Videos from handheld devices often contain abrupt and irregular camera motion, which is a limiting factor for both automated and manual video analysis. Video stabilization aims at removing these undesired motions and transforming the input sequence to a stable, visually plausible video. State of the art algorithms focus on stabilization strength, robustness, processing speed and latency, requirements which are often contradictory.
Some examples are presented below
The research behind
Video stabilization pipeline overview
The basis for the modification of optical flow
In order to compute the optical flow we start from a sparse set of point matches and interpolate at the missing positions of the dense grid. The semantic boundaries are used as a guide for the interpolation with an edge preserving geodesic distance.
- K. Karageorgos, A. Dimou, A. Axenopoulos, F. Alvarez, P. Daras, "Semantic Filtering for Video Stabilization" , – 14th IEEE International Conference on Advanced Video and Signal Based Surveillance (AVSS), Lecce, Italy, 29 Aug – 1 Sept, 2017