Dr. Joseph S. Peri received his Ph.D. in physics (1978) from The Catholic University of America where he developed a model of heat transfer between liquid helium and solids. After receiving his degree, he worked at the Computer Sciences Corporation modeling periods of satellite communications and later at Andrulis Corporation analyzing a submarine tracking system. In 1981, he joined APL and has been involved in various tasks. He analyzed the performance of communications systems that transmit emergency action messages to submerged submarines and developed an age-dependent radar track promotion algorithm still being used in the Cooperative Engagement Capability (CEC). Dr. Peri has modeled infrared (IR) propagation through the atmosphere, and has predicted the performance of various IR systems. His current efforts are in support of the Network Centric Combat Identification (NCCID) project, the goal of which is target ID by fusion of data from various sensors.
Data Fusion & Target ID: Dempster-Shafer & Probability Theories Holy War
The Network Centric Combat Identification (NCCID) is a science and technology project sponsored by the Office of Naval Research. The goal of the project is to develop a method for networking multiple sensors and integrating observational data to provide target classification and identification (ID). Classification addresses the issue of the type of target observed; ID addresses the issue of deciding whether the target is friendly, hostile or neutral. The project will demonstrate the NCCID concept by using the existing Cooperative Engagement Capability to net various sensors and to provide composite tracks together with composite ID. To achieve this goal, a reasoning engine must be utilized for combining the various pieces of evidence to produce the target classification and ID. There are several possible mathematical techniques that can be used, among them the Dempster-Shafer (DS) theory and the Bayesian theory (i.e. probability theory). There has been considerable debate about the strengths and weaknesses of the two approaches and many have viewed these as competing theories. The holy war between the DS proponents and the Bayesian supporters seems to stem from misunderstandings of the various terms used in the two theories. The purpose of this talk is to show the close relationship the two theories share in a measure theoretic setting, and to provide some examples.