Search code examples
3dscanning3d-reconstructionphotogrammetry

create 3D Object from found images from various sources


I'd like to use a program to at least partially digitally reconstruct a 10-foot-tall human statue using 15 or so images taken from different sources gathered from google images. The idea is to use these images as "scans" in order to digitally reconstruct an object. Then, the model can be assessed and refined (filling in missing spots, etc) from there.

Do you think this is possible, and if so, how successful do you think using these images from the web as scans would be? The object no longer exists, so these images are all I have. Finally, what software would be best to carry out this process?

Again, the images would be sourced from google images, and would consist of a variety of images of the object taken at different angles from different people using different cameras. It is important for me to use a program to at least partially generate an image rather than create it by hand from scratch.


Solution

  • The best approach would be a Structure-from-Motion (SfM) approach here. Unlike many traditional photogrammetry softwares SfM software can handle multiple un-calibrated cameras and has a reasonable degree of robustness to changes in light, shadows etc. The main criteria for the reconstruction is that each point to be reconstructed is seen by more than 1 camera as this allows for triangulation.

    In terms of software packages a good package would be Agisoft PhotoScan. This is reasonably cheap and is very good at always finding a solution even if the data is pretty poor. Other solutions would be Pix-4D and Capturing Reality.

    If you were looking for free open-source software you could try; OpenSfM, VisualSfM or ColMap to name just a few. If you are not familiar with SfM I would recommended using a free trial (or even purchase) one of the commercial products.

    Answering your questions:

    Do you think its possible? - Yes, absolutely so long as you can recognise the same features in multiple images.

    How successful will it be? - Completely depends on the quality of you images. If they are all similar quality and overlap well it should work quite nicely, however, if they are not you could have problems. Only one way to find out.