Single-Camera Flood Monitoring Using YOLOv5 for Water Level and Turbidity Detection

Gamma Aditya Rahardi (Department of Electrical Engineering, University of Jember)
Sherly Apta Maharani (Department of Electrical Engineering, University of Jember)
Hasanur Mohammad Firdausi (Department of Informatics, KH Bahaudin Mudhary Madura University)
Satryo Budi Utomo (Department of Electrical Engineering, University of Jember)
Dani Hari Tunggal Prasetiyo (Department of Mechanical Engineering, University of Jember)

Jurnal Ecotipe (Electronic, Control, Telecommunication, Information, and Power Engineering)
Vol 13 No 2 (2026): Jurnal Ecotipe, October 2026

Floods are one of the most common natural disasters and require efficient early warning systems to mitigate their effects on infrastructure and communities. Conventional flood monitoring systems generally consist of many physical sensors, which may incur higher maintenance cost and reduce the system robustness in a severe river environment. The purpose of this study is to create a single-camera flood monitoring system that is capable of detecting water level and water turbidity in real-time using YOLOv5. An initial dataset containing 1,711 original images was collected and augmented to 4,104 images for model training and evaluation. Four YOLOv5 variants (YOLOv5s, YOLOv5m, YOLOv5l, and YOLOv5x) were trained for 50 and 150 epochs. Experimental results show that all models have high detection performance with precision, recall and mAP values above 0.98. YOLOv5s achieved the best real-time performance with 4.26 FPS and maintained competitive accuracy. Detection failures due to overexposure were observed at high illumination levels and illumination analysis indicated that detection can be reliably performed under lighting conditions below 25,000 lux. The proposed system provides a vision-based alternative for real-time inundation early warning applications.