Optimization of position and pose estimation by using Epipolar line segmentation optimization BA algorithm

Traditional Key Frame Bundle Adjustment Based Visual SLAM Algorithm, There are many shortcomings in minimizing the reprojection error that cause the pose estimation to be inaccurate. This paper optimizes the error function of BA, modifies the traditional reprojection error function and introduces the 2D Epipolar line constraint based on the limit division on the basis of the 3D constraint, The improved algorithm is applied to the ORB-SLAM2 system, and the comparison test is carried out on the EuRoC public data set. Experiments on positioning error and algorithm performance show that the improved ORB-SLAM2 system pose estimation is more accurate and robust than before.