Dust and Reflection Removal from Videos Captured in Moving Car

Abstract

The quality of videos captured in moving cars suffer from dust on the wind screen glass. Dust contaminates the captured videos and makes the videos blurred. Removing dust and restoring high quality dust-free and reflection-free video is a challenging task in the field of video stream processing. In this work, we present a pipeline of dust and reflection removal from the corrupted video streams. To the best of our knowledge, this paper is the first study how to effectively remove dust from video streams captured in a moving car. The pipeline is comprises of two steps. In the first step, it uses the state of the art image matting to model and resolve the captured frames as the merging result of a dust layer, a background layer and a matte layer. To refine the result, the second step employs variances of pixel streams to constrain the matte layer and optimizes the extracted dust layer, background layer and matte layer to make them spatially and temporally consistent in the streams. The test results demonstrate that our proposed method can effectively remove the dust and reflections in the video streams captured in moving cars.