Letting AI Find Dinosaur Footprints: How Deep Learning Can Assist Surveys of Large Tracksites
Trace fossils and body fossils provide two very different kinds of information. Skeletal remains allow paleontologists to reconstruct an animal's morphology and body size, whereas footprints preserve information such as pace, stride, and track arrangement, which can be used to investigate how extinct animals moved and how they interacted with their environments. Yet when an entire bedding surface is covered with depressions, fractures, and sedimentary structures, researchers still have to determine one by one which marks are genuine dinosaur footprints.
Although photogrammetry and Light Detection and Ranging (LiDAR) can now be used to produce highly detailed three-dimensional models of fossil footprints, the initial process of finding the tracks still relies heavily on visual inspection by researchers. Poorly preserved tracks may retain only a few faint toe impressions, or their outlines may blend into the surrounding rock, leaving room for differences in interpretation among researchers. In 2026, a study from Chungnam National University in South Korea therefore attempted to delegate part of this task to an AI-based object detection system, allowing a computer to locate dinosaur footprints directly within full-scene field photographs.
The model used in the study was YOLOv8. Unlike conventional image classification, which simply assigns an image to a particular category, an object detection model must determine both whether a target is present and where it occurs within the image, ultimately marking its position with a bounding box. The researchers selected YOLOv8x, the largest version in terms of parameter count, to prioritize detection accuracy, and trained it using dinosaur footprint images collected through South Korea's AI-Hub dataset.
Want to read more?
Subscribe to thesoundofevolution.com to keep reading this exclusive post.
