請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/71660
標題: | 基於光達與相機融合之三維語義地圖之靈巧運動機器人 Agile Movement Mobile Robot under 3D Semantic Map Built by LiDAR and Camera Fusion |
作者: | Xiao-Yue Xu 徐瀟越 |
指導教授: | 傅立成(Li-Chen Fu) |
關鍵字: | 全向性機器人,語義地圖,激光圖像融合,社交導航, Omnidirectional robot,semantic map,LiDAR image fusion,social navigation, |
出版年 : | 2021 |
學位: | 碩士 |
摘要: | 本文針對全向性移動機器人平台,提出了一種在室內場景中,基於光達(LiDAR)和RGB相機融合的語義地圖系統。目的有二,基於語義地圖,機器人可以感知環境中物件位置,可執行基於物件語義的導航任務;其次,基於LiDAR良好的偵測能力,可以結合行人躲避算法和靜態障礙物躲避算法,實現更人性化的導航規劃。為了實現這兩個目標,機器人需要配置多線LiDAR和RGB相機,並對LiDAR和相機進行校準配准。之後,通過RGB相機進行物件偵測,將偵測結果附加到對應點雲上,從而得到具有語義資訊的3D點雲地圖。富含語義資訊的語義地圖建立完成後,將被壓縮成八叉樹格式,大大減小地圖文件存儲空間,提高索引速度。在通過語音發出導航指令時,通過構建導航決策模型,機器人將根據周圍人和靜態障礙物的關係,選擇適宜的行進路線進行躲避。 此方法克服了先前方法的缺點,和常規2D路徑規劃相比,通過構建3D地圖,我們不僅可以躲避桌子等下方為空的物件,並且具有地圖具有語義資訊,可以實現語義導航。此外,與之前的研究相比,藉助LiDAR360度的檢測能力,我們能識別到周圍行人的位置和速度,充分預測行人的未來動作,不僅提升了導航安全性,更可提高在居家環境中躲避行人的社交友好程度。 This thesis proposes an object-aware semantic mapping of indoor scenes with LiDAR and camera fusion for an omnidirectional robot. The aim is not only to perceive the position of the objects for semantic navigation task, but also to achieve a more user-friendly navigation planning based on LiDAR, which can simultaneously avoid pedestrians and static obstacles. In order to achieve the goals, the robot needs to equip and calibrate a multi-layer LiDAR and a RGB camera. The RGB camera is used for object detection, and the detection result will be attached to the corresponding point cloud, thereby obtaining a 3D point cloud map with semantic information. After the creation of a semantic map, it will be further compressed into an octree format, which greatly reduces the storage space of the map file and improves the indexing speed. When the robot is instructed to navigate, through a navigation decision model, the robot will choose a suitable route alter evaluating the geometric the relationship among the free space, the surrounding people and the static obstacles. This method overcomes the shortcomings of the previous methods. For example, the methods of constructing dense 3D semantic maps, mainly focus on the accuracy of 3D reconstruction, so they need high-precision ray, 128-layer LiDAR to obtain a very dense space map, which is unrealistic to use for navigation; On the other hand, unlike conventional 2D path planning, by constructing a 3D map, we can not only avoid the objects like tables, but also have the information to achieve semantic navigation. In addition, with the help of 360-degree lidar detection capabilities, we can identify the locations and speeds of surrounding pedestrians and fully predict these future actions. This not only improves navigation safety, but also improves the social acceptability to pedestrians in the home environment. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/71660 |
DOI: | 10.6342/NTU202100745 |
全文授權: | 有償授權 |
顯示於系所單位: | 電機工程學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
U0001-1802202116260700.pdf 目前未授權公開取用 | 4.63 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。