News

• Oct 1 (2021): a dense semantic reconstruction paper pushed to arXiv.
• May 26 (2021): a stereo matching paper pused to arXiv, accepted to RA-L/IROS' 21
• March 1 (2021): two papers accepted to ICRA' 21
• Sep 9 (2020): 1 Paper accepted to RA-L
• Jul 2 (2020): 1 Paper accepted to RA-L/IROS' 20

Student Project

Feel free to contact me if you're interested in a bachelor's or master's thesis related to my research areas.

Available
Project	Lazy Camera Calibration Toolbox Development (Yanyan Li, Federico Tombari)

Running

Finished
Project	Reconstructing the MI. Building in a Day (Yanyan Li, Federico Tombari)
Master Thesis	Memory-enhanced Categrory-Level Pose Estimation (Yan DI Yanyan Li, Federico Tombari)
Master Thesis	indoorSLAM: Robust RGB-D SLAM based on Plane-Line-Point features for indoor scenes (Yanyan Li, Federico Tombari)

Teaching

• WS 2020/21 -- Master-Seminar: Computer Vision and Deep Learning for Autonomous Driving
• WS 2019/20 -- Master-Seminar: Computer Vision and Deep Learning for Autonomous Driving

Research topics

• Camera pose estimation for robots and autonomous-driving cars based on geometric and learning methods.
• Geometric prediction and understanding.

Publications

2020
	Y. Li, N. Brasch, Y. Wang, N. Navab, F. Tombari Structure-SLAM: Low-Drift Monocular SLAM in Indoor Environments International Conference on Intelligent Robots and Systems (IROS), October 25-29, 2020. Las Vegas, USA. The first two authors contributed equally. (bib)

TOPIC: Tracking, Reconstruction and Scene Understanding for Indoor Environments

The basic idea of this topic is to propose a dense-accuracy-robust-efficient-semantic (Dares) SLAM system for indoor scenes, which makes uses of RGB-D/Stereo/Monocular and IMU sensors. Benefiting from the constraints between common features of indoor environments and new optimization strategies, we pursue to achieve a high-level accuracy. Thanks to those outstanding contributors, it is cheap to obtain a dense reconstruction map based on estimated camera poses. But how to make the map lighter and higher quality still requires more contributions. Deep neural networks have achieved impressive performance in more and more computer vision topics, so how to combine those SOTA methods with geometric-based SLAM systems to help robots understand unknown scenes is also a necessary and interesting section in this topic.

Structure-SLAM: Low-Drift Monocular SLAM in Indoor Environments[slides]

Co-Planar Parametrization for Stereo-SLAM and Visual-Inertial Odometry [code] [video]

RGB-D SLAM with Structural Regularities [code] [video]

ManhattanSLAM^?: Robust Planar Tracking and Mapping Leveraging Mixture of Manhattan Frames