A Spatially Aligned RGB-Event Modality Dataset for Roadside Traffic Object Detection
2026-08-27
Event cameras are promising for traffic perception. However, due to the limited availability of high-quality datasets, especially those with well-aligned and uniformly annotated dual-modal ones, roadside traffic applications remain underexplored. We propose the Dual-modal Roadside Traffic Dataset (DRTD) to support traffic object detection research from a fixed roadside perspective. DRTD includes 29,445 images at a 1280 × 720 resolution captured using a custom dual-modal imaging system. A collection workflow is used to achieve spatio-temporal alignment and automatic annotation based on a visual segmentation large model. All annotations are consistent across modalities enabling their use in single-modal, fusion, dual-modal, or annotation generation research. Moreover, raw event stream, uncalibrated RGB data, and camera calibration parameters are provided to facilitate studies on sensor fusion, calibration, and data representation. A benchmark using advanced detection models is presented to validate data quality and usability. DRTD addresses a current gap in publicly available event-based datasets on roadside traffic and supports reproducible research in event-based multi-modal traffic perception.