Research On Image Matching Algorithm Based On Grid Motion Statistics

As the basis of computer vision,image matching is essential in photogrammetry,medical diagnosis,driverless and so on.Image matching algorithm based on point feature has been paid more and more attention by researchers in related fields because of its fast computing speed and high matching accuracy.At present,many excellent algorithms have been proposed in feature detection and description.However,how to accurately and quickly separate true and false matching from initial matching is still a challenging and key problem in feature matching.Grid based motion statistics(GMS)algorithm is used to eliminate the wrong matching according to the principle of motion smooth constraint,which can meet the needs of high precision and fast matching in complex scenes.This paper is mainly based on GMS algorithm,aiming at the shortcomings of the algorithm to improve,and on this basis,an improved image mosaic scheme is adopted.The specific research contents are as follows:(1)Master the processing flow of image matching algorithm,compare GMS algorithm with traditional error elimination matching algorithm,and analyze the matching effect of GMS algorithm when the image changes in illumination,perspective,blur,etc.Experimental results show that GMS algorithm can solve the problem of fast matching in complex scenes and effectively improve the quality of feature matching.(2)The rotation angle of the image is determined by calculating the angle change of the main direction of the feature points when the image is rotated,and the best shape of the motion kernel is determined directly according to the rotation angle to avoid the GMS algorithm to calculate the feature matching results of different shapes of the motion kernel circularly;When the image scale changes,the scale factor between the images is determined by calculating the ratio of the distance between the feature points,and the meshing method of the image to be matched is determined according to the scale factor,so as to avoid the GMS algorithm to calculate the feature matching results of different scale factors circularly;When the image is rotated and scaled at the same time,the above two improved algorithms are combined to realize the feature matching of the two images.The results show that when the image is rotated,the average running time of the improved algorithm is 76% less than that of GMS algorithm;When the image is scaled,the average running time of the improved algorithm is 71% less than that of GMS algorithm;The average running time of the improved algorithm is 6% of that of GMS algorithm when image rotation and scale change occur simultaneously.To sum up,the improved algorithm greatly improves the matching speed on the basis of ensuring the matching effect of GMS algorithm,and solves the problem of fast matching when the image is rotated and the scale changes.(3)The feature point detection is performed in a block method,which solves the problem of uneven distribution of feature point detection in traditional algorithms.The combination of cross-check and GMS algorithm is used to eliminate false matches to obtain a high-quality feature matching set,and the RANSAC algorithm is used to calculate the homography matrix of the two images to complete the image mosaic experiment.The results show that the RANSAC homography matrix obtained by the improved algorithm is more global and improves the stability and accuracy of image mosaic.