Robust global motion estimation for video security based on improved k-means clustering

Abstract

The global motion vectors estimation is the most critical step for eliminating undesirable disturbances in unsafe video. In this paper, we proposed a novel global motion estimation approach based on improved K-means clustering algorithm to acquire trustworthy sequences. Firstly, the speeded up robust feature algorithm is employed to match feature points between two adjacent frames, and then we calculate the motion vectors of these matching points. Secondly, to remove the local motion vectors and reduce redundancy from the motion vectors, an improved K-means clustering algorithm is proposed. Thirdly, by using matching points from richest cluster, global motion vectors are calculated by homography transformation. The experimental simulation results demonstrate that the proposed method can obtain significantly higher computational efficiency and superior video security performance than traditional approaches.