SEENIC: dataset for Spacecraft posE Estimation with NeuromorphIC vision

Dataset used in the paper "Towards Bridging the Space Domain Gap for Satellite Pose Estimation using Event Sensing" (arXiv, IEEE Xplore), for the purpose of satellite pose estimation with an event camera. Both events and ground truth camera poses were captured across the 20 scenes in total. There are two trajectories, five lighting configurations and two camera speeds. All combinations of trajectory type, speed and lighting configuration were enumerated for capture. Sample event frames and dataset statistics are available in the paper linked above, along with our pose estimation method used on this dataset. Live-capture scene names use the following encoding: {satellite model}-{trajectory}-{speed}-{lighting configuration} The calibration scene (calibration.tar.gz) includes multiple views of a chessboard used to calibrate the camera intrinsics and extrinsics for the live-capture scenes. Camera parameters calibrated using this scene can be found in the calib.txt file, with the format: fx fy cx cy k1 k2 p1 p2 k3. All live-capture scenes have the same data format: scene/ poses/ -- Raw timestamped robot gripper to base transforms cam-poses.csv -- Ground truth camera poses with the format {timestamp, Rx, Ry, Rz, x, y, z} events.csv -- Event stream with the format {timestamp, x, y, polarity (0=off, 1=on)} meta.json -- Metadata file with camera frame dimensions Note: all timestamps are in microseconds. The synthetic scene (synthetic.tar.gz) has the following data format: synthetic/ poses/ -- Sequential poses captured at a constant time interval events.txt -- Event stream with the format: time (float s), x, y, polarity (0=off, 1=on) as specified at https://rpg.ifi.uzh.ch/davis_data.html camera_intrinsics.txt -- The camera intrinsic matrix (space separated) Note: please refer to the paper referenced below for further details on using this synthetic scene. When using the data in an academic context, please cite the following paper. @INPROCEEDINGS{10160531,
author={Jawaid, Mohsi and Elms, Ethan and Latif, Yasir and Chin, Tat-Jun},
booktitle={2023 IEEE International Conference on Robotics and Automation (ICRA)},
title={Towards Bridging the Space Domain Gap for Satellite Pose Estimation using Event Sensing},
year={2023},
volume={},
number={},
pages={11866-11873},
keywords={Adaptation models;Satellites;Pose estimation;Lighting;Robot sensing systems;Robustness;Data models},
doi={10.1109/ICRA48891.2023.10160531}
}

本数据集用于论文《基于事件感知的卫星位姿估计：弥合空间域差距》（Towards Bridging the Space Domain Gap for Satellite Pose Estimation using Event Sensing，收录于arXiv、IEEE Xplore），旨在开展基于事件相机（event camera）的卫星位姿估计（pose estimation）任务。数据集共采集了总计20个场景的事件数据与相机真值位姿，涵盖2种轨迹、5种光照配置与2种相机速度，所有轨迹类型、速度与光照配置的组合均被枚举用于采集流程。

论文中附带了示例事件帧与数据集统计信息，以及本数据集上使用的位姿估计方法。实时采集场景的命名采用如下编码规则：{卫星型号}-{轨迹}-{速度}-{光照配置}。

校准场景（calibration.tar.gz）包含用于校准实时采集场景相机内参（intrinsics）与外参（extrinsics）的棋盘格（chessboard）多视图数据。通过该场景校准得到的相机参数存储于calib.txt文件中，格式为：fx fy cx cy k1 k2 p1 p2 k3。

所有实时采集场景的数据格式保持一致：
scene/
├─ poses/ -- 带时间戳的机械臂夹爪到基座的原始变换
├─ cam-poses.csv -- 真值相机位姿，格式为{timestamp, Rx, Ry, Rz, x, y, z}
├─ events.csv -- 事件流（event stream）数据，格式为{timestamp, x, y, polarity (0=暗事件，1=亮事件)}
└─ meta.json -- 包含相机帧尺寸的元数据文件

注意：所有时间戳单位均为微秒。

合成场景（synthetic.tar.gz）的数据格式如下：
synthetic/
├─ poses/ -- 以固定时间间隔采集的序列位姿
├─ events.txt -- 事件流数据，格式为：time (float s), x, y, polarity (0=off, 1=on)，详见https://rpg.ifi.uzh.ch/davis_data.html
└─ camera_intrinsics.txt -- 相机内参矩阵（空格分隔）

备注：关于该合成场景的使用细节，请参阅下述引用论文。若在学术场景中使用本数据集，请引用如下文献：
@INPROCEEDINGS{10160531,
author={Jawaid, Mohsi and Elms, Ethan and Latif, Yasir and Chin, Tat-Jun},
booktitle={2023年IEEE国际机器人与自动化会议（ICRA)},
title={Towards Bridging the Space Domain Gap for Satellite Pose Estimation using Event Sensing},
year={2023},
volume={},
number={},
pages={11866-11873},
keywords={自适应模型；卫星；位姿估计；光照；机器人传感系统；鲁棒性；数据模型},
doi={10.1109/ICRA48891.2023.10160531}
}

搜集汇总

背景与挑战

背景概述

SEENIC是一个专为卫星姿态估计设计的数据集，特别适用于事件相机的应用。它包含20个场景的数据，涵盖多种轨迹、光照和速度组合，提供事件流和地面真实姿态信息，支持学术研究。