Projected Light Patterns

Using Projected Light Patterns can drastically improve the quality of the reconstruction (especially for featureless or single coloured surfaces).

(The following examples are based on 19 images of the "dog" dataset.)

Use projectors or other light sources to produce patterns on your object of interest.

Capture two images from each perspective. One with the pattern, one without pattern for texturing. Use a tripod or camera rig, as both images need to be in the same position. Do not change / move the light pattern or the object of interest while capturing.

Place the two image sets in different folders and rename the image files so corresponding images have the same file name.

For testing, the sample datasets by pi3dscan or the dataset "American Rodeo Cowboy" provided by Meshroom user hargrovecompany can be used.

Import the images with the projected light patterns in Meshroom. Duplicate the PrepareDenseScene node and connect the output to the Texturing ImageFolder input. Set the image dataset without the light patterns in the new PrepareDenseScene node (Images Folders (+) ).

Now you can start the reconstruction. (The images with patterns will be used for the reconstruction. For texturing the images set in the PrepareDenseScene node (Images Folders) will be used.)

Note: The same workflow could also be used when taking polarised and unpolarized photos. See 763#issuecomment-583904068 for a detailed comparison.

Using polarized photos, with unpolarized photos used for texturing with a new PrepareDenseScene node:

Great advice on the topic can be found here: https://www.reddit.com/r/photogrammetry/comments/n9jlf5/questions_on_using_polarizing_filters/