RivAIr: A custom-designed UAV-based sensor for real-time water area segmentation and surface velocity estimation

• RivAIr: A UAV-based edge-computing system for real-time river flow monitoring. • Integrates CNN and optimized optical flow for accurate surface velocity estimation. • Achieves 90% confidence in water detection and velocity estimation with −3% error. • Delivers real-time flow analysis in 40 s, operational up to ∼ 800 m from the pilot. Real-time monitoring of river surface velocity and flow extent is crucial for flood risk mitigation, sustainable water resources management, and ecosystem protection. However, high computational demands and limited high-resolution river cross-section data make real-time river monitoring impractical, particularly for large or ungauged rivers during emergencies, impeding rapid analysis and decision-making. A key limitation is the lack of an all-in-one, efficient tool combining machine learning-based segmentation with real-time optical flow estimation on edge-computing Unmanned Aerial Vehicles (UAVs). To address these challenges, we introduce RivAIr, a novel UAV-mounted sensor system designed for real-time river monitoring. This self-contained, edge-computing module integrates custom hardware with real-time machine learning and optical flow algorithms, enabling onboard water segmentation and river surface velocity estimation. Unlike existing methods that rely on post-processing, RivAIr is purpose-built for real-time field deployment in data-scarce or emergency scenarios. RivAIr integrates YOLOv8 with an optimized Farneback dense optical flow algorithm focused on water bodies, improving motion estimation accuracy while reducing runtime for real-time UAV processing. Validated along the Basento River (Basilicata, Italy) under controlled conditions, the optimal configuration of RivAIr achieved a 90% confidence score in water surface area detection and velocity estimation with a −3% error relative to in-situ measurements, showing strong consistency with in-situ flow observations. Notably, RivAIr completed end-to-end processing in approximately 40 s, demonstrating operational feasibility across altitudes and distances up to 800 m from the ground unit. RivAIr represents a significant advancement in real-time hydrological monitoring, essential for both routine and emergency applications.