Monocular Video Stream Depth Estimation SLAM System for Low-Light Indoor Environments

This paper proposes an RGB-D SLAM system based on video stream depth estimation. The system obtains RGB data through the camera and uses the depth estimation image as input information, achieving accurate pose estimation and dense reconstruction. Aiming at the problem of lack of correlation in depth estimation of adjacent frames, a depth estimation based on video streams is proposed, which effectively improves the smoothness of depth estimation of adjacent frames. In terms of feature extraction, the robust pose estimation method based on improved ORB is adopted, which effectively improves the accuracy of camera pose estimation and solves problems such as ghosting and misalignment in 3D reconstruction in low light environment. Experiments conducted in different characteristic scenarios show that compared with Elasticfusion and ORBSLAM2, the system has lower pose root mean square error and a more accurate reconstruction model.