About my blog

Based on novel robot vision/image recognition algorithms, it is possible to reconstruct a 3D model with a simple webcam in real-time. The AR drone, equipped with several cameras seems to be a perfect match to establish a 3D scanning system for large objects, such as cars or buildings. (Experts: Ronald Poelman&Oytun Akman)




De meningen ge-uit door medewerkers en studenten van de TU Delft en de commentaren die zijn gegeven reflecteren niet perse de mening(en) van de TU Delft. De TU Delft is dan ook niet verantwoordelijk voor de inhoud van hetgeen op de TU Delft weblogs zichtbaar is. Wel vindt de TU Delft het belangrijk - en ook waarde toevoegend - dat medewerkers en studenten op deze, door de TU Delft gefaciliteerde, omgeving hun mening kunnen geven.

Flying 3D scanner

Future recommendations

In the last 4 weeks we did some investigation on software
and the ARdrone. Because we didn’t had a lot of time to do this research we
were not able to come up with a completely working system for 3D scanning. But
we were able to create an idea how this 3d project could be improved.

We wanted to improve the software so that the software is
able to run completely by itself because now we had only proven the concept by
running it step by step. Also the software can be improved so that the flight
path of the drone can be preprogrammed. At the moment the drone could only be
controlled by hand. A second point of improvement is the implementation of an
extra software package which is able to filter the sharp pictures so that the models
which come out of the pictures/videos, will have an higher surface quality.  Also should it be nice to have a program
which is faster in converting the pictures/videos into a point cloud.  For example a PTAM based program.

The drone can be improved in several ways.  First of all we experienced that de ARdrone was
to light to fly outside. With the least amount of wind the drone became
unstable. That why we recommend to use another more heavy and stable drone. Which
is also capable of lifting heavier cameras and other equipment.  For the camera we would recommend to use a
higher quality camera, and if possible on with two lenses because than a 3d
picture is build up in just seconds.  Also we would recommend stabilizing the camera
using gyroscopes.  This will result in a
much better and faster 3d model.  Also a GPS
and a 3G receiver should be nice to have an accurate position and a larger range.
This is needed when the programming of the fight path is implemented.


These are about all the recommendations we came up with. We
hope you enjoyed our blog. And maybe it gave you some inspirations for your own


Looking back on the Science Fair

Last Tuesday, the first of November, down the basement of the Faculty of Industrial Design we showed what we achieved in the past five weeks. To do that we used a beamer and two laptops on a desk. The beamer projected a short movie about the content of our project and some recommendations. This movie will be posted on this blog asap. One of the two laptops was for running the interface we made, which is already posted on this blog. Interested people were able to test the interface. The other laptop was equipped with a 3Dscreen and 3Dgoggles, so people can see a pointcloud we made, in 3D. Ofcourse the AR.Drone was also there with the GoPro camera. The drone actually flew a few minutes during the Science Fair.

We all enjoyed the project with as highlight the Science Fair. We think that all groups did very interesting research and gave nice presentations of their findings. Our compliments to all of you!

preview interface





– Still takes a lot of time (± 1.5 hour).

– The model still shows some deviations from the real model.

– The quality of the model depends on the lighting and the
reflections on the object to scan.

 (Nuna 4)

 The light is not equal distributed, this results in a low resolution



original movie

YouTube Preview Image


YouTube Preview Image

YouTube Preview Image


Photosynth and PTAM

Repair and software research

Hi, it has been a while since our last
blog. The past few days we were very busy trying to get

the drone to fly properly. Last
Friday we went to the aula, and after a little bit of asking around, 

we had the chance to fly in one of the
college rooms.

We did this to practice flying with the drone and to make some
videos, that we can use later on to create a 3d model.


Last Monday we did some repair works
on the ARdrone. Because apparently someone had crashed

the drone before we received it from
Jouke Verlinden.

The drone was broken at the cross were the arms of the quadrocopter come

First Menzo and Roeland tried to solve the problem by gluing the
parts together, this didn’t work.

Eventually we solved the problem
using two washers and 4 nuts and bolts.

Then we tested the drone again and
we could immediately notice the ARdrone was much more stable than before.

The rest of the time we did some research on software to get a 3d model from the videos taken last Friday.


© 2011 TU Delft