Offline or deferred
solutions were often employed for high quality and reliable results in current
video stabilization. However, neither of these solutions could be used for strict
real-time applications. Here, we proposed a practical and robust
algorithm for real-time video stabilization. To achieve this, a novel and efficient
motion model based on inter-frame homography estimation was
proposed to represent the video motion. An important feature of the proposed
motion model is that it will be updated at each frame input to reduce the accumulation
errors caused by parallax or scene changes. We also proposed a novel Kalman
filter for the motion smoothing and a novel mosaic algorithm for the video
completion. The proposed Kalman filter and mosaic algorithm
enabled us to develop a practical real-time video stabilizer which not just
produce steady video but also keep the full resolution of the original video.
We verified the proposed algorithm through a broad range of video sequences,
which demonstrated that our algorithm is computationally efficient
while being able to robustly stabilize videos with various challenges.
Here is the download link for the software and results,
software: https://www.dropbox.com/s/mbnmbio491j58it/software.zip?dl=0
results: https://www.dropbox.com/s/xe76tv6rxdlg60u/results.zip?dl=0
We also upload the software and results here,
https://drive.google.com/folderview?id=0Bzcq2ug6D6hBfmNlQzkyRXlaSWR3WlZOcGRKOWl6VlRhNmdMdjRsZ2pyQUlMUlBNbnQwNDg&usp=sharing
We also upload a multi-thread version source code and results here,
https://drive.google.com/file/d/0B9_80aruh3K2RGZRbXVWM1FoY3M/view?usp=sharing
If you meet any problem about the software or the source codes please send email to dongjing1756@163.com.
