Robust and accurate river flow measurement by Space-Time Image Velocimetry (STIV) with improved deep learning technique

Space-time image velocimetry (STIV) is one of the image-based river flow measurement methods, which has been widely used because of its accuracy, robustness, and ease of use. Although fully automatic STIV measurement under adverse conditions has been a challenge for many years to establish a steady real-time measurement system, the authors recently have shown the possibility of solving this problem by the application of deep learning to STIV. In this study, to further improve the performance of STIV with deep learning, the data augmentation, updating of the deep learning model, and automatic optimization of search line settings were examined. As a result, it was confirmed that the accuracy and robustness were further improved without parameter tuning, which suggests the realization of a stable automatic flow measurement system that does not require fine adjustment by experts according to the shooting conditions.