地方辨識與定位,一直以來都是自主移動機器人研究領域,備受關注的議題之一。本論文提出一套非常接近人類思維的場景相關認知與判別方法-“場景變化偵測”。此研究主題想法來自人類在記憶路徑或指引他人路徑時,通常只採用少量的關鍵點資訊輔助。在本研究中我們僅以一台全方位攝影機,不含其他感測器,利用基於增強型凸包編碼與SURF特徵點擷取的方式,開發出經由分析增強型凸包編碼參數變化趨勢的場景變化偵測機制,使其能適用於室內、室外場景之變化偵測。一旦自主移動機器人行經一段未知路徑,即可使用此場景變化偵測機制標記未知路徑中的重要區域,並利用偵測出的節點訊息建構環境拓墣地圖。
Place recognition and location are important problems in mobile robotic research. In this thesis, we present a novel scene recognition technique that resembles human thinking-"scene change detection".Our semantic scene descriptor is based on SURF and Extend-HCT features.We build a scene change judgment system by analyzing parameter of Extend-HCT codewords.Then we use our algorithm to mark the interested regions and construct the topological map after the autonomous mobile robot passing through an unknown path.For a robot to have a topological map in an unknown environment is just like voyaging in the vast sea with compass.In the future, if a robot goes through this environment again, it can use the topological map which we built before, and assist the robot for place recognition, location and navigation tasks.