Scholarlink

<p>Perspective projection is a widely used technique in monocular 3D object detection. However, UAV imagery presents a unique challenge due to its dynamically changing viewpoints. This violates the horizontal-view assumption that most existing methods rely on, as they are primarily designed for autonomous driving scenarios. To overcome this limitation, we propose MonoFHD, a monocular 3D object detection framework that leverages flight height data. Its core is a Flight-Height-Aware Depth Estimation (FHD) module, which fuses real-time flight height data with geometric attributes such as the object’s height, projected 3D center, and rotation matrix, to estimate depth reliably. We further employ a depth offset head to improve depth accuracy. On the CARLA Drone dataset, MonoFHD achieves a 3.2% improvement in mean Average Precision (mAP) over the baseline without sacrificing inference speed. This work provides a lightweight, compatible, and practical detection framework for UAV platforms, which improves detection accuracy through enhanced depth estimation.</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

MonoFHD: leveraging flight height data for UAV monocular 3D object detection

  • 期刊论文

  • 发表日期：

    2026年1月7日

查看全文
  • Heshan Dong,
  • Hongya Tuo,
  • Lindong Wang,
  • Hongjie Zhou

摘要

Perspective projection is a widely used technique in monocular 3D object detection. However, UAV imagery presents a unique challenge due to its dynamically changing viewpoints. This violates the horizontal-view assumption that most existing methods rely on, as they are primarily designed for autonomous driving scenarios. To overcome this limitation, we propose MonoFHD, a monocular 3D object detection framework that leverages flight height data. Its core is a Flight-Height-Aware Depth Estimation (FHD) module, which fuses real-time flight height data with geometric attributes such as the object’s height, projected 3D center, and rotation matrix, to estimate depth reliably. We further employ a depth offset head to improve depth accuracy. On the CARLA Drone dataset, MonoFHD achieves a 3.2% improvement in mean Average Precision (mAP) over the baseline without sacrificing inference speed. This work provides a lightweight, compatible, and practical detection framework for UAV platforms, which improves detection accuracy through enhanced depth estimation.