Real Time 3D Reconstruction from Monocular Video

June

1st week

  • Adding OpenCV and OpenMVG as GIT submodules to the project
    • It allows local compilation and debug
    • Changes to main repositories can be submitted as contributions
  • Porting Essential matrix estimation and decomposition to OpenCV
    • Both algorithms are now implemented in 3.0 version
    • OpenCV seems up to twice times faster thanks to OpenCL support
  • Analyzing implementations of stereo matching algorithms
    • Dense depth map for current frame $Ii$ is fused with $I{0..i-1}$ in a cost volume, according to @newcombe2010dtam
    • PathMatch stereo algorithm from @bleyer2011patchmatch

2nd week

  • Implemented support for block matching in GPU using OpenCV CUDA support
    • Added a basic point cloud rendering shader to represent depths from disparity maps using cv::cuda::reprojectTo3D()
  • Added visualization using OpenGL avoiding to download the image from GPU
    • Refactored rendering code to use a "layer" drawing concept in OpenGL
    • Created a Layer to render OpenCV images coming from CPU / GPU
  • Studying and documenting a general framework to feedback reconstruction and recognition (see rec2rec.md for more information)
  • Recorded a video demoing the camera tracking using the OpenCV essential estimator

tracking

3rd week

  • Documentation:
    • Added more documentation related with fibrations and fiber bundles
    • Improved description of relations between Lie groups and Lie algebras
    • Added total variation description as the regularization term in optimization approaches
      • Added explanation of linearization of the $GL$ group using a structural group

4rd week

  • Implementation and code review:
      - Analysis of the [Urho3D](https://github.com/urho3d/Urho3D) project as a C++ rendering engine for future AR applications
      - Study JNI bridge for executing native ARM code from Java applications in Android
      - Reading optimization procedures in the $\mathfrak{SE}(3)$ algebra instead in the $SE(3)$ group
    

References