Roadkill accidents pose a serious threat to both wildlife and human safety, particularly in nighttime or adverse weather conditions where visual degradation significantly limits the effectiveness of conventional surveillance systems. Most existing object detection models assume high-quality input images and suffer substantial performance degradation when applied to low-resolution road monitoring environments. This study presents a deployment-oriented roadkill detection framework that enhances visual information prior to detection by integrating super-resolution–based image restoration, feature-enhancing preprocessing, and lightweight real-time object detection. The proposed system adopts an ESRGAN-based super-resolution module to recover fine-grained visual details from low-resolution inputs, followed by channel-expansion preprocessing to enrich structural cues. A lightweight YOLOv5-nano detector is then employed to achieve real-time animal detection under resource-constrained conditions. Comprehensive experiments conducted under diverse environmental scenarios demonstrate that enhancing visual information before detection improves robustness for small and distant animals without increasing model complexity. The results further highlight the effectiveness of combining super-resolution preprocessing with lightweight detection models, achieving a favorable balance between detection accuracy and real-time efficiency. In addition, a comparative analysis between modular pipeline designs and an end-to-end integrated variant provides insights into system-level trade-offs between flexibility, efficiency, and feature coupling. Overall, this work offers practical design insights for developing reliable and deployable wildlife monitoring systems and contributes toward mitigating roadkill accidents in real-world road surveillance environments.