Sensor Lab

中文|Chinese

A repository of implementations and notes that provide services for the math, computer vision, AI, robotics, autonomous driving related methods I have studied

just for fun

🚧

• Getting Started
  • Prerequisites
• Modules Introduction
  • cpp_test
  • Algorithm
  • Feature Detect
  • Pose Estimation
  • Kalman Filter
  • imu_and_gnss
  • Lidar_2d
  • Lidar_3d
  • pcl_test
  • cere_test
  • g2o_test
  • gtsam_test
  • Path Plan
  • Path Tracking
• Usage

Getting Started

Prerequisites

• Ubuntu 20.04 (virtual machine); ROS noetic
• add below ros messages library and don't forget source source /opt/ros/noetic/setup.bash

sudo apt install ros-noetic-cv-bridge ros-noetic-image-transport ros-noetic-roscpp ros-noetic-rospy ros-noetic-rosbag ros-noetic-std-msgs ros-noetic-sensor-msgs ros-noetic-pcl-ros ros-noetic-pcl-conversions ros-noetic-geometry-msgs

• C++14; OpenCV 4.2.0; Eigen 3.3.7; PCL 1.10.0
• Sophus, commit SHA a621ff local build, support SO(3) and SE(3) operator
• glog(release version); gflags; gtest build gtest and sudo cp libgtest*.a /usr/local/lib
• Ceres v2.2.0, deal with complicated non-linear optimization problems like Bundle Adjustment
• G2O checkout ff647bd, like Ceres
• DBoW3, images loop closure
• gtsam 4.2.0, optimize library
• Pangolin branch:v0.9 installed but rarely used
• octomap a 3D occupancy grid mapping approach
• Python 3.8.10. I use the features of matplotlibcpp and run some python scripts
• yaml-cpp use .yaml

Modules Introduction

Each module describes the methods it contains

attention: some test data comes from the network public dataset

cpp_test

including some cpp feature test

Algorithm

this module summarizes some methods for tradition image process and curve fit

• dlt camera calibration
• image undistortion
• based on Tsai-Lenz of hand-eye calibration (eye in hand)
• least square polynomial curve fit and move least square
• cubic spline interpolation curve fit
• thin plate spline
• stereo match and 3D point cloud reconstruction

stereo match and 3D reconstruction

Feature Detect

• based orb operator match

• LK option flow apply

pre image

current image

LK option flow result

Pose Estimation

• sfm: epipolar constrainc and triansgulation
• 3D points ICP
• gaussian-newton estimate camera pose

Kalman Filter

• direct kalman filter point estimate
• extend kalman filter test

imu_and_gnss

• static imu initialization
• tradition imu integration
• imu pre-integration
• gnss data convert to utm
• eskf gins (imu data predict and gnss data update)
• eskf gins with imu pre-integration optimize

in the image below, the left shows the ESKF and the right side uses IMU pre-integration, trajectory interruptions due to poor GNSS signals are reduced

Lidar_2d

• 2d lidar scans ICP (point to point & point to plane distance)
• scan to occupancy grid map
• lidar 2d mapping global map

Lidar_3d

• point cloud ICP registration
• point cloud NDT registration
• direct NDT lidar odom
• incremental NDT LO
• loosely coupled LIO

incremental NDT loosely coupled LIO mapping and without loop closure

pcl_test

• point cloud linear fit
• point cloud data convert
• nearest neighbor search

cere_test

the below image show the easy use of the ceres optimization library

• left image was source BAL data
• right image use ceres for BA optimize,reduce some noise(running in virtual machine, low performance)

g2o_test

g2o optimize library test

gtsam_test

gtsam optimize library test

Path Plan

this module conclude some common tips and knowledge in global or local path plan.

Dijkstra global path plan

A star global path plan

DWA local path plan

RRT* sample based local path plan

Bezier Curve local path plan and B spline curve local path plan and smooth

Bezier Curve

B spline curve

Dubins Curve path plan

dubins curve LSR

dubins curve LSL

Path Tracking

PID controller

LQR controller

pure pursuit controller

stanley controller

Usage

mkdir build && cd build

# enable debug
cmake ..
make -j8