Toward Real-Time Dense 3d Reconstruction using Stereo Vision
School of Electrical Engineering, Royal Institute of Technology
Royal Institute of Technology, 2011
State of the art Structure from Motion algorithms can produce a real-time sparse 3d map of the environment, in a fast, robust and efficient way. However, dense 3d maps would be very useful for accurate Augmented Reality with occlusion management. This project focus on generating accurate dense depth-maps in near real-time from the data provided by a Structure from Motion algorithm. The presented algorithm uses a TVL1 optimization scheme with a novel initialization from matched 2d keypoints. The generated depth-maps are good candidates for on-line 3d model generation and update.
January 4, 2012 by hgpu
