Spoiler alert: of course they can! The camera technology, software and hardware required to collect the data and process a 3D model is either cheap or free and there are some amazing examples of models that have been created by enthusiasts in marine habitats and underwater archaeology. That said, the techniques to do this are often learned by following instructions from other scenarios that may not be ideal for the scenario our citizen scientist is attempting to capture data for.
As well as working on methods for large area imaging using multiple cameras to scale up photogrammetry surveys, I’ve also been looking at methods for scaling up surveying by increasing the survey effort with more humans i.e., citizen scientists. In collaboration with Queen’s University Belfast, we’re investigating methods to create 3D models of some challenging habitats in the temperate rocky reefs around Northern Ireland, focusing on Rathlin Island and Strangford Lough. The abundance of kelp causes a huge issue for photogrammetry, not only entangling the diver capturing the data but also occluding parts of the image and creating areas without data. Low visibility and lack of light at depth is also problematic, but for many years local divers have been producing stunning underwater imagery of the area so why not 3D models as well?
The application of orthomosiac and 3D model data opens up the possibility for some very interesting monitoring and research projects. For example, understanding cup coral abundance and health in relation to local rugosity and site topography can be explored using data capture of small quadrats, ideal for citizen scientists to quickly gather on a site when they find a suitable area and to contribute to a collaborative dataset. A similar approach is used by Seasearch for biodiversity (in which divers submit observations of marine species they have encountered on a dive).
Below are some examples of 3D models created by recreational divers using camera setups that divers would be familiar with (either dSLR with strobes, SLR with video lights or GoPro with video lights). The results highlight the uses of photogrammetry for different scales of surveying, from fine detail of cup corals through to large scale habitat surveys.
These 2 models by Ross and Dan show what can be achieved with a dSLR and strobes to capture image footage from different viewpoints around a subject. By circling the camera around and over the subject for 30-50 images a very detailed model can be created – just don’t pick a subject that moves between the photos.
This model was generated from video footage from Bernard, taken on a high-end dSLR with powerful video lights. Despite it being video stills, the resolution of the model is excellent due to the high intensity of light and excellent quality of imagery. The camera was mostly moving in a circle over the subject so some of the height detail is lost. Also note the hydroids on the edge of the model have not reconstructed very well, partly due to the lack of data but also because they are slowly moving.
The 2 models above are good examples of what could be used to support citizen science monitoring schemes such as Seasearch. Ross has taken approx. 50 images (manual using 2 strobes) in a grid pattern on a small patch of reef (approx. 50cm square). The close-up imagery allows us to see the pores of sponges and the tentacles of the cup corals. The colour is accurate due to the strobes giving us potential for identifying different morphologies of species and to detect any disease or damage that may be present. It would also give the record verifier the ability to explore the reef in detail and from multiple angles. Although it is a bit more effort, it shows the enormous potential of citizen science photogrammetry for supporting habitat monitoring.
The 2 models above were created using a stereo camera setup: 2 GoPros approx. 0.4m apart on a pole with 2 video lights attached on the pole even further apart. The fixed distance of the cameras allows the model to be scaled in the photogrammetry software. The footage was taken as images (jpg+RAW) on 5 second timelapse, which is the fastest Gopro Hero 10 can shoot in RAW. This meant going slowly along the transect. The cameras do a good job of picking up the larger features but you lose the detail at 1cm scale. Complex structures in sedimented and dark water also create a lot of noise and backscatter that needs filtering out. The transect approach works well for surveying known areas without additional fixed markers, i.e., the same path could be taken with a similar rig in the future to compare the changes.
The final model is an approx. 30m vertical transect at 20m along the North Wall of Rathlin Island. The footage was taken with a stereo Gopro Hero 10 rig with 2 video lights mounted on the outside. 296 images were taken on a 0.5 second timelapse. The cameras were set to a manual, fixed white balance and corrected in post-processing. This has highlighted the difference in lighting of the footage, with a cream-coloured bar down the middle of the model caused by the uneven distribution of light across the camera lens. The cameras FOV was linear and this would have been worse using a wide FOV. Even so, the model shows how a difficult environment can be surveyed with simple equipment used by recreational divers.
Hopefully these models have inspired some thoughts about how citizen scientists could help support data for scientific projects. If you’d like to discuss some more then get in contact.