Skip to main navigation Skip to search Skip to main content

TA-MDet: Terrain-Aware Multimodal 3D Object Detection for UAV Platforms

  • Yanze Jiang
  • , Yanfeng Gu
  • , Xian Li*
  • *Corresponding author for this work
  • School of Electronics and Information Engineering, Harbin Institute of Technology

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Multimodal 3D object detection method combines the spatial geometric information from LiDAR with the visual semantic information from cameras. However, existing multimodal methods are mainly designed for vehicle platforms. In UAV top-down scenarios, they often suffer from inaccurate feature alignment and high computational cost. To address these challenges, we propose TA-MDet, a terrain-aware multimodal 3D object detection method for UAV platforms. It uses sparse BEV features as queries and generates near-ground 3D query points from the predicted terrain, enabling accurate cross-modal feature alignment without depth estimation. Furthermore, we construct a UAV-based multimodal dataset UAV-LC3D to evaluate the proposed method. Experimental results demonstrate that TA-MDet outperforms all compared single-modal and multimodal methods on UAV-LC3D, validating the effectiveness of the proposed approach.

Original languageEnglish
Title of host publication2026 China Aerospace Information Technology Conference, CAIT 2026
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9798319510389
DOIs
StatePublished - 2026
Externally publishedYes
Event2026 China Aerospace Information Technology Conference, CAIT 2026 - Tongxiang, China
Duration: 8 May 202610 May 2026

Publication series

Name2026 China Aerospace Information Technology Conference, CAIT 2026

Conference

Conference2026 China Aerospace Information Technology Conference, CAIT 2026
Country/TerritoryChina
CityTongxiang
Period8/05/2610/05/26

Keywords

  • 3D object detection
  • Multimodal method
  • Terrain-Aware
  • UAV platform

Fingerprint

Dive into the research topics of 'TA-MDet: Terrain-Aware Multimodal 3D Object Detection for UAV Platforms'. Together they form a unique fingerprint.

Cite this