Modified version of rpg_svo(commit d616106 on May 9, 2017) which implement a Semi-direct Monocular Visual Odometry pipeline
Modified version of rpg_svo(commit d616106 on May 9, 2017) which implement a Semi-direct Monocular Visual Odometry pipeline
[TOC]
Boost - C++ Librairies (thread and system are needed)
Eigen 3 - Linear algebra
OpenCV - Computer vision library for loading and displaying images
Sophus - Lie groups
Fast - Corner Detector
g2o - General Graph Optimization OPTIONAL
export G2O_ROOT=$HOME/installdir
rpg_vikit (git submodule)
git clone https://github.com/cggos/svo_cg.git --recursive
# or
git clone https://github.com/cggos/svo_cg.git
git submodule init
git submodule update
catkin_make
or catkin build
build vikit_common
cd 3rdParty/rpg_vikit/vikit_common
gedit CMakeLists.txt # SET(USE_ROS FALSE)
mkdir build & cd build
cmake .. & make -j3
sudo make install
build svo
mkdir build & cd build
cmake -DUSE_ROS=FALSE .. & make -j3
roslaunch svo_ros test_rig3.launch [rviz:=true]
rosbag play airground_rig_s3_2013-03-18_21-38-48.bag
# 1. calibrate your camera and modify `svo_ros/param/camera_atan_zr300.yaml`
# 2. run
roslaunch svo_ros realsense_zr300.launch [rviz:=true]
export SVO_DATASET_DIR=${HOME}/Datasets # .bashrc
source ~/.bashrc
cd ${SVO_DATASET_DIR}
[tsocks] wget http://rpg.ifi.uzh.ch/datasets/sin2_tex2_h1_v8_d.tar.gz -O - | tar -xz
cd svo
mkdir build & cd build
cmake .. & make
./test_pipeline
use the python scripts in scripts
directory
p_f = T_f_w * p_w
pos = T_f_w.inverse().translation()
In order to obtain the same performance as shown in the videos, consider the following:
@inproceedings{Forster2014ICRA,
author = {Forster, Christian and Pizzoli, Matia and Scaramuzza, Davide},
title = {{SVO}: Fast Semi-Direct Monocular Visual Odometry},
booktitle = {IEEE International Conference on Robotics and Automation (ICRA)},
year = {2014}
}
SVO 2.0: Fast Semi-Direct Visual Odometry for Monocular, Wide Angle, and Multi-camera Systems.
SVO 2.0 (IEEE TRO'17) extends the original SVO implementation (ICRA' 14) with the addition of edgletes, IMU prior, wide angle cameras (fisheye and catadioptric), multi-camera configurations, and forward looking camera motion.