Video Mosaic Algorithm Based on Invariant Feature
Gan Xin-sheng · Command Control & Simulation · 2009
Video mosaic is an integral part of aerial video surveillance.There are two steps in video mosaic,image registration and image fusion.The approach based on invariant features is proposed in this paper.Panoramic image mosaic is implemented through frame-to-frame mosaic in the proposed method.Gradient orient histograms determine the rotate angle between two neighboring frames in the first step.The corner is detected and its feature is represented by scale invariant feature transform(SIFT).Coarse match is based on SIFT and the rotate angle.The results are corrected by Random Sample Consensus(RANSAC).Transform matrix is estimated from inliers.The fusion step adopts gray weighted method.Experimental results demonstrate the effectiveness of the algorithm,even when there are moving target in videos.