Abstract
Rolling shutter distortion becomes quite common with the increasing popularity of CMOS cameras in recent years. In this thesis, we present a novel correction method to remove the artifacts of videos captured by rolling shutter cameras. The proposed algorithm estimates the camera’s 3D rotation and 2D translation vectors simultaneously in an optimization framework. Unlike previous works which only focus on either translational or rotational motion, our rolling shutter correction model takes both kinds of motion into consideration to accurately reconstruct an undistorted video. In cases of pure camera translation, we allow the depth of field to be large, which breaks the limitations of previous works, and present a depth estimation method for a rolling shutter video. To alleviate the camera jittering problem, we also proposed a simple and efficient video stabilization method that can directly produce the stabilized video from the rolling shutter video. Experimental results on both synthetic and real videos are shown to demonstrate the effectiveness of our system and superiority over other methods.