Video stabilization based on motion segmentation
Seok-Jae Kang, Tae-Shick Wang, Dae-Hwan Kim, Aldo W. Morales, Sung-Jea Ko · 2012
This paper presents a new approach for video stabilization based on motion segmentation. The performance of video stabilization depends on the accuracy of the global motion. However, the scenes with large depth variation yield the performance degradation. In the proposed method, the feature points are detected and tracked, and their displacements are examined over the two adjacent frames. Motion segmentation is performed based on the similarity of the feature point displacements. The global motion is finally estimated using the feature points with the dominant motion which are located in similar depth ranges in most cases. Experimental results show that the proposed method provides the superior performance of video stabilization over the conventional method.