Is it possible to measure window openings with Ps?

[DRONE@WORK]

Hello all,
I am requested by a construction firm to do some research about measuring window openings in a new constructed building. The 27m heigh building is constructed with large prefabricated panels. Normally each window opening is measured up manually, but they want to make this more efficient and faster.
I'm flying with an Aibot X6, so my kopter is for sure capable of performing a job like this.
Is there someone who can give me some advice about the expected accuracy? It should be about 2mm, is this possible with a horizontal mapping?

Thank you.

[SAV]

Interesting application 

Theoretically it is possible.

Your accuracy will depend on:
- the accuracy of your ground control points (GCP)
- the distance to the object
- the lens you are using (focal length)
- sensor size and resolution of your camera

Following setup would give you a resolution of 2mm/pixel.
- Sony Nex 7 (24MP) with 16mm lens
- distance to object / wall / window / building: 8m (or less)
- ground control points accuracy of 2mm (or better)
- flight line spacing 4.7m (60% side overlap, as recommended)

Because it's a building, I guess it would be best to map "vertically", which means you fly in an up/down pattern (see attached image) and have your camera facing perpendicular towards the wall (to get the best result).

Some remarks:
- Total station / or laser range finder necessary to measure ground control points accurately
- If the building has a complex shape, it gets trickier
- Flying close to a building (to get high resolution images) can be difficult / dangerous (wind!!). Workaround: You could use a different camera (more MP) or a different lens and map the building from further away.
- You would also need some good hardware to be able to process the images at best quality in a feasible way (e.g. i7 4930K, 2 x GTX 770 graphics, 64GB RAM). Workaround: Rent the hardware/workstation (see http://www.agisoft.ru/forum/index.php?topic=918.0) 
- Fly slow and use a gimbal (but I guess you have one already on your Aibot) to keep the motion blur to a minimum. Better images, better 3d model. With above settings you would get about 1.4px motion blur (which is acceptable) at 1m/s mapping speed.

I hope that helps.

Keep me informed about your progress ;-)

Cheers
SAV

[James]

Good news is that if you are only interested in measuring window openings to within 2mm, then your whole building will have a greater tolerance for accuracy.

I.e. if window openings are 2m across, and the building is 20m across, then 2mm tolerance across the 2m openings means you can get away with up to 20mm across the width of the building.

When we scan large structures (using dslr and boom lift normally) we always seem to get within 5-10mm across the whole facade, so your application should be achievable too. We generally use terrestrial laser scan data for our control. If you had access to that then i guess you wouldn't need photoscan or a UAV.

We only did one project so far using UAV (hired in a pilot), and that was using Sony NEX 7 and I agree with SAV it's an ideal camera. Wind was indeed a problem (we were in central London) and I spent a lot of time filtering out bad photos.

[JMR]

I would not be very optimistic unless the surface texture showed the best imaginable texture for feature extraction, and that sounds not so likely to happen when it comes to new building materials.
On the other hand, measure ground control points better than 2mm is not just piece of cake. I would say sfm based photogrammetric programs like Photoscan are not the best choice for such a survey unless you had many things under tight control. It would be hard to achieve the that accuracy even using scissor platform instaid of a copter. The key points to be measured to determine the openings, should be marked and pined one by one.
Try first your setup at the ground level.
Good luck

GEOBIT

[David Cockey]

Is the visual texture suitable for PhotoScan. Try a set of photos of several ground level windows and see if PhotoScan will process them with satisfactory results.

It sounds like the required level of accuracy is 2mm for each individual window opening, not an overall accuracy of 2mm. That can be a major difference.

[DRONE@WORK]

Thanks all for your reply.

The structure will be build in architectural concrete panels, but as you can see in the attachment there will be many of windows (it will be a prison). Indeed the requested accuracy is for each individual window. I will start testing soon and post the results here.

[stihl]

*snip*

I hope that helps.

Keep me informed about your progress ;-)

Cheers
SAV

Hi SAV,

Interesting and informative post surely. Though I do feel I must add something regarding your attached image. We've surveyed a building facade for water damage before and the illustration that you're showing will not produce the overlap you're searching for. The center of the facade will be generated just fine, but as you're only flying one line near the edges you'll end up with a lot of missing geometry at the sides.
Therefore in my experience with this I suggest that if anyone were to do this, make sure to do pay extra attention to the edges of the facade as you can't generate geometry of something if you've only observed it from one angle.

As for the rest, good post.