Canberra UAV Progresses with Robot Vision Test Flight

Canberra's own autonomous aerial robot team, Canberra UAV has now progressed to adding robot vision to their UAV which will be entered into the UAV Outback Challenge. Andrew Tridgell, inventor of Rsync and core contributor of the Samba project is leading up the artificial intelligence programming for the entry. As reported at a Make Hack Void Maker Meetup, Tridgell has made some progress with their machine vision camera attempting to locate an IR beacon on the ground while flying. His robot vision program is using OvenCV to detect "Outback Joe", a Canberra UAV team member laying on the ground pointing the IR beacon upwards. Tridgell explains their first attempts were about 50% accurate in locating Outback Joe-- the blurry, badly focused camera along with early algorithms (a total of 40 lines of Python) are at fault for the false positives and positive negatives. Following are some images of the team's robot vision test flight.

A successful location of "Outback Joe". Tridgell explains that the resulting "spot" of the IR beacon was only 1 pixel-- a cause for concern until they can upgrade their camera to one with higher resolution.

An image of the edge detection algorithm.