Please use this identifier to cite or link to this item:
標題: Design of Intelligent Vision and Hearing Systems for Robots
作者: 陶金旭
關鍵字: 應用研究;電子電機工程類, 資訊科學--軟體
In this project, the vision and hearing systems of an agile human symbiotic service robot will be designed based on computational intelligence algorithms. The detection of humans and objects is an important and indispensable preprocessing step; and the face detection and recognition are crucial functions for service providing and surveillance applications. With gesture recognition and voice control systems, robots can receive correct instructions in noisy environments. Therefore, they provide robust human machine interface. The major application areas include commercial service (e.g. reception personnel), recreations (e.g. robot caddie), home service, and taking care of elderly citizens. For the vision system, stereo images based on two cameras will be adopted so that the target detection can be more effective. In the face recognition system, the matching results from the two-dimensional template and the three-dimensional feature points are combined in order to achieve a higher recognition rate. For speech and gesture recognition, feed-forward and recurrent TS-type fuzzy classifiers based on the support vector training algorithm will be developed to improve the accuracy and reduce the computational load. The vision and the hearing systems will be integrated with the systems from other projects on a PC-based platform.

本子計畫主要以計算智慧(computational intelligence)為基礎,設計與人共生半人型機器人之視覺與聽覺系統。其中機器人視覺是機器人對環境的認知以及與人互動中很重要的一環。而人體與目標物的偵測是一重要的前處理步驟,人臉的偵測與辨識則是對於提供服務以及環境監控等應用所必須具備的基本功能。動態手勢辨識與語音控制則著重在噪音環境下的使用者控制命令辨識,可以提供適當的人機溝通介面,使機器人能正確的接受使用者的指令。其主要應用領域有商用服務(如接待與導覽)、休閒娛樂(高爾夫桿弟機器人,餐飲機器人)、居家服務機器人(家務處理,交談)以及老年照護機器人(照護保健,事件提醒)。在視覺系統方面,我們將採用雙攝影機的架構,使得目標物偵測更為有效。在人臉辨識系統中,我們將結合二維樣版與特徵點三維資訊比對的結果,以提高辨識率。在手勢與語音辨識系統中,我們將發展以支持向量所訓練的模糊分類器,來提高運算速度與準確性。最後我們將把所發展的系統與其他子計畫成果整合至以個人電腦為基礎的平台上。
其他識別: NSC97-2221-E005-072-MY3
Appears in Collections:通訊工程研究所

Show full item record

Google ScholarTM


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.