Supervisor of Master's Candidates
Open Time:..
The Last Update Time:..
A novel real-time matching and pose reconstruction method for low-overlap agricultural UAV images with repetitive textures
Impact Factor:10.6
DOI number:10.1016/j.isprsjprs.2025.05.009
Affiliation of Author(s):LIESMARS, Wuhan University
Teaching and Research Group:航空航天摄影测量研究室
Journal:ISPRS Journal of Photogrammetry and Remote Sensing
Funded by:国家自然科学金、国际重点研发计划、湖北省自然科学基金
Abstract:Real-time processing of UAV-based agricultural remote sensing images is essential for precision agriculture, especially for low-overlap, large-format image sequence. These images, characterized by large formats, repetitive textures, and low overlap, pose significant challenges for real-time UAV photogrammetry. To address these challenges, this paper proposes a novel real-time matching and pose reconstruction method tailored to low-overlap agricultural UAV imagery. Firstly, an adaptive algorithm for robust map initialization is introduced, dynamically adjusting matching windows and thresholds to ensure successful map initialization in difficult conditions. Secondly, a robust and rapid feature extraction method is developed, integrating global texture information to improve feature repeatability and matching accuracy, especially in weak-textured regions. Thirdly, a multi-model adaptive tracking method is proposed, automatically selecting between four tracking modes to enhance 2D-3D matching point numbers and uniformity, ensuring robust tracking in complex environments. Lastly, a hybrid local–global dual-threading optimization strategy is implemented, combining local graph optimization with global loop closure detection, improving both accuracy and global consistency of pose reconstruction while maintaining high real-time efficiency. Experimental results show that the proposed system is the only one capable of real-time processing of low-overlap large-format UAV images, achieving a processing speed of approximately 3.4 frames per second, this performance is four times faster than the Photoscan (Fast mode) software with a reprojection error of within 1.1 pixels, meeting the real-time UAV photogrammetry requirements for precision agriculture.
Note:中科院一区TOP期刊
Co-author:Gui-Song Xia,Miaozhong Xu,Zhenfeng Shao,Jianya Gong,Deren Li
Indexed by:Journal paper
Correspondence Author:Wenhu Qu
Discipline:Engineering
Document Type:J
Volume:226
Page Number:54-75
Number of Words:12253
ISSN No.:1872-8235
Translation or Not:no
Date of Publication:2025-05-15
Included Journals:SCI
Links to published journals:https://doi.org/10.1016/j.isprsjprs.2025.05.009