人臉辨識及其在家用機器人與人互動之應用

Abstract

本論文發展一套應用於家用機器人之家庭成員辨識系統。文中提出一適用於特定人搜尋之人臉辨識系統，此系統提供每個家庭成員各自擁有自己的RBF類神經網路，各個類神經網路具備分辨該所屬的家庭成員即可，不需去判斷其它家庭成員，使得辨識時間不會隨著家庭成員的增加而增加。本論文亦完成一以膚色區域分割為基礎的多人臉偵測，使系統能同時對影像中的多張人臉進行辨識。我們將家庭成員辨識系統整合影像追蹤系統於機器人進行實際測試。整合測試之實驗結果顯示本系統具有即時人臉辨識及追蹤之功能。

In the thesis, a family-member recognition and image tracking system has been developed for a home robot. We proposed a new architecture for face recognition of specific people. In this system each family member has his own RBF neural network. Each neural network only needs to recognize its own member. It does not need to judge other family members. Consequently, the recognition time will not increase as the family member grows. This thesis also develops an algorithm for the detections of multi-face based on division of skin regions. The family-member recognition system and image tracking system has been integrated to a home robot. The experimental results show that the proposed methods have real-time face recognition and tracking performance.