Author(s): Guocheng An; Tiantian Du; Yanwei Zhang

Linked Author(s):

Keywords: River surface velocity; VideoFlow; Optical flow; Illumination conditions; Attention mechanism

Abstract: In the field of hydrological monitoring, accurate assessment of river surface velocity (RSV) is crucial for flood control, disaster prevention, and soil erosion mitigation. RSV measurement techniques can be classified into contact and non-contact methods based on environmental conditions. This study proposes a deep learning-based optical flow computation method for RSV measurement, utilizing the VideoFlow optical flow estimation architecture to effectively track and identify particles in water flow. To enhance accuracy, we define the region of interest for optical flow estimation and perform image distortion correction. By simulating real river scenarios in the laboratory and collecting data, we validate the reliability of this method in complex environments. Results indicate that in river scenes with velocities not exceeding 2.5m/s, the relative error estimated by this method is below 15%. Compared to traditional methods, it offers lower cost and enables continuous real-time dynamic measurement of the entire flow field, providing a more effective solution for hydrological monitoring.

DOI: https://doi.org/10.3850/iahr-hic2483430201-256

Year: 2024