Strucutre from Accidental Motion Kit is a 3D reconstruction software kit. It is capable of reconstructing 3D sparse structure of an object from a sequence of images with limited camera translation and rotation. The reconstruction model is based on Fisher Yu and David Gallup's work in 2014, 3D Reconstruction from Accidental Motion.
Please see CVPR 2014 3D Reconstruction from Accidental Motion for more information about the theories behind the SAM Kit.
This kit uses Ceres Solver for Bundle Adjustment.
Please see ceres-solver.org for more information.
Note: It is best to build the Ceres Solver on Linux or macOS, to build on Windows is a bit troublesome, Linux Subsystem on Windows 10 is highly recommended.
Follow Installation Instructions here.
Follow Installation Instructions here.
- Save image sequences (not included) in a folder, such as
example/images. - Run following Script in Python (see
example/example.py), script will generate a bundler filebundle.out, and an optical flow tracking resultoptical_flow.jpgin the output folder (example/output).
python example.py
- Bundle adjustment will be automatically triggered once the featurue tracking is completed. BA result (
final.ply) can be found in the output folder (example/output).init.plyis the initial guess data.
- More Examples.
- More test cases to prove that it is robust, as the Kit is written from scratch. It should achieve similar reconstruction result as the version I coded for FYP.
I re-wrote SAM Kit from scratch for following reasons:
- The original one I wrote for FYP is full of hacking. I re-organised the Python Code, made it neat and structured, added some documentation in the code for quick reference, consdering it may be used for further research development
- This SAM kit is modified from Ceres Solver, removing most of the unnecesary parts, building is faster than previous version.
I did not include the multiple sparse matching, as it is not current priority.