EBSSA Dataset

The Event-Based Space Situational Awareness (EBSSA) dataset [7] contains neuromorphic event camera recordings of resident space objects (satellites, rocket bodies), planets, and stars.

Overview

Property	Value
Sensors	ATIS (304 x 240) and DAVIS240C (240 x 180)
Labelled recordings	84
Unlabelled recordings	153
File format	MATLAB .mat (per recording) or HDF5 (combined)
Event fields	x, y, polarity, timestamp
Labels	Bounding box trajectories (expert annotated)
Source	Western Sydney University, International Centre for Neuromorphic Systems

Sample Recording

The following video shows a raw EBSSA recording of a satellite (SL-8 rocket body, NORAD 21938) tracked against a star field:

Recording 20170214-21-15, SL8RB (NORAD 21938). The satellite appears as a faint streak moving across the field of view while stars produce stationary event clusters.

Expert Labels

This video shows the EBSSA expert label overlay, demonstrating the ground-truth bounding box annotations used for evaluation:

Expert label data showing bounding box annotations tracking satellites across event camera recordings.

Directory Layout

EBSSA/
├── Labelled Data/
│   ├── 20170214-20-58_22285_SL-16RB_labelled.mat
│   ├── archenar_leos_11_33_atis_td_labelled.mat
│   ├── archenar_leos_11_33_davis_td_labelled.mat
│   ├── ...                    (84 recordings)
│   └── HDF5_Format/
│       ├── plot_trajectory.py
│       └── Readme.txt
├── Unlabelled Data/
│   └── ...                    (153 recordings)
├── converted/                 (optional: pre-converted .h5 + .npy)
│   ├── train_h5_1/
│   └── val_h5_1/
└── Readme.txt

Usage

from spikeseg.data import EBSSADataset

dataset = EBSSADataset(
    root="/path/to/EBSSA",
    split="train",
    sensor="all",         # "ATIS", "DAVIS", or "all"
    n_timesteps=10,
    height=128,
    width=128,
    polarity_channels=True,   # 2 channels (ON/OFF)
    train_ratio=0.9,
)

for events, labels in dataset:
    # events: (T, C, H, W) voxel grid
    # labels: bounding box mask or dict
    pass

Configuration

data:
  dataset: "ebssa"
  data_root: "/path/to/EBSSA"
  sensor: "all"
  n_timesteps: 10
  input_height: 128
  input_width: 128
  input_channels: 2
  windows_per_recording: 1

Citation

S. Afshar, A. P. Nicholson, A. van Schaik, and G. Cohen, "Event-based object detection and tracking for space situational awareness," IEEE Sensors Journal, vol. 20, no. 24, pp. 15117--15132, 2020.