In this paper, we introduce a new method for detecting debris flow in mountainous rivers using unit vectors of RGB values of stream water surface images. In the Imokawa River in Niigata Prefecture, Japan, turbidity was evaluated relatively using unit vectors of RGB values by comparing turbidity sensor observation data with image analysis results for images of turbid water associated with low or medium level flows. In the Kurodakesawa River in Hokkaido, we examined the validity of debris flow detection using the same analysis of unit vectors of RGB values for images of debris flow and turbid water caused by low or medium level flows, and then, we revealed differences in characteristics between debris flow and turbid water that did not lead to debris flow. We also showed that there is a linear correlation between turbidity and changes in RGB unit vectors, and that there is a unique direction vector for each stream. Furthermore, by setting a plane orthogonal to this direction vector, we proposed a method to set a threshold for not only quantitative evaluation of turbidity, but also a high turbidity zone that would detect the occurrence of debris flow. Currently, quantitative evaluation of turbidity by image analysis is not suitable for use as an instrument for observing turbidity due to its coarse accuracy, but it is effective for use as a monitoring instrument. The results of this study are expected to be used effectively in the future for monitoring debris flow using images from CCTV and other monitoring cameras, such as detecting turbidity of the precursor phenomena of debris flow.