Tracking of high-speed and aggressively-maneuvered aerial targets by Kalman-Kernelized and Kalman-Discriminative Correlation Filters (K-KCF & K-DCF)

Abstract

In this work, we present an integrated target tracking method that combined the fast Kalman Filter for tracking initialization and the effective Kernerlized Correlation Filter (K-KCF) and Discriminative Correlation Filter (K-DCF) for tracking continuation. Both K-KCF and K-DCF are able to track the multiple high-speed and aggressively-maneuvered aerial targets, especially drones, even if the target may cross path to each other in camera view, in comparison to normal Kalman Filter (KF). We conducted test case and compare the results from individual KF, KCF and DCF, together with K-KCF and K-DCF combinations. Our initial results show both K-KCF and K-DCF perform better than those individual filters. For long term tracking, together with varying target (drone) sizes, K-DCF performs better as it can handle and track the drone more continuously with less discontinuties compared to K-KCF. Hence, K-DCF is the preferred choice for this tracking of fast and aggressively moving aerial targets applications.