Video stabilization algorithm using feature tracker and adaptive motion filter

Motion vector was detected by means of feature tracker and multi-resolution technique. Motion parameters were computed using nonlinear least square method. An adaptive motion filter was constructed considering the number of local extremum in a period of time as the criterion of video vibration, which filtered the motion parameters to decrease vibration by adjusting the filter length. The stabilized video was obtained after we employed the filter's output to compensate movement between the reference frame and subsequent frames. Moreover, motion correction was introduced to avoid error propagation. Experimental results indicate the algorithm is effective and robust with low computational complexity after extensive experiments over a wide variety of videos.