基於ROS之智慧安防自主巡邏履帶式機器人系統

Abstract

本論文主要將深度感測器與自主式履帶機器人整合，並針對障礙物偵測與人體動作辨識這兩方面各自提出一種系統。在障礙物偵測系統中，運用深度影像使機器人能夠偵測前方空間中的障礙物，並結合模糊控制器控制機器人安全避開。在人體動作辨識系統中，藉由Kinect v2攝影機取得人體骨架，並透過事先訓練好的模糊類神經網路進行即時動作辨識，以觀察是否作出危險動作。除了以上兩種系統外，還增加監控系統的使用者介面，並透過3台Mesh架構的路由器來跟履帶式機器人相互溝通，以此來傳遞影像資訊、地圖位置、任務要求、顯示警示燈等功能。

This thesis focuses on the integration of a depth sensor with an autonomous tracked robot, and proposes a system for both obstacle detection and human movement recognition. In the obstacle detection system, depth image is used to enable the robot to detect obstacles in the space ahead and to control the robot to avoid them safely in conjunction with a fuzzy controller. In the human movement recognition system, the human skeleton is captured by a Kinect v2 camera and a pre-trained fuzzy neural network is used to perform real-time motion recognition to see if a dangerous action is taken. In addition to these two systems, a user interface is added to the monitor system, and three mesh-based routers are used to communicate with the tracked robots to transmit video information, map locations, task requirements, display warning lights and other functions.