Adaptive Motion Tracking for the Spiri UAS

Remote controlled helicopters or drones, specifically the ones with several rotating blades, have received lotsof attention in the last few years due to their advanced navigational abilities. Many new applications, includingcommercial applications, are now being considered. To enable such applications, the system must becomemore autonomous, meaning that they can act with little or no operator input. One such autonomous ability isfollowing an objects, which is called tracking.This project is about building an advanced tracking system for a quadcopter developed in Halifax called Spiri.Such a system is also more formally called an unmanned areal system (UAS). In order to reliably track anobject with such a device that carries cameras, one needs a very good recognition system for this object, whichin itself is a big challenge. Luckily, computer vision has made much progress so that several new algorithmscan be tested and adapted for a tracking system. Another major challenge for a tracking system is that theobject to be tracked changes its appearance during the tracking process. For example, if the UAS wants to tracka bicycle rider on the street, then it needs to learn that different light conditions make the helmet look differentand the the angle for the viewpoint might change on a hill. The centre of our system is therefore a learningsystem that can adapt online to changing conditions.