Animal Detection Using Footprint

Abstract

Wildlife conservation increasingly relies on noninvasive monitoring techniques to track and identify animal species efficiently. Traditional methods, such as physical tagging and direct observation, are labor-intensive, costly, and challenging in remote or environmentally sensitive areas. To overcome these limitations, this paper presents an advanced footprint-based animal classification system leveraging YOLOv8, CSPDarkNet for feature extraction, and C2f-based feature refinement. By processing footprint images from diverse sources, including wildlife cameras, mobile captures, and field recordings, the system ensures high classification accuracy across varying terrains. CSPDarkNet enhances feature extraction by capturing essential footprint attributes such as texture, edge contours, and species-specific morphological details, while the C2f module refines these features, improving adaptability to challenging conditions like muddy, sandy, and uneven surfaces. Extensive experimentation on a dataset of over 10,000 labeled footprint images confirms the system’s effectiveness, achieving a classification accuracy of 98% and outperforming traditional tracking techniques. The
proposed model offers a scalable, automated solution for wildlife monitoring, ecological research, and biodiversity conservation while also enhancing public safety by enabling early detection of potentially dangerous wildlife in residential or trekking areas.