0 講座
(1)SLAM定義
對比雷達傳感器和視覺傳感器的優缺點(主要介紹視覺SLAM)

單目:不知道尺度信息
雙目:知道尺度信息,但測量范圍根據預定的基線相關
RGBD:知道深度信息,但是深度信息對距離也有要求
vSLAM(視覺SLAM)
攝像機(主要)+IMU+超聲波避障傳感器
2016年之后已經可以跑一點DEMO程序了(在剛體的和靜態的環境下)
視覺SLAM的幾個模塊

傳感器數據(圖像數據采集點雲)
視覺里程計(估計攝像機參數)
后端(對攝像機參數優化,因為攝像機誤差累積跟IMU一樣,因此需要對其優化)
建圖(二維圖和三維圖)
回環檢測(若能回來)
2. 視覺里程計

1. 提取特征點:FAST, SIFT, SURF, ORB
2. 匹配:KNN
3. 估計ego-運動
缺點:
1. 特征點必須存在,不能是一堵白牆或者運動太快晃得太厲害,SLAM是有這些假設條件的
2. 解不一定是最優解
ego運動估計

假設相機矩陣已知求解運動的R(旋轉)、t()、X(平移)
問題:
1. 不一定有最優解 因為方程非凸
求解最優解的方法

代數法(解析法)
迭代法(數值法)
單目三維點重建

尺度問題
假設兩幅圖之間移動了一個距離1,而如果不移動呢距離就為0了怎么辦

直接法
不提特征點直接求差的平方,這是在慢速運動、平滑運動以及光的條件改變不大的時候,計算量大
slambook
This is the code written for my new book about visual SLAM. It will come in April 2017. I am reviewing the context and codes right now.
It is highy recommended to download the code and run it in you machine because you can learn more and even modify it by doing so. In this repo, the code is stored by chapters like "ch2" and "ch4". Note that chapter 9 is a project so I stored it in the "project" directory.
If you have any questions about the code, please add an issue so I can solve it. Contact me for more information: gaoxiang12 dot mails dot tsinghua dot edu dot cn.
These codes are under MIT license. You don't need permission to use it or change it. But I recommed to cite this book if you are doing academic works.
Citation will be added as soon as the book is officially published.
Please cite: Xiang Gao, Tao Zhang, Qinrui Yan, Yi Liu, 14 Lectures on Visual SLAM: From Theory to Practice, Publishing House of Electronics Industry, 2017.
- ch1 Preface
- ch2 Overview of SLAM & linux, cmake
- ch3 Rigid body motion & Eigen
- ch4 Lie group and Lie Algebra & Sophus
- ch5 Cameras and Images & OpenCV
- ch6 Non-linear optimization & Ceres, g2o
- ch7 Feature based Visual Odometry
- ch8 Direct (Intensity based) Visual Odometry
- ch9 Project
- ch10 Back end optimization & Ceres, g2o
- ch11 Pose graph and Factor graph & g2o, gtsam
- ch12 Loop closure & DBoW3
- ch13 Dense reconstruction & REMODE, Octomap
