Sliding window based monocular SLAM using nonlinear optimization

In this paper, a sliding window based real-time monocular SLAM is proposed. In our method, latest multiple states are estimated in a sliding window by using nonlinear optimization, and the other states are marginalized out from the sliding window. Meanwhile, we convert measurements corresponding to marginalized states into prior, so as to bound the computational complexity and improve the accuracy of state estimation without loop detection. Two experiments are designed to evaluate the accuracy and effectiveness of our method. The results show that the performance of our method is much better than the monocular ORB-SLAM, and our method can effectively estimate the sparse point cloud of map structure and camera motion with unknown scale.