Non-contact canal discharge measurement using the Farneback dense optical flow algorithm

Open-channel discharge measurement underpins irrigation scheduling, water allocation, and flood management decisions, yet contact-based techniques carry real operational risk at high flows and cannot support continuous automated monitoring. Field validation of the Farneback dense optical flow algorithm for non-contact discharge estimation was carried out at Dhayari canal, Pune, Maharashtra, India, with simultaneous measurements from a Vega C-21 radar Doppler sensor and an acoustic Doppler sensor serving as reference instruments. A Nikon 16 MP camera, mounted at a fixed overhead position, recorded canal flow at 24 fps across 17 measurement grids at 1.0 m intervals, covering the full 17.0 m top width of a trapezoidal section (bottom width 9.0 m, FSL 4.5 m, cross-sectional area 58.5 m2). Surface velocities were derived from consecutive frame pairs, scaled to mean velocity using the correction factor α = 0.85, and adjusted for flow direction through cosine correction. Grid-level discharge was computed and summed across all 17 sections, yielding QF = 14.219 m3/s, which sits between QR = 13.860 m3/s from radar and QA = 14.494 m3/s from acoustic sensor, at an overall accuracy of 97.54%. Camera-based optical flow matches instrument-grade discharge measurement without placing personnel or equipment at the water surface.