Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Dur360BEV: A Real-world 360-degree Single Camera Dataset and Benchmark for Bird-Eye View Mapping in Autonomous Driving

Wenke E, Chao Yuan, Li Li, Yixin Sun, Yona Falinie A. Gaus, Amir Atapour-Abarghouei, Toby P. Breckon

Abstract

We present Dur360BEV, a novel spherical camera autonomous driving dataset equipped with a high-resolution 128-channel 3D LiDAR and a RTK-refined GNSS/INS system, along with a benchmark architecture designed to generate Bird-Eye-View (BEV) maps using only a single spherical camera. This dataset and benchmark address the challenges of BEV generation in autonomous driving, particularly by reducing hardware complexity through the use of a single 360-degree camera instead of multiple perspective cameras. Within our benchmark architecture, we propose a novel spherical-image-to-BEV module that leverages spherical imagery and a refined sampling strategy to project features from 2D to 3D. Our approach also includes an innovative application of focal loss, specifically adapted to address the extreme class imbalance often encountered in BEV segmentation tasks, that demonstrates improved segmentation performance on the Dur360BEV dataset. The results show that our benchmark not only simplifies the sensor setup but also achieves competitive performance.

Dur360BEV: A Real-world 360-degree Single Camera Dataset and Benchmark for Bird-Eye View Mapping in Autonomous Driving

Abstract

We present Dur360BEV, a novel spherical camera autonomous driving dataset equipped with a high-resolution 128-channel 3D LiDAR and a RTK-refined GNSS/INS system, along with a benchmark architecture designed to generate Bird-Eye-View (BEV) maps using only a single spherical camera. This dataset and benchmark address the challenges of BEV generation in autonomous driving, particularly by reducing hardware complexity through the use of a single 360-degree camera instead of multiple perspective cameras. Within our benchmark architecture, we propose a novel spherical-image-to-BEV module that leverages spherical imagery and a refined sampling strategy to project features from 2D to 3D. Our approach also includes an innovative application of focal loss, specifically adapted to address the extreme class imbalance often encountered in BEV segmentation tasks, that demonstrates improved segmentation performance on the Dur360BEV dataset. The results show that our benchmark not only simplifies the sensor setup but also achieves competitive performance.

Paper Structure

This paper contains 17 sections, 7 equations, 4 figures, 4 tables.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORK
  3. Autonomous Driving Datasets
  4. Vision-based BEV Approaches
  5. Multiple Perspective Camera Models
  6. Multiple Fisheye Camera Models
  7. DUR360BEV DATASET
  8. Sensor Setup
  9. Data Collection and Process
  10. Spherical Imagery
  11. Methodology
  12. Spherical-Image-to-BEV module
  13. Multi-Task Framework and Loss Functions
  14. EXPERIMENTS
  15. Experimental Setup
  16. ...and 2 more sections

Figures (4)

  • Figure 1: An example from the Dur360BEV dataset where (from left-to-right) see 3D bounding box annotation for LiDAR, an exemplar LiDAR in Bird's Eye View (BEV), the dual-fisheye image from our spherical camera and our semantic map based on OpenStreetMap.
  • Figure 2: Sensor placement. Left: the top view of the vehicle equipped with sensors. Right: our spherical camera on top of the LiDAR. Both figures show the coordinates space for each sensor.
  • Figure 3: Validation loss curves for different values of $\gamma$. From top to bottom: $\gamma = 0.2, 0.4, 0.8, 0.6, 1, 2, 5$. The curves illustrate how the choice of $\gamma$ influences the convergence behavior during training.
  • Figure 4: The inference visualisation of the Coarse/Fine sampling strategy and focal loss with $\gamma=2$ on Dur360BEV validation split. Left: Input image; Middle: Prediction; Right: Ground Truth Map.