Approximation Probability of Detection in the Janus Model.

Abstract

This report documents our approach at developing a satisfactory detection model for use in computing the Janus Wargame. Information gain measures the Blue forces' awareness of Red's disposition over time. Within a time interval (t, t+l) the measure is a distance measure between two probability distributions Pt and Pt+ 1 respectively. These distributions represent the relative discrete probability, from Blue's perspective, that a Red vehicle is in a particular area of the battlefield. The sum of the discrete probability values over all areas is 1.0 with those areas of greatest likelihood having the larger values. Information is generated by the actions of Blue sensors. When any Blue sensor scans an area of the battlefield Blue gains information about the enemy disposition. The magnitude of this new information is determined in part by the efficiency of the Blue sensor. As new information about the location of a Red vehicle becomes available to Blue we update the probability distribution using theoretical formulations based upon Bayes formula 4. Though the theory is simple, its implementation in the Janus model was very challenging. The Bayesian formulation mentioned above requires two parameters during each time stage: 1) a listing of which areas of the battlefield (cells) Blue sensors looked in and 2) the probability of detection (PD) for the sensors that did the respective scanning. Neither of these data are directly available in Janus runs, nor can they be deduced from Janus output files. For example, the Janus algorithms for line of sight computations and detection of enemy vehicles are only called when two opposing vehicles are within some threshold of proximity to each other. Since information gain credits finding where the enemy is not, we need to know at each time increment what terrain cells Blue sensors have searched regardless of the presence or absence of enemy vehicles.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Jun 01, 1997
Accession Number
ADA350243

Entities

People

  • E. T. Sherrill
  • Mickey A. Sanzotta

Organizations

  • United States Military Academy

Tags

Communities of Interest

  • Materials and Manufacturing Processes
  • Sensors

DTIC Thesaurus Topics

  • Aircrafts
  • Algorithms
  • Battlefields
  • Computations
  • Curve Fitting
  • Data Sets
  • Detection
  • Detectors
  • Ground Control Stations
  • Ground Vehicles
  • Normal Distribution
  • Operations Research
  • Probability
  • Systems Engineering
  • United States
  • United States Military Academy
  • Vehicles

Readers

  • Chemistry (specifically Chemical Fluorescence)
  • Sensor Fusion and Tracking Systems.
  • Statistical inference.

Technology Areas

  • AI & ML
  • AI & ML - Bayesian Inference
  • AI & ML - DoD AI Strategy