eTraM: Event-based Traffic Monitoring Dataset

CVPR 2024 (Highlight)

Aayush Atul Verma*       Bharatesh Chakravarthi*        Arpitsinh Vaghela*
Hua Wei        Yezhou Yang
(* Equal Contribution)
Arizona State University
Paper arXiv Code Data Dataset Description
Poster Video Supplementary Material

eTraM a novel fully event-based traffic perception dataset curated using the state-of-the-art (SOTA) high-resolution Prophesee EVK4 HD event camera. The dataset spans over 10 hours of annotated data from a static perspective that facilitates comprehensive traffic monitoring. The dataset encompasses various weather and lighting conditions spanning challenging scenarios such as high glare, overexposure, underexposure, nighttime, twilight, and rainy days. eTraM includes 2M bounding box annotations of traffic participants such as vehicles, pedestrians, and various micro-mobility.

Teaser Video

Abstract

Event cameras, with their high temporal and dynamic range and minimal memory usage, have found applications in various fields. However, their potential in static traffic monitoring remains largely unexplored. To facilitate this exploration, we present eTraM - a first-of-its-kind, fully event-based traffic monitoring dataset. eTraM offers 10 hrs of data from different traffic scenarios in various lighting and weather conditions, providing a comprehensive overview of real-world situations. Providing 2M bounding box annotations, it covers eight distinct classes of traffic participants, ranging from vehicles to pedestrians and micro-mobility. eTraM's utility has been assessed using state-of-the-art methods for traffic participant detection, including RVT, RED, and YOLOv8. We quantitatively evaluate their ability to generalize on nighttime and unseen scenes. Our findings substantiate the compelling potential of leveraging event cameras for traffic monitoring, opening new avenues for research and application.

eTraM - Data Collection Setup and Diversity


We utilized Prophesee EVK4 HD event camera, notable for its high resolution (1280×720 px), temporal resolution (over 10,000 fps), dynamic range (above 120 dB), and exceptional low light cutoff (0.08 Lux), to capture high-quality data. The event camera was strategically positioned at approximately 6m with a pitch angle of 35o to the ground. This configuration is chosen to maintain consistency with the placement of traffic cameras and ensure comprehensive coverage of interactions between diverse traffic participants.

eTraM - Statistics


eTraM encompasses three distinct traffic monitoring scenes with 5 hrs of intersection, 3 hrs of roadway, and 2 hr of local street data sequences. Data for each scene is collected at multiple locations. For instance, the intersection scene contains data from 2 four-way, threeway, daytime, nighttime, and twilight data totaling up to 10 hr of data with 5 hrs of daytime and nighttime data. eTraM contains 2M instances of 2D bounding box annotations for traffic participant detection. These annotations additionally include object IDs, making it possible to evaluate multi-object tracking, as shown in the supplementary material. The annotation classes encompass a range of traffic participants, from various vehicles (cars, trucks, buses, and trams) to pedestrians and micro-mobility (bikes, bicycles, and wheelchairs).

Sample Data Recordings

BibTeX

@InProceedings{Verma_2024_CVPR,
    author    = {Verma, Aayush Atul and Chakravarthi, Bharatesh and Vaghela, Arpitsinh and Wei, Hua and Yang, Yezhou},
    title     = {eTraM: Event-based Traffic Monitoring Dataset},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2024},
    pages     = {22637-22646}
}