T3 Season 2, The Model


Its T3 time again but first I have started tagging people that have entered T3 in the past on their member icons its a very small recognition for their valuable contributions which have driven APM forward faster than it might have without the competition. They have not only put their equipment on the line discovering new ways of doing things but countless hours as well.

Unsung heroes in the development of low cost autopilots for all.

So lets get into this round.

I waste far too much time on the most excellent GEFS there could be more 3D buildings laid on top of the Google Earth base layer. Lets change that.

The last T3 challenge "The Cube" bought some really outstanding results and the natural progression from this is automated flight around a building or natural point of interest and the creation of a 3D model perhaps in 123D catch of the subject. Last November Alan Sanchez showed us how. Improvements in the multirotor code that came into play at the Sparkfun AVC will help with the flying.

This could be achieved with a fixed wing platform with a little thought.

We should have some ground rules (pun intended) you must seek permission to make a model of the building, or natural wonder! You should fly no higher than 130m and the flight should be hands off all the way and you should stay visual with the airframe at all times. Please do not create a model of something in class A airspace unless it's at a recognized model aircraft field.

The flight must be at very least autonomous from just after take off to just before landing. If you can do it all autonomously bonus points! FPV entries will not be permitted.

If you submit your model to Google Earth and get it placed bonus points when it appears on GEFS for me to fly around it. In this case the model might be skinned with images taken from the air.

This is no simple task so it should remain open for six months. Normally I tack time on at the end so to break with tradition lets start the competition on the 1st of August and run all the way to 1st Feb 2014! That gives folks the extension weeks in advance!

Prizes to be announced when I have finished twisting CA's arm. As we know he always comes good.
I am also putting out a call for volunteer judges from the ranks of previous T3 entrants to help me decide. This one is I think going to be very subjective and how attempts are documented is going to be a big factor. 

There is no doubt automated building inspection will be a part of the future for sUAS.

Good Luck, be safe


Now let the traditional but did you mean XYZ questions begin! Oh and if you are tooling around on GEFS as well look out for me as Gary sUAS News, often slope soaring ;-)

E-mail me when people leave their comments –

You need to be a member of diydrones to add comments!

Join diydrones


  • If you have a large project, you can separate the pics by areas in a few separate folders. Then, from area/folder 1, select some pics neighboring the area/folder 2 and copy them into folder 2. Do the same for each folder. Now you have each area/folder having some overlap to adjoining areas.
    You then process one chunk per area/folder in a single project. When you're finished, you can correlate all chunks together using common cameras instead of ground points. Finally you may also merge all chunks together.
    With such process, you reduce processing time drastically compared to processing a large pic datasets which could finally fail due to memory limitation.
    I'm no photoscan specialist, having just processed a few projects until now, and working on a simple 16Gb Core I7 proc, but I was successful with this method in my last larger project.

  • T3

    Stephen, even without having any experience with Photoscan at all, I was wondering if 5000+ images will be possible to process in one run. On how many flights is your data based on? You could try to process each single flight separately and then merge them into one big file. This is no problem if you have a good georeference. Actually, that's what I wanted to do but the weather conditions did not allow any further flights so far. 

  • T3

    maybe you will get a good result in medium quality, your point cloud must be very dense

  • T3

    After 9 days of processing, my true 3D job finished but Photoscan simply said "insufficient memory" and couldn't display it.  Oh, well, starting over on medium quality instead of high quality.  And I still have the height map, I was just hoping for a true 3D.

  • T3

    Nice work Jose, and nice place

  • José, that is really cool. I am impressed (and jealous that you live somewhere with castles).

  • Let me show you one of my recent work (well, not just mine but of my team at university): this is out of competition because the model has came out of two flights. One was completely autonomous for taking vertical photos and the second, that disqualifies for the contest was done in manual mode (not FPV) for taking oblique shots. Each of flight took about 35 mins and the drone itself is also kinda out of competition not having APM inside. You will seer laser scanning at work but all data shown in the video came out from photogrmmatric processing by means of Photoscan. I hope you enjoy this non-elligible entry.

  • T3

    José asked me for a point cloud.

    This is a screenshot of the point could before dense matching:


    This is a screenshot of the clipped point cloud after dense matching (however, not the highest resolution possible):


    You can download this file here (22MB *.zip).

  • Very nice work James! My hometown looks awesome from above!

  • T3

    Sylvain, thanks for testing! 

    Meanwhile I tested it with IE11 and Safari. It works with both but you have to enable it first in Safari.

This reply was deleted.