請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/87907
標題: | 基於視覺-慣性-超寬頻融合技術的室內絕對位姿估計 Absolute Pose Regression Using Visual-IMU-UWB Fusion for Indoor Scenes |
作者: | 俞正楠 Zheng-Nan Yu |
指導教授: | 洪一平 Yi-Ping Hung |
關鍵字: | 相機定位,數據融合,特徵選擇,絕對位姿估計,超寬帶, Camera Localization,Data Fusion,Feature Selection,Absolute Pose Regression,Ultra-wideband, |
出版年 : | 2023 |
學位: | 碩士 |
摘要: | 位姿估計在導航、虛擬/增強現實、機器人、軍事等領域有著廣泛的應用,因此實現高精度的位姿估計十分重要。但由於環境複雜、噪聲幹擾大,位姿的準確估計難度較大。為了提高位姿估計系統的性能,本文搭建了一套全新的端對端位姿估計系統,集成了光學傳感器、慣性傳感器和超寬帶傳感器,並提出了一種新型位姿估計框架充分利用了不同傳感器的優勢。該位姿估計系統由視覺傳感器、慣性傳感器和超寬帶傳感器組成,通過融合多模態的測量數據實現高精度的位姿估計。本文所提出的模型能夠基於原始傳感器的數據準確估計物體位置和方向。整個模型由視覺編碼器、慣性編碼器、超寬帶編碼器和特徵融合網絡組成。實驗結果表明本文的方法在實現較高精度的同時節省了訓練時間。 The accurate pose regression of the target is vitally important, which has a wide range of applications in navigation, virtual/augmented reality, robots, and the military. However, due to the complicated environment and noise, the accurate pose regression of the target is difficult. To improve the performance of the pose regression system, we integrate optical sensors, inertial sensors, and ultra-wideband sensors, and propose a novel end-to-end pose regression framework which makes use of advantages and makes up for disadvantages. The pose regression system, which consists of visual, inertial, and ultra-wideband sensors, achieves high-precision performance through the fusion of multi-modal sensor data. The proposed model takes raw sensor data and generates estimated 6DoF results, positions and orientations. The whole model is made up of the visual encoder, the inertial encoder, the UWB encoder, and the fusion network. Experimental results demonstrate that our method shows excellent precision and saves training time. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/87907 |
DOI: | 10.6342/NTU202300994 |
全文授權: | 未授權 |
顯示於系所單位: | 資訊網路與多媒體研究所 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-111-2.pdf 目前未授權公開取用 | 1.97 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。