How much square footage can a Photogrammetry capture?


(Carowley) #1

Hello All,

I am doing some research on forensics using photogrammetry. Is it possible for a camera to capture a room? Looking at my old renders by measurements it look as if my camera captured a 9ft x 9ft area accurately with a 18-55mm lens. I know laser scanners are used for large rooms but out of curiosity do you believe a small drone such as a Spark / Mavic could capture a crime scene accurately?


(Mesheritage) #2

Hello,

There isn’t a real limitation on that regards, you can do SFM (I guess you meant that, or general photogrammetry?) on microscopic sample or entire city (consider Google map).

Problem is always "how many pictures and how well you area is covered?"
More details your room has, bigger your room is, more pictures you might have to take. This can explode your processing time.

Accurately also depends on “what you are measuring”? You might end up having an accurate measure of the room, but not of a sharp edge of an object.
Assessment of SFM techniques are a bit tricky, often it is compared with laser scan which is not perfect either (depend on position of the scanner, the performance, the reflectivity of object etc…).

Would be best to define how accurate you need it to be. Thinking of what object must be rendered well. For example if you have very reflective object or transparant, it will be much harder to be “accurate”.

Also if you get crazy on the number of picture you might consider software like reality capture. Keep in mind that different software will produce different model and will have different “accuracy”.

The best way to go will be to test, if you have a “test chamber” with the standard case scenario. Take a smartphone to mimic the type of camera a drone might have and try out.
Use of small drone will be an issue more to navigate and take enought picture and enought angle. Not so much on the “drone camera”. I managed to get accurate measurment with bad video from low smartphone and out of focus usb microscope (very hard to get worst than that!).

So maybe gather the various standard object you might encounter and try it out. At least you can have a sense on which object work/doesnt work. At the end, you might extract visual feature instead of depth for some of them.

Good luck!