Video Stabilization Algorithm Research Based On SIFT

In the process of photograph, there are always some shake of the carrier which makes the video output unsteady, it’s bad for watching or processing. Electronic video stabilization technology is to eliminate the jitter of the video, in order to obtain smooth video image.It is a new generation skill, by using image processing technology to achieve stabilized video, it has advantages in high precision,low energy, small volume and low cost,etc. And just because this technology is widely applied, it has also become a new trend of the video stabilization.Electronic video stabilization mainly include three modules which is Motion estimation, Motion filtering and Motion compensation. Motion estimation aims to get the global motion vectors between frames. Motion filter can effectively remove unintentional jitter meantime keep the camera’s normal scanning motion. Motion compensation is to use the obtained motion parameters to correct the position of each frame. Among them, Motion Estimation, which directly determines the performance of the system, is most significant. In this thesis, the current mainstream algorithms belong to the classic Block Matching methods, such as three, four steps search, diamond search were deeply studied. Based on the simulation results, the advantages and disadvantages of these methods were compared.Based on the block matching algorithm, this thesis expresses the principle, method and process of the electronic stabilization in detail. Then, taking the complexed motion vedio as processing object, a global motion estimation algorithm based on SIFT matching was proposed. This algorithm has good invariance in scale transform and strong ability to match. In improving the precision, using RANSAC algorithm, which effectively remove the wrong matching of SIFT, and applying a new repair technology based on motion recovery to make motion inpaiting to the undefined area of the image edge. And full frame stabilization with good visual effect was achieved. In real-time processing, choosing the speed up version of SIFT-SURF algorithm, which realizes real-time video stabilization, meantime ensures a certain precision.In this thesis, the peak value signal-to-noise ration between frames have been improved grately in using SIFT. Finally, using Kamplaer processing software to deal with the stabled video frames, and comparing it with the advanced Deshaker sofeware, as a result, the algotithm of this thesis achieved a more satisfying effect. Experimental results show that this algorithm can make accurate estimation to the translation, rotation and affine transformation parameters of the video, atter being disposed, the dynamic image sequence has good stability, and the jitter of it certainly can be eliminated.