An attempt to count visitors at the Ireland's famous Giant’s Causeway using AI has failed, after the software struggled to distinguish between people and the site’s distinctive hexagonal rock columns.
The AI system was deployed to analyse commercially available drone footage of crowds at the causeway, located near Bushmills in County Antrim. However, the unique geological formations share visual similarities with human figures when viewed from above, particularly from a top-down drone perspective, which led to significant confusion within the model.
Researchers believe this issue stems from limitations in the data used to train the AI. They suggest the model “may not include enough examples of the Giant’s Causeway or similar environments”, causing it to generalise incorrectly when interpreting the footage.
The study was commissioned as part of a broader initiative by the UK government’s Department for Culture, Media and Sport, which is exploring how digital technologies might be used to estimate visitor numbers at non-ticketed attractions.
A team from the University of Glasgow carried out the research using YOLO-Crowd, an open-source AI model designed for crowd counting and face detection. Despite its intended purpose, the system proved unreliable in this setting.
The researchers concluded that the object detection model performed poorly when applied to footage from the Giant’s Causeway, ultimately failing to produce a realistic estimate of visitor numbers.
Read the full story here.