Introduction

As a outcome of ADEX/OzTek 2025 in Sydney (Australia), the Wombats were involved in a excursion to South Sulawesi (Indonesia). Robin Cuesta (Sulawesi Dive Trek) has been working with local partners to develop cave diving in the area. Josh (Grubby) embarked on an adventure with Robin and his team for a little bit more than a week in the area. Apart from the long airport overlays, the area is definitely worth the visit and has many unexplored or barely explored leads that we believe might take us to a major dicovery right around the corner. Robin has been working closely with local authorities and government bodies in order to improve and regulate access and conditions of use of the undergroung water table.

Line Survey and Photogrammetry

Lapahia is usually the first dive site Robin takes his guests to when travelling with him. An explorer himself, Robin has run MNemo in several caves in the area, including LaPahia. The Wombats did not run new MNEmo surveys as the info would not add to the information already available in the cave.

As the invitation and decision to engage was an impromptu one, the only tool taken in the trip was the Insta360 camera and 2 BigBlue video lights. Although a light packing set-up, this was enough to generate enough data for several caves in the region. Due to the proximity to the accommodation location, easy of gathering data and relatively small size of the underwater cave, Gua LaPahia was selected as the first one to have it’s processing done via Agisoft Metashape.

This was the first model we worked on based 100% on Insta360 images. Josh set the rig to take timelapse photos, which we extract one every second to create the database for the point cloud. Point cloud completed and checked, it is time to filter and clean the scan. Often times there are points that way of the mark. This happens in every model we worked on so far, and it is easy to spot them as the majority of the points are going in the same direction but these are wide spread and off the mark. This will reduce the number of points the software has to manage to produce the model optimising processing time and generating a cleaner model to work with.

The last part of the photogrammetry workflow is to generate the textured model. This will be based on the photos again and here is where we notice the biggest difference between imagery from GoPros (Wombats std for photogrammetry) and Insta360. GoPro data gives us a way higher resolution in a very large processing time, while Insta360 gives us quicker results at lower resolution. It is not an issue with how we set the GoPros vs how we set the Insta360. It is basic maths (and rough numbers): we use 4 GoPros shooting 1 photo/sec each. GoPro photos resolution is 25MPixels. A single photo generated via timelapse in the Insta360 will give us a 12MPixels photo that is 3 times the size of the GoPro picture. Of course, this depends on which GoPro and Insta360 cameras you are using, and if we had 1 vs 1, the result would be most likely better via Insta360.

Insta360 photos are a perfectly good source for a mapping project. They will give you an quicker result and most times the level of detail is enough for a map. And so we followed this new ‘Wombat std’ for photogrammetry. General mapping tasks will be developed with Insta360 imagery. Detailed sections will be supplemented by GoPros imagery.

For surface and dry cave, Insta360 hasn’t provide easy results to stich with underwater scans. Again, the lack of definition doesn’t help the software to understand that points are the same above and under water. Drone footage (DJI Mini) and GoPros have given us the best results. The geo-referenced photos from the drone help us to drop the cave in the real world position without any manual adjusts. From there, GoPro scans can be tied nicely to it and we keep developing the model. Each tool for a job. If you want to know more about photogrammetry and the Wombats’ workflow, have a look at the blog entry: Photogrammetry, a conversation.

Drone imagery for this project was gathered by Robin Cuesta. Another exemple on how miles-away colaboration can work and produce great team work.

Models and Maps