FLO-EMD: Hybrid Traffic Congestion Classification Framework
Researchers propose FLO-EMD, a hybrid framework for traffic congestion classification that integrates motion-guided attention with empirical mode decomposition. The method uses dense optical flow to guide channel and spatial attention, refining RGB features toward motion-relevant regions, while aggregated flow statistics characterize temporal dynamics. This approach addresses limitations of vision-based methods that rely on static appearance cues and signal-based methods lacking spatial context. The framework jointly captures roadway scene context and non-stationary traffic motion for improved scene-level localization.
Key facts
- FLO-EMD is a hybrid approach coupling motion-guided attention with empirical temporal decomposition
- Dense optical flow guides channel and spatial attention to refine RGB features toward motion-relevant regions
- Aggregated flow statistics are used in parallel for temporal characterization
- The framework addresses limitations of vision-based and signal-based methods
- It jointly captures roadway scene context and non-stationary traffic motion
- The study is published on arXiv with ID 2605.04752
- The approach is designed for accurate traffic congestion classification
- It improves scene-level localization by linking motion evidence to spatial feature selection
Entities
Institutions
- arXiv