Calculating volume from a photo

Hi! I want to ask is there a way to calculate the volume of wine inside the wine glass like in this photo? Or do I need more photos of this wine glass from different perspectives?

Answer 1

[Image Analyst]

We can't see your photo. But if you assume the glass is round, and if you've calibrated spatially, then you should be able to calculate approximate volume. You may need to throw in a small fudge factor to account for the refraction near the edges. To do that you could fill the glass with known volumes, then measure the area in pixels and calculate the volume. Prepare a function to transform the pixel volume into an actual volume in milliliters.

[kkk]

Thanks for your quick reply!

Actually I am wondering if I could make it into a smartphone application. How could I make use of the phone camera? Any suggestion is appreciated. The photo is like this

[Image Analyst]

I don't know how to make a smartphone app with MATLAB. There is no way to get the volume of that glass unless you have something of known size in the image. I don't know the size of the glass - there could be one ounce in there or 100 liters. That's why you need something of known size to calibrate against, like in my attached spatial calibration demo.

[José-Luis]

I'm curious. How would something of known size help if you don't know the depth of field? Wouldn't you need both its size and position relative to the glass and camera? It is relatively easy to make perspectives lie.

[Image Analyst]

If you put a ruler at the same plane as the glass, and assume the glass is round, and assume the vanishing point issue is minimal over the range of camera distances we're dealing with (from front to back of the glass), then you can get the width at each row and convert it to a volume of a disc with V = (lineSpacing)*pi*Diameter^2/4. Then do that for every line in the wine to get a rough volume. Of course if you have an angle view so that you're seeing the top surface of the glass, it gets a lot more complicated - not so straightforward as if you have a perfectly perpendicular viewpoint. That's why I said to calibrate by getting a transfer function of true volume vs. pixel area.