In the recent years we have developed a real-time CPU-based stereo reconstruction algorithm which reconstructs depth information from two camera views. The general idea of the algorithm is to perform accurate and efficient stereo computation of the scene by employing fast stereo matching through an adaptive meshing scheme. The mesh is created based on the detected variance of different image regions. Special properties of the mesh used in our encoding, allows for fast and robust implementation of post-processing algorithms. Our mesh-based algorithm also encodes the depth information with high compression ratio to minimize the size of packets sent over the network.
More information on the stereo reconstruction work can be found in the following publications: