Performance Evaluation of various Temporal Derivatives for Stabilization of Videos with Large Moving Objects

The efficiency of the various video stabilization methods depends open the efficiency of the motion estimation stage.. Various Temporal derivatives are designed to efficiently estimating the inter-frame motion or optical flow for video stabilization. The performance of the video stabilization methods degrades under the presence of large moving objects in the scene. Therefore, this paper compares the performance of video stabilization method by estimating the motion using the various temporal derivatives. The video stabilization method in this paper uses differential Taylor series expansion based motion estimation followed by IIR filter for smoothening the motion. A 2D-affine parametric model is used for motion estimation. For performance evaluation the MSE and temporal derivatives in X and Y directions using different derivative operators are compared. it is found that the four point temporal derivative gives the minimum inter-frame error under large object in scene