayes is a small research project organized by Charles Twardy. Our goal is to generate accurate probability maps, and use them quickly, automatically, and well to find missing people on land more quickly. As the name suggests, we use a Bayesian approach to modeling and statistics.

We develop models and algorithms to support missing-persons searches on land, using data that we and others have collected about missing-person behavior and search operations. In 2008 the large ISRID database become available, and in 2009 we have made a copy clean enough for machine learning.

Some of our past projects include:

  • collecting data on Australian searches into an online database
  • making and testing Bayesian network models to predict lost-person behavior
  • developing a general-purpose library for optimal resource allocation (SORAL)
We have discovered by experience that we are not well-suited to end-user software development, but we have developed some prototypes, and are always looking for real programmers.

Search is a classic case of Reasoning Under Uncertainty and the core of the problem is generating and maintaining a probability map for the current location of the lost person. Search theory and optimal resource allocation presume that such a map exists, but there has been no good way to make one for land search.

The database of lost-person incidents is one approach to generating such a probability map — by matching current case data to a statistical profile generated from past incidents. We use Bayesian networks to turn the bare statistics into dynamic models, and hopefully improve on the very rough statistical profiles now used. There are other approaches besides case-history, such as modified random walk simulations, and of course, expert scenario analysis at the scene. An ideal system would combine all of these approaches. And Bayesian statistics provide a clear and principled way to do that too.

Our goal is to be of real use to the SAR community. We expect our models to beat the current state of the art, and we hope to contribute to software that makes it possible to use those models on live searches.

History

SARBayes began in late 2000 in the Reasoning Under Uncertainty Group, a part of the Monash Data Mining Centre) at Monash University in Melbourne, Australia, with the cooperation of the Victorian Police Search & Rescue Squad, and VicWalk's Bushwalkers' Search & Rescue. The project has received invaluable assistance from other individuals and organizations, especially Bob Koester, Jack Frost, and Alan Washburn.

In 2005, Charles returned to the U.S. on a SBIR grant with Jim Donovan and Bob Koester, and finished the Australian Lost Person Behaviour Report (2006). Those cases helped to prototype the ISRID database. In 2008, Charles joined George Mason University as research faculty, where he hopes to create an active research project again. Online collaboration is also welcome.

Comments are welcome (remove the underscores): c_t_w_a_r_d_y at sarbayes.org

News

2009: Use the Blog: The news here was pretty old. The navbar now links to the blog, which is updated occasionally.

22 Feb 2007: CVSDude for SORAL: The SORAL source is going online with its own trac site on CVSDude. Bug-tracking and Wiki are up. Code to arrive soon.

Key Links Washington State and New Zealand (2004-2005 data). Also, the 2006 Australian LPB report, data (with identifying info removed) and Python programs are under Downloads. SORAL: see the Downloads pages.


SARBayes Website
© Charles R. Twardy and the SARBayes project, 2003-2007.
Page design by David Stokes
This page last modified May 30, 2009
Last modified: Wed Nov 12 15:20:25 EST 2003