Abstract:Aiming at the problem of real-time and accuracy of image mosaic,this paper proposes an image mosaic method based on improved ORB feature point detection.The method is improved in the way of extracting the surrounding points of image noise points and judging the condition of feature points.Firstly,according to the characteristics of ORB algorithm,16 points are selected to select only 4 points,which greatly reduces the number of comparison between centroid and peripheral pixels.Then,according to the characteristics of extensibility of feature points,four related points are selected in the outer circle for comparison,which can quickly distinguish noise points or feature points,and improve the matching accuracy of feature points.This paper also improves the random sampling consistent RANSAC(Random Sample Consensus)algorithm,adding appropriate new feature points in the iterative process,improving the internal point ratio,reducing the number of iterations,which not only improves the efficiency,but also can calculate a more suitable transformation matrix to achieve better splicing effect.Experiments show that the improved ORB algorithm,improved RANSAC algorithm and Poisson fusion algorithm can achieve real-time,efficient and accurate image mosaic.