视觉slam的一些论文整理

原文链接：https://blog.csdn.net/weixin_37724881/article/details/98752309

常规的VSLAM关键算法点研究整理：
（1）图象校正
采用现成的矫正算法（180度广角）
相关论文：
• Multi‐robot Mapping Using Omnidirectional‐Vision SLAM Based on Fisheye Images
来自 https://onlinelibrary.wiley.com/doi/full/10.4218/etrij.14.0114.0584
本文提出了一种基于Lucas-Kanade光流运动检测和通过安装在机器人上的鱼眼镜头获取图像的目标提取方法的多机器人全向视觉同步定位映射(SLAM)全局映射算法。多机器人映射算法使用从所有单个机器人获得的地图数据绘制全球地图。全局映射需要很长时间来处理，因为它在搜索所有区域时交换来自单个机器人的地图数据。全向图像传感器能够同时测量机器人周围的所有信息，在目标检测和映射方面具有许多优点。通过对目标特征点的修正，改进了修正算法的过程计算。该算法包括两个步骤:首先，基于面向单个机器人的全向‐视觉SLAM方法创建局部地图。其次，通过合并来自多个机器人的单个地图生成全局地图。通过与实际地图的比较，验证了该算法的可靠性。
• Multi-Robot Avoidance Control Based on OmniDirectional Visual SLAM with a Fisheye Lens Camera.pdf
• Large-Scale Direct SLAM for Omnidirectional Cameras.pdf
• 2018-12-11-CubemapSLAM: A Piecewise-Pinhole Monocular Fisheye SLAM System 鱼眼镜头slam
我们提出了一个实时的基于功能的SLAM(同时)定位与测绘)系统为鱼眼相机的一大特色视场(FoV)。大型FoV摄像机适用于大型户外活动SLAM应用程序，因为它们增加了视觉上的重叠连续帧，捕捉更多属于静态部分的像素的环境。然而，目前基于功能的SLAM系统就是这样因为PTAM和ORB-SLAM将他们的相机模型限制在针孔上。来为了弥补这一空缺，我们提出了一种新的SLAM系统利用完整的FoV而不引入失真的cubemap模型

（2）相机标定
内参：棋盘格标定法
• 2018-12-29-Online Inertial-Aided Monocular Camera Self-Calibration 相机自标定
在本文中，我们提出了一种新颖的执行方式一个在线相机校准，没有任何事先知识的环境，基于IMU的读取和跟踪一个单一的3D点投射到相机的图像帧摄像机四处移动。我们打算使用我们的方法与视觉测程或视觉惯性导航一起使用系统中，由于这类算法对相机的固有参数的精度，但他们通常采用离线标定方法，并将标定结果作为常量。但是我们知道有几个因素可以随着时间改变相机的固有参数。我们的方法可用于在更改这些值时纠正这些值因为这是一个在线的方法。的附加贡献本工作中提出的方法，是将其估计，与摄像机的本征参数，由运动执行算法运行时的摄像头。我们的方法使用了EKF来执行它的估计。我们证明了在现实世界的实验中，我们可以使用这种方法利用该方法对摄像机进行标定和运动估计

（3）特征点提取算法
ORB特征点提取
• 2018-11-12-Evaluation of Lightweight Local Descriptors for Level Ground Navigation with Monocular SLAM对比不同描述子对slam影响
• [ICRA 2018] A Monocular SLAM System Leveraging Structural Regularity in Manhattan World国内慢慢也有实验室开源了原理与code
曼哈顿世界的结构特征编码有用的几何信息的并行性，正交性和/或场景中的共面性。通过充分利用这些结构特点，提出了一种新颖的单目SLAM提供精确的相机姿态估计系统和3 d地图。
• 2018-09-14-Good Line Cutting: towards Accurate Pose Tracking of Line-assisted VO/VSLAM 对于线特征提取线内信息最多的一段做SLAM，声称对低纹理，运动模糊鲁棒

（4）特征点匹配算法（降低特征点的标准，降低匹配的标准）暴力匹配和FLANN
数据关联方面:
特征点的选取方法
• [ICRA 2018] Feature-constrained Active Visual SLAM for Mobile Robot Navigation路径规划时候考虑特征点数量,代码在此
• DOVO: Mixed Visual Odometry Based on Direct Method and Orb Feature: 通过特征点个数判断用直接法与光流法。交替使用
• 2018-12-07-Two-pass K Nearest Neighbor Search for Feature Tracking 增强的ＫＮＮ特征匹配方法（sift）

（5）深度估计
• Adaptive Baseline Monocular Dense Mapping with Inter-Frame Depth Propagation
来自 https://ieeexplore.ieee.org/abstract/document/8593936
第一种是自适应基线匹配代价计算，使用序列输入图像为每个像素提供宽基线观测。第二种是帧间传播深度滤波器，它以一种鲁棒的概率方式集成了同一物理点的序列深度估计。两个贡献被集成到一个单眼密集的测绘系统中，该系统为针孔和鱼眼摄像机实时生成深度地图。我们的系统是完全并行的，可以在Nvidia Jetson TX2上以超过25 fps的速度运行
• 2018-09-14-Deep Virtual Stereo Odometry:Leveraging Deep Depth Prediction for Monocular Direct Sparse OdometryCremers出品，用单目深度估计+DSO

（6）VO位姿估计
• 2018-11-26-Fast and accurate visual odometry from a monocular camera
• 2018-10-11-A Structureless Approach for Visual Odometry
三点改进： 1) the complexity of our solution is lower than those of the state-of-the-art methods, 2) no extra matrix operations required to eliminate map points, 3) no need guesses on map points’ initial locations.
• [CVPR 2018] Unsupervised Learning of Monocular Depth Estimation and Visual Odometry with Deep Feature Reconstruction Ian Reid出品，代码在此。无监督学习的单目深度估计与VO估计
• Lightweight Visual Odometry for Autonomous Mobile Robots
• 2018-09-14-Stereo Vision-based Semantic 3D Object and Ego-motion Tracking for Autonomous Driving 沈老师出品，既估计相机位姿，也估计物体位置，对动态物体鲁棒
• 2018-10-20-Four- and Seven-Point Relative Camera Pose from Oriented Features 利用特征点方向信息计算相对位姿

（7）闭环检测
（可结合深度学习)
• [ICRA 2018] Assigning Visual Words to Places for Loop Closure Detection 使用GNG clustering
• A Loop Closure Detection Algorithm in Dynamic Scene： 动态环境下回环检测algorithm

（8）优化算法（BA等）
• [CVPR 2018] ICE-BA: Incremental, Consistent and Efficient Bundle Adjustment for Visual-Inertial SLAM新的优化库
• 2018-11-12-A review on graph optimization and algorithmic frameworks

（9）周围场景的快速三维重建算法
（sfm）
• 2019-01-12-Real-time Monocular Dense Mapping of Small Scenes with ORB Features
单目ORB SLAM重建
• 2018-12-20-Real-time 3D scene reconstruction and localization with surface optimization
• 2018-09-26-GPU-accelerated feature tracking for 3D reconstruction GPU 加速
• 2018-09-28-Real-time Graph-Based 3D Reconstruction of Sparse Feature Environments for Mobile Robot Applications

（10）点云
• PCR-Pro: 3D Sparse and Different Scale Point Clouds Registration and Robust Estimation of Information Matrix For Pose Graph SLAM https://sites.google.com/view/pcr-pro

（11）六自由度跟踪模型
模型的体现方式
• CVPR 2019 | Stereo R-CNN 3D 目标检测
https://mp.weixin.qq.com/s/3JzwA2HAzoWtE_j2UhZCSw

（12）建图

（13）应用平台
• 2018-11-12-Incremental Feature Forest for Real-Time SLAM on Mobile Devices
• MULTICOL-SLAM - A MODULAR REAL-TIME MULTI-CAMERA SLAM SYSTEM

（14）路径规划
• Flying on point clouds: Online trajectory generation and autonomous navigation for quadrotors in cluttered environments
• 2018-10-08-Review of Wheeled Mobile Robots’ Navigation Problems and Application Prospects in Agriculture 路径规划review
• [ICRA 2018] Online Safe Trajectory Generation for Quadrotors Using Fast Marching Method and Bernstein Basis Polynomial 沈老师连续两篇路径规划
（15）VSLAM几何体系与深度学习的融合
1、与帧间估计，计算VO（端对端）
• 2018-12-30-Epipolar Geometry based Learning of Multi-view Depth and Ego-Motion from Monocular Sequences
• 2018-12-28-Deep Global-Relative Networks for End-to-End 6-DoF Visual ocalization and Odometry端到端VO
• 2018-12-28-SfMLearner++: Learning Monocular Depth & Ego-Motion using Meaningful Geometric Constraints sfmLearner加强版，加入了对极约束，减少参数和模型大小
• 2018-12-18-Self-Improving Visual Odometry magic leap出品，可以线上训练网络，自我提升vo性能，厉害了！
• 2018-12-03-MagicVO: End-to-End Monocular Visual Odometry through Deep Bi-directional Recurrent Convolutional Neural Network端到端VO
• 2018-09-25 GANVO: Unsupervised Deep Monocular Visual Odometry and Depth Estimation with Generative Adversarial Networks
• UnDeepVO: Monocular Visual Odometry through Unsupervised Deep Learning github:https://github.com/drmaj/UnDeepVO
• Learning-based Image Enhancement for Visual Odometry in Challenging HDR Environments
• 2018-11-06-Anytime Stereo Image Depth Estimation on Mobile Devices 代码
• DeepTAM: Deep Tracking and Mapping
Tracking和Mapping两个模块都使用了深度学习的卷积神经网络的方法，都是单单从数据中解决了Tracking和Mapping的问题。

2、与语义SLAM（图像理解）
• 2018-12-31-Semantic Monocular SLAM for Highly Dynamic Environments
• 2018-12-29-A Unified Framework for Mutual Improvement of SLAM and Semantic Segmentation
• 2018-12-28-Simultaneous Localization and Mapping in the Epoch of Semantics: A Survey
语义SLAM综述
• 2018-12-28-MID-Fusion: Octree-based Object-Level Multi-Instance Dynamic SLAM分割出动态物体 ，Andrew Davison
• 2018-11-06-Semantic Mapping with Simultaneous Object Detection and Localization
• 2018-10-17-Semantic-only Visual Odometry based on dense class-level segmentation
• 2018-10-09-DS-SLAM: A Semantic Visual SLAM towards Dynamic Environments
• 2018-09-25 A Variational Observation Model of 3D Object for Probabilistic Semantic SLAM
• 2018-10-08-Efficient Constellation-Based Map-Merging for Semantic SLAM
• 2018-09-26-An Orientation Factor for Object-Oriented SLAM
• 2018-10-09-Real-Time Monocular Object-Model Aware Sparse SLAM Ian Reid
确定物体的方向
• 2018-12-05-SLAM method: reconstruction and modeling of environment with moving objects using an RGBD camera去除动态物体
• 2019-02-15-Semantic and 3D Understanding of a Scene for Robot Perception 硕士论文
http://search.proquest.com/openview/3a5804693015d4d0b42f1e4089a02267/1?pq-origsite=gscholar&cbl=18750&diss=y
• 2019-02-14-VUNet: Dynamic Scene View Synthesis for Traversability Estimation using an RGB Camera估计机器人未来可通行区域
• 2018-09-14-VSO: Visual Semantic Odometry

3、与闭环检测
• 2018-12-28-Learning to Fuse Multiscale Features for Visual Place Recognition 回环检测
• 2018-12-03-Loop Closure Detection with RGB-D Feature Pyramid Siamese Networks

4、深度估计（第四章）
多任务CNN生成单目深度图与并分割，重建出有语义信息的地图

• 2019-01-25-Towards Building the Semantic Map from a Monocular Camera with a Multi-task Network
• 2018-12-11-Inferring Point Clouds from Single Monocular Images by Depth Intermediation
• 2018-10-20-Reactive Obstacle Avoidance of Monocular Quadrotors with Online Adapted Depth Prediction Network线上自适应CNN估计深度
• 2018-12-18-Unsupervised Learning of Monocular Depth Estimation with Bundle Adjustment, Super-Resolution and Clip Loss CMU Michael Kaess高清深度图恢复
• 2018-10-06-CNN-SVO: Improving the Mapping in Semi-Direct Visual OdometryUsing Single-Image Depth Prediction 用CNN 减少深度估计误差 代码在此
• 2018-10-08-Semi-dense Stereo Matching using Dual CNNs深度估计
• Real-time Dense Monocular SLAM with Online Adapted Depth Prediction Network
• 2018-07-25 Deep Virtual Stereo Odometry: Leveraging Deep Depth Prediction for Monocular Direct Sparse Odometry.TUM DL+DSO大作

5、深度学习设计特征点
• 2019-01-22-DF-SLAM: A Deep-Learning Enhanced Visual SLAM System based on Deep Local Features
In our DF-SLAM system, learned local feature descriptors are introduced to replace ORB, SIFT and other hand-made features.
• 2018-12-18-From Coarse to Fine: Robust Hierarchical Localization at Large Scale
HF-Net, a hierarchical localization approach based on a monolithic CNN that simultaneously predicts local features and global descriptors foraccurate 6-DoF localization 有代码
• 2018-11-22-A FAST-BRISK Feature Detector with Depth Information 融合深度的特征点提取方法
• 2018-12-06-Matching Features without Descriptors: Implicitly Matched Interest Points (IMIPs)Davide Scaramuzza Michael Bloesch 发表不需要描述子使用CNN就能匹配特征点方法
• Graph-Based Place Recognition in Image Sequences with CNN Features

6、重定位（有用）
• 2018-11-06-Real-Time RGB-D Camera Pose Estimation in Novel Scenes using a Relocalisation Cascade随机森林做重定位
• 2018-10-19-Scene Coordinate Regression with Angle-Based Reprojection Loss for Camera Relocalization同下，合在一起读，可以加强理解
• 2018-10-19-DSAC - Differentiable RANSAC for Camera Localization 深度学习重定位，代码
• 2018-10-19-PoseNet: A Convolutional Network for Real-Time 6-DOF Camera Relocalization
深度学习重定位，代码

7、BA
• 2018-12-10-BA-Net: Dense Bundle Adjustment Network
• 2018-10-15-Learning to Solve Nonlinear Least Squares for Dense Tracking and Mapping
• Andrew J. Davison 的 learned optimizer

8、点云分割
• Fusion++: Volumetric Object-Level SLAM
• 2018-10-08-Deep Learning Based Semantic Labelling of 3D Point Cloudin Visual SLAM SLAM 加上点云标签