基于ORB特征的改进误匹配剔除算法OA

An improved eliminating mismatching algorithm is proposed to solve the problem of ineffiency and mismatching in complex environment when ORB-SLAM is used in feature registration algorithm.First,when extracting orb feature points,the glob-al threshold of algorithm is changed to adaptive threshold to improve the speed of feature point extraction.Then,the grid based on motion statistics(GMS)algorithm is used for reducing the loss time of processing key frames,and the progressive uniform sampling(PROSAC)algorithm is used for correcting and optimizing the situation where GMS algorithm is prone to mismatch.The experimen-tal results show that this method can effectively eliminate the mismatches of GMS algorithm,which can compare with the random sampling agreement(RANSAC)algorithm at the same time.Under the same conditions,the average time consumption is reduced by 81.56％.Finally,the method is integrated into the binocular sparse stereo matching module of ORB-SLAM3.Compared with the original system,this scheme has higher positioning accuracy and stronger real-time performance.