Cheap onboard tracking system tested on UAV

More and more uses for drones and UAVs seem to be cropping up, but there are limitations. The Department of Defense in coordination with the University of North Dakota has developed a UAV that is capable of recognising specific objects on the ground and tracking them accordingly. Traditionally this was a hard exercise, requiring large computation powers, but the duo have come up with an imaging processing machine that's light enough to be installed on the UAV itself.

The device is capable of tracking specific objects such as cars or houses without any additional processing needing to be performed on the ground. The team tried to boil down the problem to the bare essentials, recognising that cars and houses don't change shapes, but do change their orientation relative to the camera. The processing machine predicts the path of the object based on previous frames if it's in motion, but it also creates a new set of images rotated by 10 degrees when it first locks on to the target.

A total of 36 pictures are generated, which leaves the image processor to try and match the current target against the estimations. The team uses a simple Linux computer on a circuit board with a camera and gimballing system, testing it on a UAV with a wingspan of 350 cm and carrying capacity of 11kg. In testing, the team says the system worked well, with targets being locked on to using an air-to-ground video link, sent back at around 25 frames per second at an altitude of 200 feet.

[via Technology Review]