Aiming at the decrease of positioning accuracy caused by key-point matching error of left and right images of binocular SLAM,an optical flow algorithm based on projection transformation is proposed.Firstly,the corner points on the left image are extracted,and the positions of the corner points on the right image are calculated by the forward and inverse LK optical flow.Then,the homography matrix from right image to left image is calculated by using the successful matching corner points by RANSAC algorithm.Finally,the projection transformation image of the right image is calculated,and the corresponding position of the corner point of the left image is calculated by the forward and reverse LK optical flow,and the corner point is mapped to the original image of the right image.The experiments use partial sequences of the public datasets Euroc and KITTI,and the localization error of SLAM is evaluated by EVO tool.The results show that the improved algorithm can effectively reduce the corner point matching error with a certain increase in time complexity,and the root mean square error of binocular positioning is reduced respectively by 24.7% and 28.1% on average on partial sequences of Euroc and KITTI.