ubuntu16.04+ROS kinetic下跑LSDSLAM

SLAM

字数统计: 1.6k阅读时长: 7 min

 2018/09/30 

LSD-SLAM是一个semi-dense的直接法SLAM算法。其主页见https://vision.in.tum.de/research/vslam/lsdslam?redirect=1 其代码见https://github.com/tum-vision/lsd_slam

官方在github上给出默认环境有两个 * ROS fuerte + Ubuntu 12.04 * ROS indigo + Ubuntu 14.04

因此推荐使用上面的环境，出错较少。下面是我在ROS kinetic + Ubuntu 16.04环境下实现LSD-SLAM的过程

1.安装ROS kinetic

这个地方一般不会有问题，请参考ROS wiki或者一些博客，如博客,根据博客安装、初始化并测试ROS。如果出错，请换个源或者VPN试试。

2.编译LSD-SLAM

1.创建rosbuild工作空间

# python默认版本应为2.7,或者下载python３-rosinstall
sudo apt-get install python-rosinstall

# 具体路径请自己更改
mkdir ~/rosbuild_ws
cd ~/rosbuild_ws
rosws init . /opt/ros/kinetic
mkdir package_dir
rosws set ~/rosbuild_ws/package_dir -t .
echo "source ~/rosbuild_ws/setup.bash" >> ~/.bashrc
bash

2.安装依赖

1
2

# 注意LSD-SLAM依赖qt4而不是qt5，请自行安装正确版本libqglviewer-dev-qt4
sudo apt-get install ros-kinetic-libg2o ros-kinetic-cv-bridge liblapack-dev libblas-dev freeglut3-dev libsuitesparse-dev libx11-dev

3.获取LSD-SLAM

1 2	cd package_dir git clone https://github.com/tum-vision/lsd_slam.git lsd_slam

4.编译

1	rosmake lsd_slam

编译过程中不会一帆风顺，可能出现许多错。我出现的错如下：

error 1

# 报错信息
ERROR [gendeps] 1 Finding dependencies for /home/gjh/code/rosbuild_ws/package_dir/lsd_slam/lsd_slam_viewer/cfg/LSDSLAMViewerParams.cfg
......
Exception: quotes not allowed in description string `log10 of threshold on point's variance, in the respective keyframe's scale.

#　解决方法[1]
删除lsd_slam_viewer/cfg/LSDSLAMViewerParams.cfg和lsd_slam_core/cfg/LSDDebugParams.cfg两个文件中出现的单引号

error 2

# 报错信息
main_stitchVideos.cpp:(.text+0xd65c)：对‘cv::imshow(cv::String const&, cv::_InputArray const&)’未定义的引用
main_stitchVideos.cpp:(.text+0xd668)：对‘cv::String::deallocate()’未定义的引用
main_stitchVideos.cpp:(.text+0xd672)：对‘cv::waitKey(int)’未定义的引用
main_stitchVideos.cpp:(.text+0xd715)：对‘cv::fastFree(void*)’未定义的引用
main_stitchVideos.cpp:(.text+0xd7b5)：对‘cv::fastFree(void*)’未定义的引用
main_stitchVideos.cpp:(.text+0xd84d)：对‘cv::fastFree(void*)’未定义的引用
main_stitchVideos.cpp:(.text+0xd8e5)：对‘cv::fastFree(void*)’未定义的引用
main_stitchVideos.cpp:(.text+0xd985)：对‘cv::fastFree(void*)’未定义的引用
CMakeFiles/videoStitch.dir/src/main_stitchVideos.cpp.o:main_stitchVideos.cpp:(.text+0xda25): 跟着更多未定义的参考到 cv::fastFree(void*)
CMakeFiles/videoStitch.dir/src/main_stitchVideos.cpp.o：在函数‘inlayVid(int, char**)’中：
main_stitchVideos.cpp:(.text+0xdeb4)：对‘cv::String::allocate(unsigned long)’未定义的引用
main_stitchVideos.cpp:(.text+0xdedf)：对‘cv::Mat::deallocate()’未定义的引用
main_stitchVideos.cpp:(.text+0xdeec)：对‘cv::Mat::deallocate()’未定义的引用
main_stitchVideos.cpp:(.text+0xdefd)：对‘cv::Mat::deallocate()’未定义的引用
main_stitchVideos.cpp:(.text+0xdf0e)：对‘cv::Mat::deallocate()’未定义的引用
main_stitchVideos.cpp:(.text+0xdf1f)：对‘cv::Mat::deallocate()’未定义的引用
CMakeFiles/videoStitch.dir/src/main_stitchVideos.cpp.o:main_stitchVideos.cpp:(.text+0xdf30): 跟着更多未定义的参考到 cv::Mat::deallocate()
CMakeFiles/videoStitch.dir/src/main_stitchVideos.cpp.o：在函数‘inlayVid(int, char**)’中：
main_stitchVideos.cpp:(.text+0xdfb7)：对‘cv::String::allocate(unsigned long)’未定义的引用
main_stitchVideos.cpp:(.text+0xe212)：对‘cv::String::deallocate()’未定义的引用
main_stitchVideos.cpp:(.text+0xe23e)：对‘cv::String::deallocate()’未定义的引用
main_stitchVideos.cpp:(.text+0xe26e)：对‘cv::String::deallocate()’未定义的引用
main_stitchVideos.cpp:(.text+0xe282)：对‘cv::String::deallocate()’未定义的引用
main_stitchVideos.cpp:(.text+0xe2a6)：对‘cv::String::deallocate()’未定义的引用

#　解决方法
对lsd_slam_viewer中的CMAKELISTS最后增加与OpenCV的链接：
find_package(OpenCV REQUIRED)
rosbuild_add_executable(videoStitch src/main_stitchVideos.cpp)
target_link_libraries(videoStitch ${OpenCV_LIBS})

error ３

# 报错信息
void getPosition(qreal& x, qreal& y, qreal& z) const;
     ^

#　解决方法[1]
lsd_slam_viewer/src/PointCloudViewer.h中，第135行：将float x, y, z修改为qreal x, y, z　　　　　　　　　　　
lsd_slam_viewer/src/PointCloudViewer.cpp中，第326行：将float x, y, z修改为qreal x, y, z

error 4

# 报错信息
enum Status
     ^

#　解决方法[1]
将文件lsd_slam_viewer/src/KeyFrameDisplay.cpp的头文件：
# include "opencv2/opencv.hpp"
注释掉

error 5

# 报错信息
/usr/bin/ld: CMakeFiles/viewer.dir/src/main_viewer.cpp.o: undefined reference to symbol '_ZN5boost6system15system_categoryEv'
/usr/lib/x86_64-linux-gnu/libboost_system.so: error adding symbols: DSO missing from command line
collect2: error: ld returned 1 exit status

#　解决方法
对lsd_slam_viewer中的CMAKELISTS最后增加boost_system库的链接
target_link_libraries(viewer ${QGLViewer_LIBRARIES} ${QT_LIBRARIES} ${QGLVIEWER_LIBRARY} GL glut GLU boost_system)

error ６

# 报错信息
/usr/bin/ld: CMakeFiles/live_slam.dir/src/main_live_odometry.cpp.o: undefined reference to symbol '_ZN5boost6system15system_categoryEv'
/usr/lib/x86_64-linux-gnu/libboost_system.so: error adding symbols: DSO missing from command line
collect2: error: ld returned 1 exit status
......
/usr/bin/ld: CMakeFiles/dataset_slam.dir/src/main_on_images.cpp.o: undefined reference to symbol '_ZN5boost6system15system_categoryEv'
/usr/lib/x86_64-linux-gnu/libboost_system.so: error adding symbols: DSO missing from command line
collect2: error: ld returned 1 exit status

#　解决方法与上一个类似
对lsd_slam_core中的CMAKELISTS最后增加boost_system库的链接
# build live ros node
rosbuild_add_executable(live_slam src/main_live_odometry.cpp)
target_link_libraries(live_slam lsdslam boost_system)
# build image node
rosbuild_add_executable(dataset_slam src/main_on_images.cpp)
target_link_libraries(dataset_slam lsdslam boost_system)

error 7

# 报错信息
../lib/liblsdslam.so：对‘cv::waitKey(int)’未定义的引用
../lib/liblsdslam.so：对‘cv::namedWindow(cv::String const&, int)’未定义的引用
../lib/liblsdslam.so：对‘cv::destroyAllWindows()’未定义的引用
../lib/liblsdslam.so：对‘cv::resizeWindow(cv::String const&, int, int)’未定义的引用
../lib/liblsdslam.so：对‘cv::imshow(cv::String const&, cv::_InputArray const&)’未定义的引用

#　解决方法与error２类似
find_package(OpenCV REQUIRED)
# build live ros node
rosbuild_add_executable(live_slam src/main_live_odometry.cpp)
target_link_libraries(live_slam lsdslam boost_system ${OpenCV_LIBS})
# build image node
rosbuild_add_executable(dataset_slam src/main_on_images.cpp)
target_link_libraries(dataset_slam lsdslam boost_system ${OpenCV_LIBS})

error 8

# 报错信息
DebugWindow DEPTH [Not Responding]

#　解决方法[2]
可以把lsd_slam_core/src/util/settings.cpp中改成
bool displayDepthMap = false;
但是只能正常显示点云地图，无法显示DebugWindow了

另一种方法是重新编译OpenCV，不带QT
cmake -D WITH_QT=off ..

如果还有其他问题，也大可以在这两个链接中找到解决方法：

1 Iridescent18的博客

2 github/issues issues中的问题比较全，基本可以解决遇到的问题

需要注意的是lsd-slam分两个部分编译，先编译lsd_slam_viewer，再编译lsd_slam_core。我们必须对错误的地方定位准确，当出现下面的信息(两部分都pass)，则编译通过：

[rosmake-2] Finished <<< lsd_slam_viewer [PASS] [ 4.35 seconds ] [ 4 warnings  ]                                                                     
[rosmake-2] Starting >>> lsd_slam_core [ make ]                                                                                                      
[rosmake-2] Finished <<< lsd_slam_core [PASS] [ 17.97 seconds ]                                                                                      
[ rosmake ] Results:                                                                                                                                 
[ rosmake ] Built 47 packages with 0 failures.                                                                                                       
[ rosmake ] Summary output to directory

最后，我将自己修改后的代码托管到github上了，使用方法没变： https://github.com/gjgjh/lsd-slam

3.运行示例程序

下载并解压缩数据示例数据

#分别打开三个终端运行下面四条命令
roscore # start the ROS master

rosrun lsd_slam_viewer viewer # start a node, usage: rosrun package node

rosrun lsd_slam_core live_slam image:=/image_raw camera_info:=/camera_info

#选择正确的数据路径
rosbag play ~/LSD_room.bag # playback a bag, usage: rosbag play/record bag

4.理论和代码的学习

相关的有三篇论文 2013 Semi-dense Visual Odometry for a Monocular Camera 2014 Semi-dense visual odometry for AR on a smartphone 2014 LSD-SLAM: Large-Scale Direct Monocular SLAM

另外推荐hitcm的博客,其中有比较详细的介绍

个人理解错误的地方还请不吝赐教，转载请标明出处

原文作者：GJGJH

原文链接：https://gjgjh.github.io/lsdslam.html

发表日期：September 30th 2018, 9:42:13 pm

更新日期：February 2nd 2022, 7:37:38 pm

Next Post

LSDSLAM算法解析

CATALOG

1. 1.安装ROS kinetic
2. 2.编译LSD-SLAM
3. 3.运行示例程序
4. 4.理论和代码的学习