Research And Application Of Homography Matrix Estimation Algorithm Based On Voting Mechanism

Homography transfers points of the same scene from one image to another.Homography can efficaciously establish the relationship between two images of one scene.It also can relate to camera internal parameters and camera external parameters,which means we can get them through homography.This paper takes homography estimation as the core.We use the result of homography estimation to stereo matching and get the 3-D coordinate of correspondences.Working on this paper shows as follow:(1)Robust homography estimation bases on V_RANSAC algorithm.For the correspondences used estimating homography is not accuracy,direct homography estimation result has a quite big error,which can not meet the practical application requirements.At present,many studies of homography estimations concern on using parameter optimization method to estimate the homography.Among them,the RANSAC algorithm has more application in this respect.However,if the quality of correspondences is quite bad,the iterations of RANSAC are too big.What is more,the result is not enough robust,so in this paper,we add voting thought to RANSAC.We use the result of homography to vote correspondences and use the result of voting guide the selection of correspondences.We can use fewer iterations choosing high-quality correspondences.Through simulated data experiments and real data experiments,V_RANSAC algorithm can reduce iterations,and then reduce running time.It has more robust and precision at the same time.(2)Boost matching based on homography and Gray Scale Truncation Number.In one aspect,too big searching area leads to too much computing time cost.In other aspects,more matching and no matching lead to the result of matching non-uniqueness.In this paper,we use the homography result of last part weeding our wrong correspondences.Then,guiding the eliminated correspondences with homography and Gray Scale Truncation Number effectively narrows the search scope and greatly shortens the running time.(3)This paper offers a kind of new measure function called Gray Scale Truncation Number.This measure function has low computing cost.According to experiments,boost matching can increase correspondences effectively.By comparing different measure functions under the same conditions,our measure function used to boost matching can enlarge correspondence precision effectively,especially when the precision of the initial matching is quite low.(4)Acquisition of 3-D coordinates of correspondences and visualization of experimental results.For we do not do camera calibration,we use the picture having camera parameters to do experiments.We use the result of stereo matching result of the last part to do 3-D reconstruction based on correspondences.We visualize the result.The visual result shows that boost matching can correct many obvious wrong correspondences.Use the triangle method can get the 3-D coordinate of space points.