Mixing Datasets – Photogrammetry and Multibeam

The one nagging thought while scanning is “Did I get enough overlap?”. No overlap of points and the scan data just won’t build into a single, seamless model. For anyone scanning topside data the opportunity to reshoot the missing images can be straightforward, but with shipwrecks its not so easy. The distances involved, boat availability plus cost of gas all add up. Chuck into the mix some bad weather and loss of underwater visibility and it can take weeks, if not months to reshoot.

Back in November Dee3D paid a visit to Scapa Flow. One of the sites scanned was the salvage barge YC-21 and part of the German destroyer F-2. It became apparent that the two wrecks had a gap. First up, the barge and stern of the F-2:-

The results were pleasing, but when you set out to scan something and it doesn’t work the disappointment is there. Here’s the bit that was missing, the bow of the F-2:-

There was a gap between the two sections of the F-2. No amount of tweaking the settings was going to invent data and a dead end was reached.

But then an offer of a multi beam point cloud of the site appeared on Facebook. Could the data be merged? Well it took a while, but the answer was yes.

Step by step.

First step was to trim the point cloud. The scanned area was massive and all we were interested in was the barge and F-2. Once trimmed (using CloudCompare), we ended up with a working point cloud.

The point cloud was great, but are quite low in detail when compared to photogrammetry models. To make life easier, the point cloud was converted into a mesh using MeshLab. The result was blocky, but gave us more than enough detail to work with.

Next, we exported the barge and F-2 models from Agisoft Photoscan and imported both into Meshlab. All that was left to do was align the two sections of wreck with the point cloud mesh.

Finally, for ease of display on Sketchfab the models were merged (again in Meshlab) to produce a single model:-

This process sounds quite simple, but a lot of head scratching and reiterations were tried before it worked. Given more time the quality of the final model can be improved and the textures added, but it was time to put this one down and look at another job. More news on that shortly.