Optical flow estimation technique for hazy scenes

Abstract

Hazy scenes limit the visibility and make the optical flow estimation task challenging due to low contrast images. Previous methods are not robust when applied to hazy image sequences because the two assumptions (brightness constancy and smoothness) break down due to the low contrast outdoor effects. Based on the pre-processing work on images, we applied an image dehazing method on image sequences first to handle haze and then estimated optical flow. The proposed method is based on a classical pyramidal technique. Results show that our method performs well on hazy scenes as compared to the other existing method.