Poster:Multi-Modal-Based Video Stabilization for Mobile Devices in Dynamic Scenes

Handheld devices such as smartphones and cameras often produce shaky videos due to reasons such as hand tremors and movement. Current video stabilization methods mainly estimate camera motion based on feature points, such as corner points, and then smooth the motion to obtain a stabilized video. However, in reality, most videos are shot in dynamic scenes, where some feature points are extracted from moving objects. They seriously affect the estimation of camera motion, and result in unsatisfactory video stabilization. In this paper, we propose a multi-modal-based video stabilization method for dynamic scenes using video frames and gyroscope. The basic idea is to eliminate dynamic feature as their estimated rotation significantly differs from the actual rotation obtained from the gyroscope data. The experimental results show that our method effectively enhances the quality of shaky videos in dynamic scenes.