Monocular SLAM Using Ground Textures

We implemented and enhanced the monocular ground-texture-based SLAM algorithm described by Hart et al. (ICRA 2023). Key components of our work included: - Using ORB for detecting and describing keypoints, along with experimenting with alternative algorithms such as SIFT, SURF and HARRIS for better accuracy and performance. - Employing FLANN for efficient keypoint matching. - Incorporating a Visual Bag of Words to identify revisited locations and improve loop closure detection. - Leveraging Factor Graph Optimization with M-Estimators to refine pose estimates and handle outliers effectively. We improved the original algorithm by introducing additional ground textures to evaluate robustness and tested alternative keypoint and descriptor algorithms to optimize performance in diverse conditions. After validating our implementation with public datasets, we collected and tested data using a custom TurtleBot setup. This setup enabled us to assess the algorithm's capabilities in complex scenarios, such as paths with changing textures, uneven surfaces, and varying lighting conditions. Our work demonstrated improved SLAM performance in environments with sparse features and dynamic textures, with potential applications in accessibility assessment and navigation in challenging settings like tunnels.