CGABC

Observability and Inconsistency in a Nutshell

[TOC] Overview What is observability ? In control theory, observability is a measure for how well internal states of a system can be inferred by knowledge of its external outputs. What is consistenc

2022-07-16

State Estimation

#State Estimation #Observability and Consistency #FEJ #Lie Theory #Observability Matrix #OC-VINS #Degeneracy

基于RGBD相机的多模态结构特征融合定位 (Draft)

[TOC] 概述目前基于RGBD相机的视觉SLAM系统，都是基于特征点或者稠密点云进行定位，计算量大，而且在无纹理或弱纹理的环境，容易跟丢，鲁棒性较差。本文提出了一种基于RGBD相机的多模态结构化数据融合定位方法，算法主要框架如下图所示，通过充分利用视觉特征，融合二维特征点、二维曲线（边缘点）、二维线段和三维平面等，结合状态估计算法进行实时、鲁邦、准确的位姿估计。在二维特征点的基础上，同时提取边

2022-03-06

MSF

#Image Features #Visual SLAM #Multi-Sensor Fusion #Parameterization #Struct SLAM

From MAP, MLE, OLS, G-N to IEKF, EKF

[TOC] Overview 本文主要包括以下几个方面：将线性高斯系统的状态空间方程(运动和观测方程，即多个似然因子)组合为一个新的观测方程，其实就是将多个约束项放在一起推导了MAP到MLE，再到OLS的历程以上面系统的状态估计为出发点，推导了最小二乘、高斯牛顿、IEKF和EKF的区别与联系，将优化与滤波联系在一起，导出了高斯牛顿海塞矩阵H与EKF协方差矩阵P的关系，证明了IEKF与高斯牛

2021-12-30

State Estimation

#EKF #State Estimation #MAP

Marginalization and FEJ in Sliding Window VIO

[TOC] Overview 滑动窗口与共视图我们根据时空对局部BA 进行分类：时间：滑动窗口空间：共视图通过企业管理，解释滑窗的边缘化以及共视图为啥没有边缘化：基于滑窗优化的边缘化，可以用公司小组成员离职类比，小组相当于滑窗，离职交接相当于保留共视信息，没有交接就回影响小组未来的发展；交接不好，重新搞；员工贡献不大直街裁掉，就是remove或throw；而基于共视图的，相

2021-07-04

SLAM

#Visual SLAM #VIO #Sliding Window #Observability and Consistency #Marginalization #FEJ

Bundle Adjustment (BA) in vSLAM or SFM

[TOC] Overview Bundle adjustment (wikipedia) BA is a key ingredient of Structure and Motion Estimation (SaM), almost always used as its last step It is an optimization problem over the 3D structur

2021-02-04

State Estimation

#Graph #Visual SLAM #SFM #Optimization #State Estimation #Bundle Adjustment #OLS

Linear Algebra in a Nutshell

Matrix 厄米特矩阵埃尔米特矩阵等于自己的共轭转置厄米特矩阵 \(A^H = A\) 实对称矩阵 \(A^T = A\) 反厄米特矩阵 \(A^H = -A\) 实反对称矩阵 \(A^T = -A\) 正规矩阵 \[ A^H A = A A^H \] 酉矩阵 \(A^H A = A A^H = I\) 正交矩阵 \(A^T A = AA^T = I\) 特殊正交矩阵：\

2019-01-28

Maths

#Matrix Decomposition

双目立体视觉原理

[TOC] Overview 双目立体视觉的整体流程包括：图像采集双目标定双目矫正立体匹配三维重建 1. 图像采集双目相机采集左右目图像 2. 双目标定通过双目标定工具对双目相机进行标定，得到如下结果参数：内参外参相机矩阵 \(K_1, K_2\) 旋转矩阵 \(R\) 畸变系数 \(D_1, D_2\) 平移向量 \(t\) 《Lear

2018-07-22

Computer Vision

#Computer Vision #Stereo Vision

Spatial AI Benchmark

Overview https://github.com/youngguncho/awesome-slam-datasets KITTI Vision Benchmark Suite Lifelong Robotic Vision SLAM Dataset Sample Params TUM VI TartanAir ADVIO NEAR-VI A

2024-06-09

Bechmark

#Computer Vision #Deep Learning #Bechmark #SLAM

When SLAM meets XR and Robotics

When SLAM meets XR When SLAM meets Robotics

2022-12-24

SLAM

GPG Overview

Overview PGP目前支持的算法非对称算法: RSA, ELG, DSA, ECDH, ECDSA, EDDSA 对称算法: IDEA, 3DES, CAST5, BLOWFISH, AES, AES192, AES256, TWOFISH, CAMELLIA128, CAMELLIA192, CAMELLIA256 哈希算法: SHA1, RIPEMD160, SHA256, SHA3

2022-12-12

DevOps

#GPG

移动机器人常见底盘及其运动学

Overview [1] 通常移动机器人依赖电机驱动车轮实现行走功能。机器人底盘结构不同，其运动学也完全不同。根据不同类型车轮，常见的底盘结构差速运动模型、滑移运动模型、阿克曼运动模型、全向轮运动模型等等。下图中，（a）双轮差速式机器人，（b）阿克曼式机器人，（c）四轮驱动机器人，（d）双履带式机器人，（e）麦克纳姆轮全向机器人，（f）全向轮全向机器人，（g）四轮驱动四轮转向机器人 [2]

2022-11-22

Robotics

#Kinematics #RobotBase

On-Manifold Optimization: Local Parameterization

Overview Manifold Space vs Tangent Space Jacobian w.r.t Error State Jacobian w.r.t Error State vs True State According [1] 2.4, The idea is that for a \(x \in N\) the function \(g(\delta) := f (x

2022-09-04

State Estimation

#Optimization #Parameterization #Kinematics #State Estimation #Manifold #Lie Theory #Quarternion

Mapillary API at a Glance

Embed images [1] The iframe tag below would show a 640x480px image with ID 550092599700936. 123456<iframe src="https://www.mapillary.com/embed?image_key=550092599700936&style=classic" heigh

2022-07-24

GIS

#GIS #Mapillary

PyTorch on Ubuntu 18.04

[TOC] Install PyTorch locally https://pytorch.org/get-started/locally/ Conda (Python) 12345678910conda create -n torch python=3.9conda activate torch# select oneconda install pytorch -c soumith c

2022-06-25

#Machine Learning #Deep Learning #PyTorch

Yolo Test on Ubuntu 18.04

[TOC] YOLO Darknet: Open Source Neural Networks in C YOLO: Real-Time Object Detection Build / Install Detection Using A Pre-Trained Model 1./darknet detect cfg/yolov3.cfg yolov3.weights data/dog.jpg

2022-06-25

#Machine Learning #Deep Learning #Object Detection #Yolo

Image Exposure Controller

msckf-vio Issue: About image exposure controller #41 [1] The auto exposure code is not open source because for one the actual implementation is ugly, and second because it ties into other software

2022-06-22

Computer Vision

#Computer Vision #Camera #Camera Control #Driver

A-LOAM

[TOC] Overview A-LOAM is an Advanced implementation of LOAM (J. Zhang and S. Singh. LOAM: Lidar Odometry and Mapping in Real-time), which uses Eigen and Ceres Solver to simplify code structure. Cod

2022-06-19

SLAM

#LiDAR SLAM #LIO

OpenMVG + OpenMVS build & run

[TOC] Overview OpenMVG provides an end-to-end 3D reconstruction from images framework compounded of libraries, binaries, and pipelines. a library mainly focused on Multiple-View-Geometry and Structu

2022-06-18

Mapping

#3D Reconstruction #Mapping #SFM

Mesh Texturing in a Nutshell (Let There Be Color)

[TOC] .graphviz { display: flex; justify-content: center; } Overview digraph { TV [label="TextureView"]; TVs [label="N TextureViews"]; Texturing [style=filled, shape=box]; ColorImg->TV;

2022-06-12

Mapping

#Graph #3D Reconstruction #Mapping #Texturing #MAP-MRF #Dynamic Programming #Belief Propagation

Kinect Fusion

[TOC] Overview Kinect Fusion 描述三维空间的方式叫 Volumetric。它把固定大小的一个空间（比如3𝑚×3𝑚×3𝑚）均匀分割成一个个小方块（比如512×512×512），每个小方块就是一个voxel，存储TSDF值以及权重。最终得到的三维重建就是对这些voxel进行线性插值。重建流程如上图所示： Depth Map Conversion: 读入的深度

2022-06-10

Mapping

#3D Reconstruction #Visual SLAM #Mapping #TSDF