We ran into multiple issues, both with taking the photos and with modeling them, that we would like to highlight in case others hope to do a similar project.
Photos
As the first part of our process, taking photos proved to be more challenging than anticipated. One issue was that we struggled to find a time in which no one was using the pool, as we frequently went to the pool only to find out that there was a diving practice or people swimming in the pool. We also hoped to take pictures on the roof of the offices in the pool area, and after some deliberation the janitor was able to let us on to the roof; however, likely due to the big change in angle compared to the other photos, we couldn’t end up using the pictures from the roof as they did not produce a clean model. We had to take multiple rounds of photos, as we didn’t use proper photogrammetry practices for the first couple of rounds, producing insufficient models. Finally, for the final round of photos we ended up using, we took so many pictures that the camera started to lag, which caused some of the pictures to be slightly blurry, hindering how well of a model we were able to produce as well.
Modeling
Less surprisingly, we also experienced some challenges with creating our model. Before even adding photos to a model, we realized that the software we intended to use, Matterport, now required a subscription to embed a model onto a website. This was a tough challenge because Matterport allowed for tours through a 3D space, which is exactly what we were hoping to do, so we needed to find alternative software to use. We tried using both RealityCapture and Metashape, and with greater features and tools, Metashape ended up producing better models for us to use.
One main issue was that with the amount of photos we took, we would need a computer with a more robust GPU to process the pictures and align each of them. We were able to get our hands on Apple iMacs in Carleton’s CMC, and even with those the model took over a day to create. However, besides these issues, the modeling stage went alright for the pictures we had; there was some photo alignment issues that caused our model to not look as clean as we were hoping (such as not being able to use roof pictures, poor photogrammetry practices at first, and blurry photos), but for the most part the model turned out alright. We then had to manually clean up floating artifacts from the model, which was a manual and timely process, but this produced a much cleaner model for us in the end.
