高翔博士slambook2 ch9 编译运行笔记

一、主要内容

ch9讲如何把多个相机位姿和大量三维地图点放在一起优化，使整体重投影误差最小。Bundle Adjustment，简称BA，光束法平差 / 捆集调整，是视觉SLAM和SFM中非常重要的后端优化方法。

在视觉SLAM中，通常有多个相机位姿、多个三维地图点、很多图像观测，这些数据一开始都是有误差的，因此用BA来同时调整相机位姿和三维地图点，使它们投影到图像上的位置尽量接近真实观测位置。

min Σ || u_observed - u_projected ||²

BA 是一个很大的非线性最小二乘问题。

二、安装依赖

下载 Ceres 2.1.0 + Eigen 3.3.7

最重要的放在最前面，我在编译ch9代码时掉坑里了，一直报错Eigen的版本号对不上，要求版本3.3.7，结果读出来是空值。

-- Found required Ceres dependency: absl version 20240116 in /usr/local/lib/cmake/absl CMake Error at CMakeLists.txt:11 (Find_Package): Found package configuration file: /usr/local/lib/cmake/Ceres/CeresConfig.cmake but it set Ceres_FOUND to FALSE so package "Ceres" is considered to be NOT FOUND. Reason given by package: Found Eigen dependency, but the version of Eigen found () does not exactly match the version of Eigen Ceres was compiled with (3.3.7). This can cause subtle bugs by triggering violations of the One Definition Rule. See the Wikipedia article http://en.wikipedia.org/wiki/One_Definition_Rule for more details

于是我apt重装了Eigen，fail。

于是我源码重装了Eigen，指定版本3.3.7，fail。（这里已经确定了Eigen版本正确，就是3.3.7)

cd ~/Desktop rm -rf eigen git clone https://gitlab.com/libeigen/eigen.git cd eigen git checkout 3.3.7 rm -rf build mkdir build cd build cmake .. \ -DCMAKE_BUILD_TYPE=Release \ -DCMAKE_INSTALL_PREFIX=/usr/local sudo make install #确认eigen版本 grep -R "EIGEN_WORLD_VERSION\|EIGEN_MAJOR_VERSION\|EIGEN_MINOR_VERSION" /usr/local/include/eigen3/Eigen/src/Core/util/Macros.h

报错如下

Found package configuration file: /usr/local/lib/cmake/Ceres/CeresConfig.cmake but it set Ceres_FOUND to FALSE so package "Ceres" is considered to be NOT FOUND. Reason given by package: Found Eigen dependency, but the version of Eigen found () does not exactly match the version of Eigen Ceres was compiled with (3.3.7). This can cause subtle bugs by triggering violations of the One Definition Rule. See the Wikipedia article http://en.wikipedia.org/wiki/One_Definition_Rule for more details -- Configuring incomplete, errors occurred!

找AI看了下，是因为Ceres是之前编译的，和当前的Eigen版本没有绑定，于是我重新编译了2.2.1版本的Ceres。

#安装依赖 sudo apt update sudo apt install -y build-essential cmake git sudo apt install -y libsuitesparse-dev sudo apt install -y libgoogle-glog-dev libgflags-dev cd ~/Desktop rm -rf ceres-solver #下载Ceres源码 + 指定版本2.1.0 git clone https://github.com/ceres-solver/ceres-solver.git cd ceres-solver git checkout 2.1.0 #用eigen3.3.7重新编译 rm -rf build mkdir build cd build cmake .. \ -DCMAKE_BUILD_TYPE=Release \ -DEigen3_DIR=/usr/local/eigen-3.3.7/share/eigen3/cmake \ -DCMAKE_PREFIX_PATH=/usr/local/eigen-3.3.7 \ -DBUILD_TESTING=OFF \ -DBUILD_EXAMPLES=OFF \ -DMINIGLOG=OFF \ -DCMAKE_INSTALL_PREFIX=/usr/local make -j$(nproc) sudo make install sudo ldconfig

三、编译代码

前面的依赖问题解决了以后，就可以开始编译ch9的代码了。

cd ~/Desktop/slambook2/ch9 rm -rf build mkdir build cd build cmake .. \ -DEigen3_DIR=/usr/local/eigen-3.3.7/share/eigen3/cmake \ -DCeres_DIR=/usr/local/lib/cmake/Ceres \ -DSophus_DIR=/usr/local/lib/cmake/Sophus \ -DCMAKE_PREFIX_PATH="/usr/local/eigen-3.3.7;/usr/local" make -j$(nproc)

编译结果如图：

四、运行实例

查看生成的exe文件

find . -type f -executable

结果如图：

4.1 运行Ceres BA

ch9程序的运行一般需要输入BAL数据集文件，在ch9文件夹下。

./bundle_adjustment_ceres ../problem-16-22106-pre.txt

运行结果如图

4.2 运行g2o BA

./bundle_adjustment_g2o ../problem-16-22106-pre.txt

运行结果如图

五、总结

ch9 开始进入SLAM后端优化，将前面的位姿、地图点、投影模型、优化方法全部结合起来，做全局或局部优化。