reconstruction
reconstruction copied to clipboard
3D reconstruction with openCV and SFM
3D reconstruction from multiple 2D images
The current structure from motion (SFM) module from openCV's extra modules only runs on Linux.
As such, I used docker on my Mac to reconstruct the 3D points.
Current docker environment uses Ceres Solver 1.14.0 and OpenCV 3.4.1
Docker Dev Environment
# Build the docker image
docker build -t opencv-sfm .
# Run the docker container mounting `reconstruction` folder to `/app`
docker run -it -v <path_to_reconstruction_folder>:/app opencv-sfm /bin/bash
Run
-
Download 2D temple images from http://vision.middlebury.edu/mview/data
-
Save list of images to
images.txt
:
# images.txt will contain lines of filepath
# /app/temple/temple0302.png
ls temple/*.png > images.txt
sed -i -e 's/^/\/app\//' images.txt
- In the docker container, compile the cpp file
g++ example_sfm.cpp -L/usr/local/lib/ -lopencv_core -lopencv_sfm
- Run the example with the list of images
./a.out # Prints the help text
./a.out images.txt 350 240 360
Test
# Test eigen (http://eigen.tuxfamily.org/dox/GettingStarted.html)
g++ -I /usr/local/Cellar/eigen/3.3.4/include/eigen3 eigen_test.cpp -o eigen
./eigen
# Test with full includes
g++ example_sfm.cpp -I /usr/include/eigen3/ -I/usr/local/include/opencv -I/usr/local/include/opencv2 -L /usr/local/share/OpenCV/3rdparty/lib/ -L/usr/local/lib/ -L /usr/include/eigen3/ -lopencv_core -lopencv_imgproc -lopencv_highgui -lopencv_ml -lopencv_optflow -lopencv_sfm -lopencv_viz