Scholarlink

<p>Autonomous spacecraft pose estimation is critical for non-cooperative satellite servicing and rendezvous operations, yet existing datasets lack coverage for DFH-4 satellite platforms, which are widely used as high-orbit commercial communication satellites. This paper presents D4PED (<b>D</b>FH-<b>4</b> <b>P</b>ose <b>E</b>stimation <b>D</b>ataset), the first comprehensive synthetic dataset specifically designed for DFH-4 spacecraft pose estimation. Our methodology integrates physics-based rigid body dynamics modeling with high-fidelity ray-tracing rendering using Blender’s Cycles engine, creating realistic sequential pose estimation data that captures complex rotational behaviors of non-cooperative spacecraft. The dataset comprises 59,960 images generated from SPEED+ compatible scenarios and 10,000 images from five distinct dynamics simulation scenarios, ensuring compatibility with existing benchmarks while covering realistic operational conditions. We employ advanced 3D domain data augmentation techniques including random background rotation and brightness, lighting variation, as well as image-domain noise injection and blurring to enhance dataset diversity and realism. Extensive experimental validation using YOLOv8 with EPnP as a generic pose estimation pipeline, as well as SPEED+ solutions including SPNv2 and UDA, demonstrates effective performance, ensuring the utility of D4PED for spacecraft pose estimation tasks involving DFH-4 satellites. D4PED enables seamless integration with existing SPEED+ methodologies while providing the first dedicated resource for developing robust pose estimation algorithms tailored to China’s satellite infrastructure and enhancing the diversity of existing pose estimation datasets, filling a critical gap for non-cooperative DFH-4 spacecraft scenarios.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

D4PED: A Realistic Pose Estimation Dataset for DFH-4 Satellite Driven by Dynamics

  • 期刊论文

  • 发表日期：

    2026年2月27日

查看全文
  • Xuyang Chen,
  • Yuan Xue,
  • Zihao Liu,
  • Liang Xu,
  • Xiaofan Wang

摘要

Autonomous spacecraft pose estimation is critical for non-cooperative satellite servicing and rendezvous operations, yet existing datasets lack coverage for DFH-4 satellite platforms, which are widely used as high-orbit commercial communication satellites. This paper presents D4PED (DFH-4 Pose Estimation Dataset), the first comprehensive synthetic dataset specifically designed for DFH-4 spacecraft pose estimation. Our methodology integrates physics-based rigid body dynamics modeling with high-fidelity ray-tracing rendering using Blender’s Cycles engine, creating realistic sequential pose estimation data that captures complex rotational behaviors of non-cooperative spacecraft. The dataset comprises 59,960 images generated from SPEED+ compatible scenarios and 10,000 images from five distinct dynamics simulation scenarios, ensuring compatibility with existing benchmarks while covering realistic operational conditions. We employ advanced 3D domain data augmentation techniques including random background rotation and brightness, lighting variation, as well as image-domain noise injection and blurring to enhance dataset diversity and realism. Extensive experimental validation using YOLOv8 with EPnP as a generic pose estimation pipeline, as well as SPEED+ solutions including SPNv2 and UDA, demonstrates effective performance, ensuring the utility of D4PED for spacecraft pose estimation tasks involving DFH-4 satellites. D4PED enables seamless integration with existing SPEED+ methodologies while providing the first dedicated resource for developing robust pose estimation algorithms tailored to China’s satellite infrastructure and enhancing the diversity of existing pose estimation datasets, filling a critical gap for non-cooperative DFH-4 spacecraft scenarios.