OpenMVG + OpenMVS的配置与使用
项目地址：
https://github.com/openMVG/openMVG OpenMVG
https://github.com/cdcseacave/openMVS OpenMVS
Pipeline:
Structure from Motion(SfM) -> Multi-View Stereo(MVS) -> Surface Generation(SG) -> Texture Mapping(TM)
OpenMVG完成SFM，后续的MVS/SG/TM 由OpenMVS实现
所用的环境：
Ubuntu16.04 (Win10子系统版本)
OpenMVG：
配置过程中注意，不要用从github上下载解压的zip，要用git clone的形式
按照官方文档进行：
https://github.com/openMVG/openMVG/blob/master/BUILD.md#linux
此部分根据官方文档就应该能完成
OpenMVS:
照例按照官方文档进行：
https://github.com/cdcseacave/openMVS/wiki/Building
注意点1：
main_path='pwd' 这一步中‘pwd’更改为VCGLib所在的位置（这是后续要安装的一个库，推荐先装这个库）
注意点2：
Ceres部分
这里面的git clone里面的链接挂了，换成
https://github.com/ceres-solver/ceres-solver
注意点3：
后续编译找不到Eigen库，用下面的命令安装
sudo apt-get install libeigen3-dev
注意点4：
如果没有GPU，用不了CUDA那么需要在openMVS目录下找到CmakeLists.txt文件，找到CUDA开关，设置为OFF：
SET(OpenMVS_USE_CUDA OFF CACHE BOOL "Enable CUDA library")
运行脚本：
相机不在database中时我的版本：
#!/bin/bash
# reconstruction.sh
IMAGE_PATH='/home/chenyufeng/jieditanshuan'
CAMERA_SENSOR_WIDTH_DIRECTORY="/home/chenyufeng/openMVG/src/openMVG/exif/sensor_width_database"
OPENMVG_SFM_PATH='/home/chenyufeng/openMVG_Build/software/SfM'
OPENMVG_BIN_PATH='/home/chenyufeng/openMVG_Build/Linux-x86_64-RELEASE'
OPENMVG_OUT_PATH=$IMAGE_PATH/out
MATCHES_DIR=$OPENMVG_OUT_PATH/matches
REC_DIR=$OPENMVG_OUT_PATH/reconstruction_incremental
OPENMVS_BIN_PATH='/home/chenyufeng/openMVS_build/bin'
FOCAL_LENGTH=7200
echo "--- Reconstruction Begin ---"
cd $IMAGE_PATH && mkdir out
cd $OPENMVG_OUT_PATH && mkdir matches reconstruction_incremental
cd
echo "--- OpenMVG Intrinsics analysis 1/12 ---"
$OPENMVG_BIN_PATH/openMVG_main_SfMInit_ImageListing -i $IMAGE_PATH -f $FOCAL_LENGTH -o $MATCHES_DIR
#$OPENMVG_BIN_PATH/openMVG_main_SfMInit_ImageListing -i $IMAGE_PATH -o $MATCHES_DIR -d $CAMERA_SENSOR_WIDTH_DIRECTORY/sensor_width_camera_database.txt
echo "--- OpenMVG Compute features 2/12 ---"
# $OPENMVG_BIN_PATH/openMVG_main_ComputeFeatures -i $MATCHES_DIR/sfm_data.json -o $MATCHES_DIR -m AKAZE_FLOAT -p HIGH
$OPENMVG_BIN_PATH/openMVG_main_ComputeFeatures -i $MATCHES_DIR/sfm_data.json -o $MATCHES_DIR
echo "--- OpenMVG Compute Matches 3/12 ---"
$OPENMVG_BIN_PATH/openMVG_main_ComputeMatches -i $MATCHES_DIR/sfm_data.json -o $MATCHES_DIR
# $OPENMVG_BIN_PATH/openMVG_main_ComputeMatches -i $MATCHES_DIR/sfm_data.json -o $MATCHES_DIR -n ANNL2
echo "--- OpenMVG Do Incremental reconstruction 4/12 ---"
$OPENMVG_BIN_PATH/openMVG_main_IncrementalSfM -i $MATCHES_DIR/sfm_data.json -m $MATCHES_DIR -o $REC_DIR
echo "--- OpenMVG Colorize Structure 5/12 ---"
$OPENMVG_BIN_PATH/openMVG_main_ComputeSfM_DataColor -i $REC_DIR/sfm_data.bin -o $REC_DIR colorized.ply
echo "--- OpenMVG Structure from Known Poses (robust triangulation) 6/12 ---"
$OPENMVG_BIN_PATH/openMVG_main_ComputeStructureFromKnownPoses -i $REC_DIR/sfm_data.bin -m $MATCHES_DIR -f $MATCHES_DIR matches.f.bin -o $REC_DIR robust.bin
echo "--- OpenMVG Compute SfM Data Color 7/12 ---"
$OPENMVG_BIN_PATH/openMVG_main_ComputeSfM_DataColor -i $REC_DIR/robust.bin -o $REC_DIR robust_colorized.ply
echo "--- OpenMVG to OpenMVS 8/12 ---"
cd $OPENMVG_OUT_PATH/reconstruction_incremental
$OPENMVG_BIN_PATH/openMVG_main_openMVG2openMVS -i sfm_data.bin -o scene.mvs
echo "--- OpenMVS: Dense Point-Cloud Reconstruction 9/12 ---"
$OPENMVS_BIN_PATH/DensifyPointCloud scene.mvs
echo "--- OpenMVS: Rough Mesh Reconstruction 10/12 ---"
$OPENMVS_BIN_PATH/ReconstructMesh scene_dense.mvs
# $OPENMVS_BIN_PATH/ReconstructMesh -d 4 scene_dense.mvs
echo "--- OpenMVS: Mesh Refinement 11/12 ---"
$OPENMVS_BIN_PATH/RefineMesh --resolution-level=4 scene_mesh.mvs
# $OPENMVS_BIN_PATH/RefineMesh scene_mesh.mvs
echo "--- OpenMVS: Mesh Texturing 12/12 ---"
$OPENMVS_BIN_PATH/TextureMesh scene_dense_mesh.mvs
cd
echo "--- Done ---"
相关参数
对脚本中的部分参数作些说明：
- -f $FOCAL_LENGTH
在第一步，如果图像的EXIF信息缺失或相机型号不在数据库列表中，则需要手动设置焦距，粗略的计算方法为：图像长边值乘以1.2，比如1024*768的图像数据集，FOCAL_LENGTH=1228.8（1024*1.2）
最后的结果
在图片文件夹下的out中，复制出
scene_dense_mesh_texture.ply
和scene_dense_mesh_texture.png
后可以用MeshLab查看
供参考的博客
http://leohope.com/%E8%A7%A3%E9%97%AE%E9%A2%98/2018/08/03/openmvg-openmvs/