Integrating YOLOv8m and ResNet50 for Persistent Stray Dog Detection and Re-Identification

Abstract

The uncontrolled growth of stray dog populations in urban areas presents challenges to public safety, sanitation, and animal welfare. This study introduces a scalable AI-powered computer vision pipeline that detects and re-identifies stray dogs using a hybrid model architecture: YOLOv8m for object detection and ResNet50 for identity-preserving feature extraction. Our system assigns a persistent UUID to each dog using cosine similarity and maintains structured logs for cross-session tracking. The fine tuned YOLOv8m model achieved a validation [email protected] of 90%, while our re-identification pipeline reached over 70% accuracy across repeated sightings. Asigned for real-time performance and ease of deployment, the system supports live webcam input, video uploads, and public web interface access, making it suitable for NGOs, municipalities, and smart-city initiatives.