Building continuous surface models utilising Structure-without-Motion Frameworl

ANU Solar Dishes

Multi-view 3D reconstruction is a central and fundamental topic in computer vision research, where the goal is to reconstruct 3D information of a scene from its multiple 2D images captured by a moving camera. The overarching aim of this project is to develop a new theory, as well as a novel computational framework, that are specifically suitable for efficiently reconstructing 3D from massive unorganized images obtained from distributed sources.

This is part of ARC DP grant (DP120103896, CIs: Hongdong Li and Jonghyuk Kim) and this project is focused on developing methods to build smooth dense surface model utilising Gaussian Process. This project advocates a novel use of Gaussian Process to model 3D surface in continuous form. This will lead to more robust technique for obtaining surface models (than by conventional dense stereo), and will be beneficial for possible later processes such as texture mapping and rendering. 

Software

GPmap Software is publicly released here

References

S. Kim, J. Kim, Global and Local Gaussian Processes for Robotic Map Building, IEEE/IROS 2015 (under review).

S. Kim, J. Kim, Recursive Bayesian Updates for Occupancy Mapping and Surface Reconstruction, Australasian Conference on Robotics and Automation, Melbourne, Dec. 3-5, 2014.

S. Kim and J. Kim, “GPmap: A Unified Framework for 3D Mapping Based on Sparse Gaussian Processes" Field and Service Robotics, Springer Tracts in Advanced Robotics, L. Mejias, P. Corke, J. Roberts (Eds), Volume 105, 2015, pp 319-332, ISBN 978-3-319-07487-0

S. Kim and J. Kim, “Occupancy Mapping and Surface Reconstruction using Local Gaussian Processes with Kinect Sensors,” IEEE Transactions on Systems, Man and Cybernetics - Part B, Special issue on computer vision for RGB-D sensors, 43 (05), 1335-1346, Oct. 2013

J. Kim, Y. Dai, H. Li, X. Du, J. Kim, “Multi-View 3D Reconstruction from Uncalibrated Radially-Symmetric Cameras,” International Conference on Computer Vision (ICCV 2013), Dec. 2-5, Sydney 2013

Updated:  8 September 2015/Responsible Officer:  Dean, CECS/Page Contact:  CECS Marketing