Abstract

TAMPERE UNIVERSITY OF TECHNOLOGY

Department of Information Technology

Digital and Computer Systems Laboratory

Pekkarinen, Mikko: Multiple Model Approaches to Multisensor Tracking

Master of Science thesis, 87 pages, 15 enclosure pages.

Examiners: Prof. Jukka Saarinen, Lic.Tech. Petri Korpisaari, M.Sc. Jouko Saikkonen

Funding: Tampere University of Technology, Finnish Air Force

April 2000

Keywords: multisensor tracking, passive tracking, IMM, IAC, two-stage estimator

This thesis has been made in the Digital and Computer Systems Laboratory at Tampere University of Technology in cooperation with the Finnish Air Force. The research is part of a larger project where tracking techniques have been studied to create more expertise of the field in Finland.

Tracking is the process of estimating the position and velocity of targets of interest using measurements provided by some sensor system. A tracking system should be able to follow the targets even though the targets may maneuver, the measurements are inaccurate, the origin of the measurements is unknown and some measurements may be missing.

In this thesis the problem of tracking airborne maneuvering targets in clutter using angle-only measurements is considered. Three different tracking algorithms for maneuvering targets were studied, namely the Interacting Multiple Model algorithm, Interacting Acceleration Compensation and two-stage algorithms. The algorithms were adapted for handling passive measurements from multiple sensors in cluttered environment. For the two-stage algorithm a new maneuver detector was also devised. Another problem studied was the tracking of targets having very dissimilar kinematics, using a single model bank in IMM. Two model banks capable of tracking both stationary and maneuvering targets were designed.

The performance of the algorithms in heavily cluttered environment was evaluated with computer simulations. It was found out that the algorithms are unable to track maneuvers properly, but the performance compared to Kalman filter is improved when targets do not maneuver. The main reasons for the poor maneuver tracking were found to be the linearisation errors in measurement handling, and the weakness of the local data association in passive tracking context.