Author(s): Jumpei Yagi; Kojiro Tani; Ichiro Fujita; Keisuke Nakayama
Linked Author(s): Ichiro FUJITA
Keywords: Optical flow; River flow measurement; LSPIV; STIV; Image velocity measurement
Abstract: Optical flow technique (OF) is one of the image-based techniques basically for detecting motion of moving objects. In recent years, accuracy and efficiency of the technique has improved significantly with developments of various types of novel algorithms and it has come to be used in quantitative flow measurements. The techniques are categorized into sparse and dense methods. The latter method based on the Horn-Schunck method are examined in this research. The dense-type OFs have the advantage that the optical flow vector is obtained at each pixel very efficient compared with the particle image velocimetry (PIV). Although there are many OF algorithms, we focused on DeepFlow in this study. DeepFlow was applied to river surface flow and compared with LSPIV and STIV. DeepFlow’s accuracy was comparable to PIV, which is one of conventional flow measurement techniques. In this case, obliquely viewed surface images of the Shinano River in Japan were shot from a riverbank, in which surface textures composed of air bubbles or surface ripples were used as natural tracers instead of supplying tracers. It was made clear that OF is applicable to the river surface flow as long as a clear texture representing the surface flow is visible at the measurement site.