Adaptive video stabilization based on feature point detection and full-reference stability assessment

Yiming Wang, Qian Huang, Jiwen Liu, Chuanxu Jiang, Mingzhou Shang

Abstract: Video stabilization is an important video enhancement technique that removes shaky motion and produces stable videos with good visual quality. Previous feature point matching methods are proposed to estimate motion information. However, video stabilization based on feature points matching will decrease or fail to match feature for video sequences lacking feature point, which causes poor results. Also, there is not a recognized method to measure the performance of video stabilization. To solve these problems, we propose an adaptive video stabilization based on feature point detection and full-reference stability assessment. In the proposed method, appropriate video stabilization algorithms are firstly selected by detecting the number of feature points and camera trajectory optimization is used to retain original motion information. Secondly, we propose a full-reference stability assessment to measure video stabilization performance. Furthermore, video stabilization is assessed from three aspects on the video dataset DeepStab. Finally, experimental results demonstrate the promising performance of our proposed algorithm in terms of subjective and objective evaluations.