UNI-3D

Abstract:

3D scene representations, or 3D maps, are a key component of any type of intelligent system (agent). Significant progress has been made in terms of 3D mapping, enabling a wide range of applications, including self-driving cars and augmented reality systems. Yet the current generation of 3D maps have fundamental limitations that can be traced to the fact that mapping algorithms and scene representations are tailor-made for a specific sensor setup. This makes sharing 3D maps between intelligent systems with different sensor setups and true crowdsourced mapping essentially impossible. Further, algorithms and functionality such as visual localization, semantic segmentation, handling scene dynamics, and ensuring privacy need to be re-invented for each map representation. This project proposes an unified map representation to address these limitations. It will be able to integrate data captured from heterogeneous sources, handle the fact that the world is dynamic, and provide full privacy control. The unified map representation will thus enable the next generation of intelligent agents.