基於深度學習影像辨識之兩輪平衡車追隨系統

No Thumbnail Available

Date

2020

Journal Title

Journal ISSN

Volume Title

Publisher

Abstract

為了達到人類搬運行李及短程移動時省力的效果,本論文提出兩輪平衡車架構的電子輔具,且基於Robot Operating System (ROS)之軟體架構下進行開發,具備四種功能模式提供使用者做使用。該平衡車使用模糊理論來達到自主平衡的控制,由於模糊控制器的設計並不需要運動學的模型,僅需要使用專家經驗並由機器的實際輸出表現來微調參數即可,所以在控制器的設計相當簡便且能達到良好的控制表現。本論文設計了四種功能模式,分別為手拉車模式、遙控模式、載人模式和追隨模式,其中以手拉車模式和追隨模式為主要的輔具功能。手拉車模式為兩輪平衡車上方安裝手把來供使用者進行行李車的拖拉,手把上有一搖桿用來控制拖拉車的轉彎,可以用來當作省力的行李搬運車;追隨模式則是使用單眼攝影機結合深度學習來辨識目標物並且追隨,而深度學習影像辨識的模型是使用Single shot multi-box detector (SSD)。最後,透過實驗來驗證兩輪平衡車成為電子輔具可行性。
The purpose of this thesis is to achieve labor saving for humans when carrying luggage and short-range movements. This thesis proposes a two-wheeled self-balancing mobile robot which is developed based on Robot Operating System (ROS). It has four modes for operators and utilize fuzzy theory to achieve autonomous balance control. Because the design of the fuzzy controller does not require the model of the mobile robot, it only needs to utilize the expert experience and fine-tune the parameters based on the actual output performance of the mobile robot. The controller design is quite simple and achieves good control performance. This thesis has designed four functional modes, namely handcart mode, remote mode, vehicle mode, and following mode. Among them, handcart mode and following mode are the main functions. The handcart mode is a handlebar installed above the two-wheeled self-balancing mobile robot for the operator to drag. There is a joystick on the handlebar to control the direction of moving. In this mode, the two-wheeled self-balancing mobile robot can be utilized as a labor-saving luggage carrier. The following mode utilizes a monocular camera with deep learning to identify and follow the operator. The model of deep learning image recognition is Single shot multi-box detector (SSD). Finally, the feasibility of the proposed following system using a two-wheeled self-balancing mobile robot is verified by some experiments.

Description

Keywords

兩輪平衡車, 機器人作業系統, 深度學習影像辨識, Two-wheeled self-balancing mobile robot, ROS, Deep learning image recognition

Citation

Collections