Mitigating Roadkill Using Computer Vision and Machine Learning Techniques: A Case Study on Turtles in Australia

Roadkill, which is caused by Animal-Vehicle Collision (AVC), is a prominent issue. However, a panacea is still at the experimental level. AVCs pose significant threats to biodiversity, human safety, and the economy. Hence, mitigating roadkill is in the best interest of both animals and humans. Current strategies to identify roadkill hotspots are often biased toward larger animals, rely on citizen science, and focus mainly on animal remains, leading to an incomplete understanding of the risks to smaller species like turtles. To address these limitations, this study uses advanced computer vision techniques, specifically YOLO models, to detect turtles on roadsides in real-time, achieving a mean Average Precision (mAP) of over $90 \%$. Using data augmentation and robust validation, our results demonstrate the potential for scalable, improved ecological surveys and automated roadkill mitigation systems. These findings support the United Nations’ Sustainable Development Goals for biodiversity conservation and road safety.