Introduction

Last week, we looked at capturing 2d image data for photogrammetry. We took lots of photos of objects, and in this week’s lesson, we worked to produce some 3D graphics from the 2D images. This process of creating 3D models from image data is carried out mostly by software like Reality Capture, which is the software used in this lesson.

Reconstruction

Whereas some software is paid for the entire package, Reality Capture is pay per input – which means that you pay to keep each individual 3D model that you create using the software. The cost of exporting one model is around 1-2 dollars, depending on how many images were used to create the model.

To start, I was given a set of images of a gingerbread house that I would use for my first try at photogrammetry. I imported them into reality capture and pressed the ‘Draft’ button under the ‘Registration’ menu. This is a quicker way of aligning all of the images.

This gave me a point-cloud based on all the images which could be used to create a model.

Then, under the Mesh Model tab, I selected the ‘preview’ option in the ‘Create Model’ menu. This creates a basic model out of the images – selecting normal detail or high detail would create a more detailed model, but would take more time to produce.

In the same mesh model tab, under the ‘Mesh Color & Texture’ menu I selected the Texture option, which would use all of the image data I provided to create textures for the model.

That is a basic walkthrough of the photogrammetry process using Reality Capture.

Trying Meshroom

The next task was to use the images that we took last week to create 3D models using the same process. I tried using Reality Capture, but it generated a very un-defined point cloud with the images I gave it and I didn’t think that it would generate a clear model.

So instead, I tried using Meshroom, which is a more straightforward process. In Meshroom, you import all of the images and then it runs them through a flow chart-like system. The green line in the chart represents portions of the process that are finished, the amber is portions that are being processed, and red is parts of the process that hit an error. Blue is the steps that haven’t been started yet.

As you can see, when I used the same data set to create a 3D model in Meshroom, it gave me a defined and clear point cloud.

Conclusion

For some reason however, the software found several errors towards the end of processing my image data and I couldn’t get a model out of it. This could be because some of the images are blurry, or maybe inconsistent in some way. Next time, I would make sure to capture more images of the object so that there was more reference material for the software.