Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

eTraM: Event-based Traffic Monitoring Dataset

Aayush Atul Verma, Bharatesh Chakravarthi, Arpitsinh Vaghela, Hua Wei, Yezhou Yang

TL;DR

eTraM addresses the lack of static, event-based traffic datasets by providing 10 hours of high-temporal-resolution, static roadside data with over 2M bounding-box annotations across vehicles, pedestrians, and micro-mobility. Collected with a Prophesee EVK4 HD camera under diverse lighting and weather, the dataset is denoised, frame-binned, and annotated to support detection and tracking tasks. Baseline experiments with RVT, RED, and YOLOv8 show event-based methods, especially those leveraging temporal information, outperform frame-based approaches and demonstrate strong nighttime generalization when trained with nighttime data. Overall, eTraM establishes a valuable resource to advance event-based traffic perception research and ITS applications.

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 hr 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 the ability of event-based models 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 is available at https://eventbasedvision.github.io/eTraM

eTraM: Event-based Traffic Monitoring Dataset

TL;DR

eTraM addresses the lack of static, event-based traffic datasets by providing 10 hours of high-temporal-resolution, static roadside data with over 2M bounding-box annotations across vehicles, pedestrians, and micro-mobility. Collected with a Prophesee EVK4 HD camera under diverse lighting and weather, the dataset is denoised, frame-binned, and annotated to support detection and tracking tasks. Baseline experiments with RVT, RED, and YOLOv8 show event-based methods, especially those leveraging temporal information, outperform frame-based approaches and demonstrate strong nighttime generalization when trained with nighttime data. Overall, eTraM establishes a valuable resource to advance event-based traffic perception research and ITS applications.

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 hr 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 the ability of event-based models 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 is available at https://eventbasedvision.github.io/eTraM
Paper Structure (16 sections, 1 equation, 7 figures, 4 tables)

This paper contains 16 sections, 1 equation, 7 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Datasets
  3. eTraM Dataset
  4. Dataset Acquisition Framework
  5. Preprocessing and Annotation
  6. Dataset Statistics
  7. Dataset Evaluation
  8. Baseline Evaluation
  9. Generalization Evaluation
  10. Generalization on Night time
  11. Generalization on Unseen Scenes
  12. Discussion
  13. Event Cameras in Traffic Monitoring
  14. Static and Ego Event-based Datasets
  15. Conclusion
  16. ...and 1 more sections

Figures (7)

  • Figure 1: Unveiling the Dynamic World of Road Traffic: A glimpse into our event-based traffic monitoring dataset featuring diverse traffic participants including pedestrians, various sized vehicles, and micro-mobility that include cycles, wheelchairs, and e-scooters.
  • Figure 2: Data Collection Setup: The first four images from the top left display daytime data collection sites, the center image shows the Prophesee EVK4 HD event camera and the last four images depict nighttime data collection sites.
  • Figure 3: Event-time Distribution and Object Occurrence Statistics of eTraM: (a) Histogram of event frequency of eTraM (static event dataset) as compared to 1 Megapixel and DSEC (ego-motion event datasets), (b) Shows the object density of various classes across the frame, (c) Power-law distribution of the number of instances within an image for most predominant classes - cars (c1) and pedestrians (c2), and (d) Distribution of two major categories across various traffic sites.
  • Figure 4: Qualitative Results: Showcasing detection of vehicles (cyan), pedestrians (yellow), and micro-mobility (blue) on eTraM.
  • Figure 5: Demonstrating Effectiveness of Event Camera for Traffic Scenarios: Yellow circle (top row) tracks a car that halts at a stop sign with a lack of motion captured in the third frame, red circle (bottom row) tracks a car that violates the stop sign where motion is continuously captured in every frame. Additionally, a green arrow (top row) shows a car traveling at a high speed, resulting in a high event density.
  • ...and 2 more figures