Our drone produces hundreds of hours of footage, which may or may not, contain dolphins. In order to speed up the process of finding and counting dolphins, we trained a model to automatically do the counting for us! Louise Wilson is a researcher at the University of Auckland’s Institute of Marine Science, who had a similar problem – but in her case, she was interested in counting boats. Almost half of the Aotearoa New Zealand population (45%) goes boating in their spare time. This puts a lot of pressure on our coastlines, not only due to pressure from fishing — these boats also generate high levels of underwater sound. This sound pollution is of conservation concern because it can cause hearing loss and behavioural changes in crustaceans, fishes, and mammals.
Louise installing a time-lapse camera at one of her study sites.
To investigate how sound pollution from small boats is impacting animals living in Tīkapa Moana Te Moananui-ā-Toi Hauraki Gulf, Louise selected five sites. At each site, she recorded the underwater sound, whilst also capturing images of boats on the sea surface. This meant that she could relate activity on the sea surface to the sound heard by animals living in these habitats. However, this is not an easy task! To do this, Louise needed to count boats in the thousands of images that were captured on each camera. Doing this by hand would take months – and would be very dull. MAUI63 was able to help her automate this process, using the same approach we used to identify and count dolphins in drone footage! As can be seen in our computer vision AI post.
The boat counting model we devised achieved high accuracy and meant that the hundreds of thousands of images collected could be analysed in a matter of days! Whilst our drone footage uses aerial images to count dolphins, where the background may be simpler (2D sea surface), we showed that our model can also be used effectively when working with landscape photos, which may have more complicated backgrounds e.g., trees and clouds. We are currently in the process of sharing the output of the model so that the tagged images can be used by other researchers interested in similar questions.
If you are interested in learning more about this collaborative project, the results of our work with Louise can be explored in further detail here.
Use of our model to automatically count boats in images.
Comments