Developer: Wenyu Li, Zijun Wang
Email: uestc_wenyu@foxmail.com
We modified the kitti publisher code from A LOAM. Then a python depth completion algorithm, IP Basic, is integrated in real time with a SLAM application, ORB SLAM2.
A depth completion module helps visual SLAM work with RGB-D mode in outdoor scenes. It detects a longer and more accurate distance compared with stereo or RGB-D cameras.
This is part of my master thesis research SLAMMOT. But this repository is just an engineering incremental work, so feel free to use it.
Having test it on Ubuntu 18.04 and ROS Melodic.
Please install Eigen3,Opencv,PCL.
Default version provided in ROS in enough.
sudo apt install libeigen-dev
sudo apt install libopencv-dev
sudo apt install libpcl-dev
We use kitti odometry dataset for testing. If you want to use another dataset, you need to extract the depth completion code in C++ and integrate into your own program.
For kitti odometry dataset, download and unzip them.
Organize the kitti odometry data set like:
├── dataset
│ ├── poses
│ │ ├── 00.txt
│ │ ├── 01.txt
│ │ ├── xx.txt
│ └── sequences
│ ├── 00
│ │ ├── calib.txt
│ │ ├── image_2
│ │ ├── image_3
│ │ ├── times.txt
│ │ └── velodyne
│ ├── 01
│ │ ├── calib.txt
│ │ ├── image_2
│ │ ├── image_3
│ │ ├── times.txt
│ │ └── velodyne
│ ├── xx
open and edit line 3-5 in : orb_slam2_ros/ros/launch/orb_slam2_kitti_image_lidar.launch
folder refers to the dataset folder.
save path refers to the trajectory save path when it is finished.
open and edit line 207 in : slammot/src/kittiHelper.cpp
PyRun_SimpleString("sys.path.append('/home/wenyu/catkin_ws/src/slammot/src')");
change the following path where the depth_map_utils.py file actually is.
I guess there is an elegant way for using python script in C++ program, but I haven't got an idea.
Please follow the ORB SLAM2 ROS repositiory.
ORB SLAM2 ROS requires to install other ROS packages by rosdep.
sudo rosdep init
rosdep update
rosdep install --from-paths src --ignore-src -r -y
If you failed or get an error when update rosdep, you might install fishros and use rosdepc instead of rosdep.
It is build on catkin tools, install catkin tools and run:
sudo apt install python-catkin-tools
If you build ros workspace using catkin_make, you may get errors. Just remove folder build and devel in your current workspace. Then run the following code:
catkin init
catkin build
And start these ros nodes by a launch file:
roslaunch orb_slam2_ros orb_slam2_kitti_image_lidar.launch
you can see results in rviz like this:
