Reproduce “VOOM: Robust Visual Object Odometry and Mapping using Hierarchical Landmarks” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yutong Wang, Chaoyang Jiang, Xieyuanli Chen
VOOM, a visual object odometry and mapping framework developed by researchers at Beijing Institute of Technology and National University of Defense Technology, tightly integrates low-level feature points and high-level dual quadric objects to achieve superior localization accuracy compared to state-of-the-art feature-based visual SLAM systems like ORB-SLAM2, while also building a semantically rich map.
Key insights
- Integrating objects and features in a coarse-to-fine hierarchical manner allows objects to assist and improve feature point processing, rather than merely supplementing it.
- Using a comprehensive metric like normalized Wasserstein distance for object data association significantly enhances robustness and accuracy compared to traditional IoU.
- Objects can effectively reduce odometry drift and improve mapping even without explicit loop closure, by providing robust high-level context for local map optimization.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.