engleski

A comparison of a curve fitting tracking filter and conventional filters under intermittent information

This technical report accompanies the tracking filter based on geometric properties of targets' maneuvers developed in our submission for the Journal of Autonomous Robots. Herein, we compare estimation quality, processing load, scalability and complexity of our tracking filter with Unscented Kalman Filter (UKF) and Particle Filter (PF). The filters have to estimate targets' positions given limited knowledge and intermittent detections of the targets, and measurement noise. It is shown that our filter requires substantially less information and processing power than UKF and Sampling Importance Resampling PF (SIR PF) while providing comparable tracking performance in the presence of limited information and noise. In addition, corroborating numerical simulations are provided.

curve fitting; adaptive sampling; unscented kalman filter; sampling importance resampling particle filter

nije evidentirano