應用於遠端監控之無線網路遠端互動系統之設計

Abstract

本論文所開發的遠端互動系統（RIS），是一個可提供使用者透過無線網路針對特定區域進行遠端監視或溝通的監控系統。而本論文的主要目標是設計出可以安裝在子成員上，並連線至多媒體伺服器(Flash Media Server) 的使用者導向應用軟體，包含視訊通話(Video Calls App) 以及人臉偵測(Human Face Detection App)。由於多媒體伺服器可透過無線網路連結多個子成員，並且透過RTMFP協定來提供子成員間快速的多媒體傳輸。此外RIS藉由人臉偵測功能可讓使用者查看他們的孩童或者與其進行遠端溝通，而為了提升人臉偵測的準確率，本論文開發了基於類神經網路的AdaBoost Cascade 分類演算法，將Cascade AdaBoost分類演算法的結果進行二值化後，取出其特徵，再利用類神經網路進行人臉偵測。最後，運用SAMSUNG Galaxy SII 的智慧型手機和Acer ICONIA 的平板電腦，來執行視訊通話以及人臉偵測的應用軟體，根據實驗結果，人臉偵測的準確率可以達到97.23%，而人臉偵測的傳輸速率可以達到約每秒12張影像。

This thesis develops a monitoring system, called the Remote Interaction System (RIS), for users to communicate or watch a specific location remotely via the wireless Internet. The proposed RIS adopts Flash Media Server which has the advantage of containing multiple sub-members (Kinder-Robot, webcams, clients) through the wireless Internet and providing fast multimedia transmission through RTMFP protocol among the sub-members. The main objective is to design client-oriented RIS Apps, such as Video Calls App and Human Face Detection App, installed on the sub-members to connect to the Flash Media Server. With the function of human face detection, the clients are allowed to watch and communicate with their children remotely. To increase the accuracy of human face detection, the NN-Based AdaBoost Cascade Classification is developed to process the binary features related to the result obtained from AdaBoost Cascade Classification. To demonstrate the performance of RIS system, the RIS Apps are executed on SAMSUNG Galaxy SII phones and an Acer ICONIA Tab. Under the frame rate of 12 frames per second, the accuracy rate of human face detection is 97.23%, much higher than 93.7% of the conventional AdaBoost Cascade Classification.