An Improved RANSAC in Simultaneous Localization and Mapping

Simultaneous Localization and Mapping (SLAM) has been playing a more and more important role in intelligent driving and robotics. In order to reduce the error caused by missing frames and improve the accuracy of loop-closure detection, an improved random sample consensus (RANSAC) method (named LO*-RANSAC) is proposed. Firstly, inliers produced from conventional RANSAC are filtered iteratively to further narrow down the selection of qualified inliers. Secondly, an additional bundle adjustment (BA) optimization is performed on the estimated model to minimize the reprojection error. It is evaluated with ten sequences from TUM RGBD dataset and KITTI dataset, which cover both small and large scale, indoor and outdoor environment. Experimental results show that the proposed method generally performs better than ORB-SLAM2 and several other systems with higher accuracy and computational efficiency.