|Topic:||Využití hustého mračna bodů pro lokalizaci ve vnitřních prostorách|
|Supervisor:||Ing. Michal Polic|
|Announce as:||Diplomová práce, Bakalářská práce, Semestrální projekt|
|Description:||The accurate localization of a camera device plays an essential role in Augmented Reality (AR), i.e., the correct alignment of virtual objects into the real scene. Most of the current approaches work well in environments rich in textures. However, some indoor places, e.g., construction, corridors, factory floor, and textureless rooms are hard for localization. Therefore, the student will focus on usage of dense pointcloud for this task. The work consist of comparing the state of the art methods for indoor localization from RGB-D images. Further, the student focuses on found problems of existing approaches and measure the quality of localization by employing SoTA pointcloud alignment methods.
Review state of the art approaches for the localization from RGB-D images. Find the origins of the localization mistakes and propose a new solutions of found problems. Compare the localization accuracy by using SoTA approaches for dense point cloud alignment (e.g., PREDARTOR (CVPR2021), FCGF (CVPR2019), TEASER++ (IEEE Transactions on Robotics 2020)).
|Bibliography:||* Choy, Christopher, Jaesik Park, and Vladlen Koltun. "Fully convolutional geometric features." Proceedings of the IEEE/CVF International Conference on Computer Vision. 2019.
* Yang, Heng, Jingnan Shi, and Luca Carlone. "Teaser: Fast and certifiable point cloud registration." IEEE Transactions on Robotics 37.2 (2020): 314-333.
* Huang, Shengyu, et al. "PREDATOR: Registration of 3D Point Clouds with Low Overlap." Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2021.