Robust optical flow estimation using underwater color images

Abstract

Accurate and efficient optical flow estimation is a major step in many computational vision problems, including 3-D mapping applications. While processing of gray scale image sequences has been the main approach, some authors have recently investigated and reported the use of color images. The main advantage in the use of color images is to exploit redundant information at each pixel, namely 3 channels, as potential cues for visual motion. Moreover, several spectral-dependent factors including medium attenuation may justify the use of color cues in underwater, calling for a physics-based approach to optical flow analysis. This paper explores a simplified underwater image formation model 1) to provide some preliminary insight on the information encoded in various color channels, 2) draw some general conclusions on the use of color and its various representations, and 3) verify if the results are consistent with the experiments with 7 sets of coral reef video recoded in relatively clear shallow waters of the Bahamas. These results suggest that optical flow algorithms based on the HSV representation can provide improved localization and accuracy.