The main flying event of the year was the dish farm flight. The dish farm is being torn down in 2011 & replaced by flea startups.

Don't think it emits radio waves anymore & VicaCopter is invisible when she's stationary. Nevertheless, got serious radio dropouts after a certain amount of time. Seems the XBee wire antennas are extremely directional, end-on antenna pointing even at 300ft kills the signal & the only way to consistently get over 100ft is diversity.


After using Cinelerra's 10 year old motion tracking algorithm satisfactorily for 10 years, it was time for an upgrade better suited to the low framerate, wildly rolling aerial photos we now encounter most of the time & the need to penetrate haze. Introducing the Marcy algorithm. There is no Marcy algorithm in the literature. We made it up.

The old motion tracker swept a macroblock around to test all possible translations & took the 1 with the highest match. Then it rotated the macroblock in the translated position to find the rotation with the highest match. It was a recursive problem, requiring accurate translation to detect accurate rotation & accurate rotation to detect accurate translation. Too much rolling & the translation accuracy deteriorated. You could improve the accuracy with multiple passes to a limit.

Also, we had the translation of multiple macroblocks giving rotation & scaling. That suffered the same problems.

The Marcy algorithm sweeps around a macroblock containing all the possible rotations, testing all possible rotations for each possible translation before giving the translation result. It's actually a 3D macroblock with the 3rd dimension being all the possible rotations, a "Marcy macroblock" you might say.

In exchange for a huge increase in computation, the payoff is a huge increase in accuracy for image stacking in 1 pass & deeper haze penetration than before. When is China going to get those octocores out?

This is our first use of the Marcy algorithm on haze. Stacked 170 images fully autonomous where previous algorithms needed tons of hand matching, usually missed, & weren't nearly as accurate.

This is San Francisco from the day job in extreme haze. 75 aerial photos stacked.

This mission, expertly flown by the Air Force's 50th wing (GPS), was the Marcy algorithm in previous frame/incremental block mode where the macroblock follows a point.

This was the complete flight in previous frame/same block mode where the macroblock stays in the same position.

This doesn't show anything useful but looks neat.

The Marcy algorithm is really just another step towards the ultimate goal of using databases of known polygons rotated in every possible 3D orientation. These future Marcy algorithms on steroids will give 3D volumes from single photos but require a hardware implementation. The photosynth program sort of does this by exhaustively drawing single vectors & comparing them with photos.

Views: 232

Comment by Jack Crossfire on November 13, 2009 at 9:28pm
Snagged 2 shots from the haze with 100+ aerial photo stacking.

Since the lens aberration is still unknown, we're still limited to just tracking small points. It's much less than you would see from a ground based 1600mm lens. The next step for the aerial coverage would be gyro stabilized camera gimbals & longer lenses.

This was a 1600mm lens on the ground.


You need to be a member of DIY Drones to add comments!

Join DIY Drones

© 2020   Created by Chris Anderson.   Powered by

Badges  |  Report an Issue  |  Terms of Service