Abstract: Wildlife-vehicle collisions (WVCs) represent a significant global challenge, leading to millions of animal fatalities annually, substantial economic losses, and posing threats to biodiversity. Traditional mitigation strategies, such as fencing, wildlife crossings, and static signage, have demonstrated localized effectiveness but often face limitations in scalability, cost-effectiveness, and long-term efficacy across extensive road networks. This study investigates a novel centralized artificial intelligence (AI) detection system designed to provide real-time wildlife alerts to drivers. The proposed framework integrates thermal and RGB imaging with the YOLOv8 object detection model, consolidating video processing at a central hub rather than distributing computational resources to numerous roadside units. This architectural choice aims to reduce deployment costs and simplify maintenance, particularly in remote or challenging environments. Preliminary evaluations indicate that dual-spectrum imaging enhances detection robustness under varying environmental conditions, though challenges related to false positive rates and the acquisition of species-specific training data persist. The system incorporates an adaptive learning mechanism that continuously augments the training dataset based on verified field detections, thereby improving model generalization over time. While initial results demonstrate the potential for WVC reduction in controlled settings, real-world deployment necessitates addressing critical obstacles, including reliable power infrastructure, robust data transmission, and unpredictable driver behavioral responses. This research contributes to the evolving field of intelligent transportation systems for wildlife conservation, highlighting the complex transition from laboratory-based performance to practical, operational deployment.

Keywords: Wildlife-vehicle collisions, YOLOv8, thermal imaging, centralized AI framework, intelligent transportation systems, real-time detection, wildlife conservation, computer vision, deep learning, road ecology

Downloads: | DOI: 10.17148/IJIREEICE.2026.14386