請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/88633
標題: | 無人機風險地圖之建置與視距內操作之規劃 Establishment of UAV Risk Map and Planning for Line-of-Sight Operations |
作者: | 閻憲廷 Hsien-Ting Yen |
指導教授: | 周瑞仁 Jui-Jen Chou |
關鍵字: | 無人機,深度學習,視線分析,路徑規劃,U型卷積網路, Unmanned Aerial Vehicles,Deep Learning,Line of Sight Analysis,Path Planning,U-Net, |
出版年 : | 2023 |
學位: | 碩士 |
摘要: | 本研究旨在協助無人機操作人避免飛航場域風險,提供參考路徑及視距內的操作或觀察點。近年來,無人機使用量急速增加,但也帶來了一些負面影響,如噪音、隱私和人身安全疑慮。現行法規已針對無人機飛航制定規範,但操作人往往難以預判失去視線接觸或進入風險區域。本研究建置了一套系統由操作人輸入飛航任務需求與可接受之風險程度後,系統產出完整任務所需之參考資料,如路徑、操作點、目視觀察點等,操作人可在符合法規下完成任務。除了由系統建議之路徑、操作點、目視觀察點等飛航計畫參數外,操作人也能輸入參數由系統判斷操作人規畫是否在風險可承受範圍內及是否全程維持視線接觸。整體來說本研究主要分為三大部分,包含風險地圖之建置、路徑搜尋、及人機視線分析。風險地圖係以U-Net深度學習建置而成。以Python程式爬取便利商店的人潮資料,取得動態人流密度,結合禁限航區、建物等環境資訊套入前述風險地圖中。根據操作人設定之條件,系統之路徑搜尋演算法自動生成避開風險之參考路徑,人機視線分析輔助操作團隊找到能與無人機保持視線接觸之操作人或目視觀察員之位置與視野。分析生成結果證明,建置之風險地圖反映了實際場域中的風險資訊。系統能夠有效生成飛航路徑避開高風險及非法操作區域。同時本系統建議之飛航任務能夠在操作團隊與無人機保持直接目視接觸下完成。 The aim of this research is to assist drone operators in avoiding the risks associated with operation location and provide reference paths along with ground control station and visual observer location which is within visual line of sight. In recent years, the use of unmanned aerial vehicles (UAVs) has increased, which raised many concerns such as noise, privacy issues, and potential safety risks. Regulations have been implemented for UAV flight operations but still, it is challenging for operators to anticipate where they may lose visual contact with the UAV or entering hazardous areas. Therefore, a system has been developed in this study, allowing operators to input basic mission parameters to generate comprehensive information for the mission, including paths, operational points, and visual observation location based on the operators' input which enables operators to complete their missions in accordance with regulations. In addition to the suggested parameters provided by the system, operators can also use the system to determine whether manually designed parameters fall within an acceptable risk range and ensure that visual line of sight can be maintained throughout the entire mission. Overall, this research is mainly divided into three major parts, including the risk map, path planning, and human-drone visual line analysis. The risk map is built using the U-Net architecture, which is a deep learning algorithm. The crowd density in convenience stores is collected through data scraping to estimate dynamic pedestrian density. This data is then combined with environmental information such as no-fly zones and buildings to be integrated into the aforementioned risk map. Based on the operator’s input, the system automatically generates reference paths to avoid risks with the path planning algorithm. The human-drone visual line analysis assists the operating team in identifying ground control station and visual observer location to maintain visual contact with the UAV.Analysis of the results demonstrates that the constructed risk map reflects actual risks in the field. The developed system effectively generates flight paths that can avoid high-risk and restricted location. Furthermore, the flight missions suggested by the system can be successfully completed with maintained visual contact between the operation team and the UAV throughout the mission. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/88633 |
DOI: | 10.6342/NTU202302669 |
全文授權: | 未授權 |
顯示於系所單位: | 生物機電工程學系 |
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 | 6 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。