EDPLVO:Efficient Direct Point-Line Visual Odometry note

ABSTRACT

This paper extends the photometric error for lines.  we proved that the 3D points of the points on a 3D line are determined by the inverse depths of the endpoints of the 2D line, and derive a closed-form solution for this problem.   This property can significantly reduce the number of variables to speed up the optimazition, and can make the collinear constraint exactly satisfied. Furthermore, we introduce a two-step method to further accelerate the optimization, and prove the convergence of this method.The experimental results show that our algorithm outperforms the state-of-the-art direct VO algorithms.

能否在不增加计算负荷的情况下利用共线性？

为此，他们在之前工作（DPLVO）的基础上进行了进一步的研究。并提出了一种新的算法——EDPLVO。

特征法和直接法参考：EDPLVO: Efficient Direct Point-Line Visual Odometry论文笔记_努力glow .的博客-CSDN博客_line matching slam

 

作者提出了一种使用点和线的高效的直接视觉里程计（visual odometry，VO）算法—— EDPLVO .他们证明了，2D 线上的 3D 像素点由 2D 线端点的逆深度决定，这使得将光度误差扩展到线变得可行。

论文中使用的线段提取方法是LSD，而LSD本身存在的问题是受到运动模糊或者取整的影响，线段提取本身存在一定的不确定性，不确定性反应在线段的宽度上。这里个人理解是线段可能并不准确，会由于抖动、模糊等原因，让坐标不准确，也就是说提取出来的一条线段，其周围一定范围可能仍然属于这条线段。基于此，论文对线段增加了一个支持域，支持域宽度用ρ表示，以此来衡量线段的不确定性，ρ越大，说明线段提取结果周围很大的范围都可能是线段的一部分，所以不确定性大。

所谓共线约束，实际上就是指一条线段在所有观测到的图像上检测的2d线段，反过来投影其端点应该保持共线.

Experiments

The ICL-NUIM dataset contains 8 indoor sequences, and the TUM mono VO dataset has 50 loop-closed sequences from indoor and outdoor environments. 

Related-Work:

Although deep learning based line detection methods [19]–[23] show promising results, these methods are generally computationally demanding.

"deep hough transform for semantic line detection,” IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021

"Holistically-attracted wireframe parsing," in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition,2020,

"learning point-pair graph for line segment detection," in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019,