Ego-Motion Estimation for Vehicles With a Rolling Shutter Camera

Modern automotive vehicles are often equipped with a budget commercial rolling shutter (RS) camera. These devices usually produce distorted images due to the camera's inter-row delay while capturing the image. They will defeat the following computer vision tasks, such as ego-motion estimation. In this paper, we present a closed-form solver for a vehicle-mounted RS camera ego-motion estimation problem with only one point-match between two consecutive distorted RS images, called RS-1 pt. To the best of our knowledge, this is the first work to address the ego-motion estimation problem for vehicle-mounted RS cameras specifically. Thanks to the 1 pt minimal solver, extensive experiments demonstrate the proposed method surpasses classical global shutter based (GS-based) solvers and existing RS-based solvers by a significant margin in terms of both efficiency and effectiveness. Besides, we synthesize and self-collect three vehicle-mounted RS camera datasets with ground-truth GS images and trajectory, which are absent from the existing works and will be publicly available as a benchmark to the research community.