Tt-slam: dense monocular slam for flat environment (IEEE 2021)

Published in ：2021 IEEE International Conference on Robotics and Automation (ICRA)

link ：https://ieeexplore.ieee.org/document/9561164

author ：Xi Wang    Univ Rennes, Inria, CNRS, Irisa, France

Marc Christie     Univ Rennes, Inria, CNRS, Irisa, France

Eric Marchand    Univ Rennes, Inria, CNRS, Irisa, France

translate ： Lucky stone

Abstract

This paper presents a new vision system for dense planar reconstruction using monocular camera SLAM Method ：TT-SLAM. This method uses a tracker based on plane template (TT) To calculate camera pose and reconstruct multi plane scene representation . By clustering a set of template trackers supported by super pixelated regions , Simultaneous estimation of multiple homography . And based on RANSAC Multi homography method of [1] comparison , Data association and key frame selection are handled by the continuity of template tracker . The nonlinear optimization process is applied to all homography to improve the accuracy of attitude estimation . Experiments show that , The proposed method is better than that based on RANSAC Multi homography method of [1] And other intensive methods SLAM technology , Such as LSD-SLAM or DPPTAM,

Introduce