Vision restoration of dangerous head light glare affected scenario

摘要

Vehicle's head light glare is very dangerous problem of driving at night. It instantly reduces driver sight to obstruction on road. The high luminance in glare distorts a driver's vision in large area and very difficult to restore with single frame image processing. In this paper, we present a vision restoration for glare affected scenes by using previous frames information regarding to motion vector (MV) that are derived from video decoder. We adopt three-dimensional recursive search (3DRS) motion estimation method to re-estimate new motion vectors (MV1) to select effective information from the previous non-glare affected frames. Then MV1 are extrapolated and refined. The defected area is restored using motion compensation. The results show that our proposed solutions can restore vision from glare area and increase significant driver's obstruction detection.