Colored-Noise Tracking of Floating Objects using UAVs with Thermal Cameras

Abstract

Tracking of floating objects using a fixed-wing UAV equipped with a thermal camera requires precise knowledge about the position and attitude of the UAV. Errors in the navigation estimates reduce the accuracy of the tracking system. Navigation errors are usually correlated in time and can propagate colored noise into the tracking filter. This work analyzes two approaches that seek to mitigate colored noise and they are compared experimentally with a third approach which assumes that the noise in the tracking system is purely white. Two independent flight experiments have been carried out where a small marine vessel was used as target. Thermal images of the target were captured and the position and velocity of the target have been estimated in an Earth-fixed coordinate system only using the images. The results show that objects can be tracked with an accuracy of a few meters when measurements are available, and that the estimates do not drift significantly in periods without measurements. Moreover, the results demonstrate that colored noise need to be accounted for in the measurement model to estimate the covariance precisely and maintain filter consistency, which is critical in multi-target tracking.