Vision-based Drones Tracking Using Correlation Filters and Linear Integrated Multiple Model

This paper presents the use of Correlation Filters and Integrated Multiple Model (IMM) for filtering the position measurement of fast moving drones acquired by computer vision, with probability for model selection. The maneuvering movement of the drones are often non-linear making it hard to be estimated by a simple Kalman filter. Instead of using a non-linear filter which are more complex and non-universal, this paper attempt to integrate multiple filters to estimate the drone position using low computational cost. The IMM switches between the Constant Velocity (CV), Constant Acceleration (CA) and a Constant Turn (CT) model with a Markov Chain in different flight scenario depending on the drone movement. Other filters i.e. Kernelized Correlation Filter (KCF), Particle Filter (PF) and Discriminative Correlation Filter (DCF) models are also presented for direct comparison.