Home

Welcome to the VoxelHashing_demo wiki!

The goal of this project is to have a fast + scalable 3D reconstruction pipeline. It's an implementation of this paper : VoxelHashing

The pipeline is divided in following stages :

1. Acquisition + point cloud generation : acquire depth(+ possibly RGB) frames from 3D sensor and generate point clouds for further processing.
2. Tracking : Compute dense correspondence between frames n and n-1, use these correspondences to
   a) solve a non-linear least squares formulation of point-to-plane distance metric
   b) project the solution in prev step to get a valid rigid-body motion matrix (SE(3) group object) that describes camera movement
3. Reconstruction : The camera-movement matrix such obtained is used as input to a depth-integration system. We voxelize the point clouds and store them into a global-accumulation buffer
4. Visualization : Enable rendering of said voxel-geometry
5. Streaming output : to network/secondary storage -- most likely I won't implement this