Our approach reconstructs dense geometric and photometric texture maps of the environment whilst simultaneously performing localisation. The main advanges include :
- Robustness and accuracy : Exploiting all the information in the images within state-of-the-art non-linear estimation algorithms leads to highly robust and accurate results.
- Real-time efficiency : This is obtained via an optimized implementation that has been easily parrellelised on the GPU.
- Scalability : This is easily obtained by adjusting the resolution of the image leading to a robust decrease (or increase) in performance.
- Cross-platform architecture : The implementation presented here is based on OpenGL which is cross-platform allowing the software to be run on both high and low-end platforms.
- Place recognition and Loop-closure : Using state-of-the-art regonition and bundle-adjustment approaches we can correct the map over large-scales when places are re-visted.
Some videos of this work can be found here:
Related publications :
Maxime Meilland, Andrew I. Comport. On unifying key-frame and voxel-based dense visual SLAM at large scales. International Conference on Intelligent Robots and Systems, 2013, Tokyo, Japan. 2013. BEST PAPER AWARD <hal-01357359>
Tommy Tykkala, Andrew I. Comport, Joni-Kristian Kamarainen. Photorealistic 3D Mapping of Indoors by RGB-D Scanning Process. International Conference on Intelligent Robots and Systems, 2013, Tokyo, Japan. 2013. <hal-01357355>
Andru Putra Twinanda, Maxime Meilland, Sidibé Désiré, Andrew I. Comport. On Keyframe Positioning for Pose Graphs Applied to Visual SLAM. 2013 IEEE/RSJ International Conference on Intelligent Robots and Systems, 5th Workshop on Planning, Perception and Navigation for Intelligent Vehicles, 2013, Tokyo, Japan. 2013. <hal-01357358>