Monocular Semi-Dense SLAM (Tutor: Baosheng Zhang, Wechat: XXXX204682)

Introduction

The task of this project is to build a semi-dense SLAM (Simultaneous Localization and Mapping) system using a single monocular camera, enabling real-time 3D environment reconstruction and camera tracking.

Requirements

1> Implement real-time camera pose estimation and semi-dense 3D map reconstruction using only a monocular camera.

2> The system should robustly operate in both indoor and outdoor environments, adapting to different lighting conditions and scene textures.

3> The implementation should be deployed on an actual camera hardware in real-world settings.

Advanced Requirements

1> Implement loop closure detection and global bundle adjustment to reduce drift over long trajectories.

2> The system should maintain real-time performance on embedded systems or mobile platforms while handling both indoor confined spaces and outdoor open areas.

Reference Materials

1. LSD-SLAM: Large-Scale Direct Monocular SLAM: https://vision.in.tum.de/research/vslam/lsdslam
2. ORB-SLAM: A Versatile and Accurate Monocular SLAM System: https://webdiis.unizar.es/~raulmur/orbslam/
3. Direct Sparse Odometry: https://vision.in.tum.de/research/vslam/dso
4. OpenCV Camera Calibration: https://docs.opencv.org/4.x/dc/dbb/tutorial_py_calibration.html
5. Semi-Dense Visual Odometry: https://arxiv.org/abs/1607.02565
6. DeepFactors: Real-Time Probabilistic Dense Monocular SLAM: https://arxiv.org/abs/2001.05049
7. TANDEM: Tracking and Dense Mapping in Real-time using Deep Multi-view Stereo:https://arxiv.org/abs/2111.07418

Created on: Nov. 19, 2025