How do we evaluate numerically the accuracy of a surface reconstruction algorithm in MeshLab?

Question

I already performed several surface reconstructions using data obtained from 3D scanners in MeshLab obtaining great results (visual evaluation).

However, my main question is if there is any way to know the accuracy of the surface reconstruction algorithm (I use Poisson) other than visual evaluation?

In the article https://doi.org/10.1145/2487228.2487237, the authors used the point-to-reconstruction error defined in Berger et al. (2011). Does MeshLab have available a similar mean of evaluating surface reconstruction algorithms?

Answer 1

I found later the answer to my question. In MeshLab, it is possible to compute the Haussdorf distance between 2 meshes:

• http://meshlabstuff.blogspot.com/2010/01/measuring-difference-between-two-meshes.html
• Comparing Models, somebody can give me an exmple?
• http://meshlabstuff.blogspot.com/2010/01/measuring-difference-between-two-meshes.html

(sources)

So, the steps to do that in MeshLab are the following ones:

1. Filters > Sampling > Haussdorff Distance
   • Sampled Mesh: the model generated by a surface reconstruction algorithm
   • Target Mesh: original point cloud
2. Render > Show Vertex Quality Histogram
   • This allows you to evaluate the quality histogram (the lower dispersion and values, the better) between different generated models
   • Use the maximum value for the coloration of the datasets
3. Filters > Colorize by vertex quality
   • note that the minimum and maximum values should be the same for all models (for a direct comparison between two models)!