Stereo images obtained from sequential aerial photographs of river freshets contain a lot of useful information for river hydraulics as well as river engineering works. The image correlation analysis is a standard technique for analyzing the images, but it sometimes produces anomalous results for the area where color fluctuation of the water surface is small. On the other hand, a classical manual analysis through stereo eyeglasses usually works well even in such unclear parts of the images, but the manual analysis costs much labor and time.
In this study, a new technique to analyze stereo images of river flows is developed by combining the above mentioned two methods: A small amount of suggestions are made by using the results from the manual analysis for the image correlation analysis on a computer, and then the results of the second trial of the latter is much improved without any significant time consumption.