Reproduce “PMOF: A Dataset and Benchmark for Passenger Monitoring Using Overhead Fisheye Cameras” 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
Stella Katharina Wermuth, Qazi Arbab Ahmed, Klaus Neumann, Thorsten Jungeblut
The paper introduces PMOF, a novel public dataset comprising overhead fisheye imagery captured inside a moving vehicle, to facilitate passenger monitoring in autonomous transport. Benchmarking reveals a substantial domain gap between static and dynamic environments, with the dataset and a rotation-aware augmentation strategy improving person detection performance and cross-domain generalization.
Key insights
- A significant domain gap exists between static surveillance environments (e.g., offices) and dynamic in-vehicle settings, leading to performance degradation when models are transferred without domain-specific training.
- The proposed rotation-aware augmentation strategy, designed to preserve fisheye geometry, consistently enhances detection performance and model robustness, both for in-domain and cross-domain tasks.
- Combining data from both static (CEPDOF) and dynamic (PMOF) fisheye environments, especially with rotation-aware augmentation, substantially improves cross-domain generalization capabilities for person detection.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.