Scanning a fence in Melbourne

This is a fence at the side of an apartment block in East Melbourne, painted with a mural of blue deer and these ethereal yellow tentacles.

The model was created using 80 photos shot on a Nikon D5100 with the stock 18-55mm lens, processed with Agisoft Metashape and finished with Blender. The final mesh was decimated from some 1.3 million faces to about 100,000.

The scan is, in a way, the result of a failed experiment. The idea was to capture the fence and create a fully realised mesh and texture, while retaining the building and foliage behind it as a point cloud, all in the same model.

There is often a lot of peripheral data in a photoscan capture that, as a point cloud, looks quite visually pleasing as the density of points decreases as they get further away from the subject. Seen below in a previous project on the White Cliffs of Dover

Those points are also indicative of a bigger scene while clearly being seperate from the main focus. When converted to a mesh however, sparse point clouds generally look ugly. The idea was to make better use of the data surrounding the subject of the scan to provide visual context to the final model in a way that looked good without taking up too many resources.

There was no straightforward and elegant solution to mix a textured 3D mesh and a coloured point cloud together in a rendered model. A few different methods were tested that included a couple of addons for Blender and some work with CloudCompare but none produced the results in simple enough way that would lead to an exportable model.

It would still be nice to have a way to do this, so get in touch on Twitter if you know a suitable way.

Either way, less can sometimes be more. I think the final, simple model came out looking pretty nice and rendered well with HDR lighting.

Model viewable below: