Video Stabilization with GPU

This paper proposes a fast computational method of video stabilization using the Graphics Processing Unit (GPU) that removes unwanted vibrations from videos. The video stabilization is composed of estimation of the global motion, removal of the undesired motion and mosaicking. When They are processed with CPU, the computational cost for the global motion estimation is very high. We improve the speed of this computation with GPU that enables parallel processing. Our method can obtain the result by forwarding the frame image of the video to GPU as texture data, and drawing the calculation result to the offscreen buffer. Although the transfer speed from GPU to CPU is very slower than the other way around, the method only has to transfer one pixel data from GPU.

[1]  J.-Y. Bouguet,et al.  Pyramidal implementation of the lucas kanade feature tracker , 1999 .

[2]  Harry Shum,et al.  Full-frame video stabilization with motion inpainting , 2006, IEEE Transactions on Pattern Analysis and Machine Intelligence.

[3]  Robert Strzodka,et al.  Real-time motion estimation and visualization on graphics cards , 2004, IEEE Visualization 2004.