Trytry Visual Odometry

A project about Visual Odometry, which uses ICP, PNP and BA to solve the rotation matrix R and vector T.

Build && Run

Operation System: Ubuntu 18.04

ROS: Melodic

Required

g2o

git clone https://github.com/RainerKuemmerle/g2o.git
mkdir build
cd build
cmake ..
make
sudo make install

Sophus

git clone https://github.com/strasdat/Sophus.git
mkdir build
cd build
cmake ..
make
sudo make install

Glog

Building from Source

Eigen3

sudo apt-get install libeigen3-dev

Opencv

Get the latest version here Github.

The project used Opencv that comes with ROS.

Build

This is a ROS packages.

mkdir -p ~/catkin_ws/src
cd ~/catkin_ws/src
git clone https://github.com/heiheishuang/trytry_vo.git
cd ..
catkin_make
source devel/setup.bash # Change to `source devel/setup.zsh` if you use zsh.

Run

roslaunch trytry_vo viewer.launch

Parameters

Dataset

data_set: The folder where the dataset is stored

ground_truth_path:

Camera

camera calibration and distortion parameters

Frontend

estimate_pose: Estimation pose using BA PnP SVD or RANSAC

ransac_max_iterators: Set maximum number of iterations

number_optimize_frame: Using g2o to optimize frame poses. Set number of frames

System

Initial pose required for pose initialization

write_file: T: If you want to write odometer data in a file; F: If you don't want to write a file

file_name: The name of file to write

Result (Green: ours; Red: ground_truth)

Using SVD

Using SVD and Ransac

Using BA (frame-to-frame PnP)

Using PnP and Ransac

Using PnP and Optimize 5 frames

Using PnP and Optimize 30 frames

Tranks

https://github.com/LittleDang/mySlam

https://github.com/gaoxiang12/slambook2