Research on Distance Calculation Method of autonomous vehicle Based on Direct Sparse Vision

It is very important for autonomous vehicles to obtain their own accurate positioning. Currently, visual SLAM positioning technology is considered as the mainstream positioning means due to its rich scene information and low cost. In this paper, the principle of visual SLAM is analyzed. This article briefly introduces the direct method and feature point method in visual SLAM, outlines the general development status of visual SLAM, and ultimately chooses to conduct subsequent research on the basis of DSO-SLAM. Aiming at the problem of cumulative error, a monocular direct sparse visual odometry with loopback is proposed. Through the transformation of Bag Of Words (BOW) model, the loopback frame is found out, and more constraints are entered to optimize the pose, aiming at eliminating the accumulated errors in the long-term operation of the algorithm and improving the positioning accuracy. This article conducted comparative experiments with DSO on public datasets, verifying that in the presence of loopback scenarios, DSO-IMP can significantly eliminate cumulative errors and improve positioning accuracy. This article also conducted experiments and found that DSO-IMP can correctly detect loops in actual motion scenes, and use loops to correct trajectories, which has certain practicality.