Real-time implementation of an asynchronous vision-based target tracking system in an unmanned aerial vehicle

Abstract

Currently, small unmanned aerial vehicles developed by NPS have been able to locate and track stationary and moving targets on the ground. New methods of continuous target tracking are always being developed to improve speed and accuracy, ultimately aiding the user of the system. This thesis describes one such method, utilizing an open loop filter as well as an external correction source: Perspective View Nascent Technologies (PVNT). While the PVNT correction can theoretically improve the accuracy from 20-30 meters to 1-2 meters, it does have a disadvantage in that the target position updates are delayed anywhere from 1-10 seconds. In order to account for the delay, an asynchronous filter is used to update the target position data given the external position correction from PVNT. Two cases have been tested including the general filter and one that utilizes a road model in the calculations. While an earlier thesis developed the basic simulation for the system, this thesis discusses improvements and corrections to the simulation model as well as the necessary steps for real-time implementation.