人機互動透過身份辨識和驗證應用於智慧型服務機器人

Abstract

隨著科技發展，機器人在我們的生活中扮演一個重要的角色，機器人會採取不同的行為，是取決於不同的辨識結果。因此，如何設計一個具有身分辨識功能的互動服務機器人成為很好的研究議題。本篇論文即針對身分辨識和智慧機器人進行研究。 本論文大致可分為兩大部分，前半部份會針對我們的性別辨識、人臉辨識、身高偵測演算法來說明，性別和人臉辨識都是使用影像辨識，使用多種影像特徵，採用類神經網路或支持學習向量機來學習分類，最後也使用整體學習的架構提升整體效果。其中，針對性別辨識，採用的資料庫為FERET的人臉資料庫，我們改變許多參數像是：不同影像大小、不同影像特徵、不同整體學習架構，對各種參數的改變做比較。除此之外，我們另外蒐集亞洲的人臉樣本，將我們的性別和人臉辨識演算法實現在我們的機器人上，搭配雷射測距儀偵測近似身高，完成身份辨識，結合Win32的語音系統，達到人機互動。 後半段對本實驗室自行開發的機器人做研究，以智慧型居家機器人為主，智慧型機器人要進入家庭，必須具備定位、避障、路徑規劃等功能，我們針對這些功能整合開發成一個完整系統。為了使用者操作方便，我們設計簡易的操作介面，並且搭配智慧型手機，讓使用者可容易自行操作。本論文所提出的軟體架構與設計模式都是以標準 C/C++/Java 程式語言，在Visual Studio和Eclipse軟體中進行開發。

As technologies advance, robot plays a more and more important role in our life. Many interactions with robots depend greatly on correct perception. Therefore, person recognition for interactive functions is an essential topic in terms of service robotics applications, and how to design interactive service robotics becomes a good issue. In this thesis, we conduct researches of person recognition and service robotics. There are two topics in this thesis. First, we introduce methodologies which we have developed, including gender recognition, face recognition, and height measurement. Not only gender recognition but also face recognition are classified by vision. In our research, our visual recognition adopts different kinds of image features, and the used classifiers are based on Neural Network or Support Vector Machine. Also, we develop ensemble learning methods in attempt to facilitate classification rates. We study the effects of various features, face image sizes, and ensemble learning methods on gender classification accuracies. Furthermore, we implement our visual recognition algorithms on our robots, and we use laser range finder to measure one’s height. Thus, Human-Robot interaction can be done through speech system (Microsoft Speech API) after identifying important differences that impact Human-Robot interaction. The second topic is about the introduction of the intelligent mobile service robots developed by the Intelligent Robotics and Automation (IRA) Laboratory at National Taiwan University. We break down our research efforts into four categories in this topic: (1) localization, (2) obstacle avoidance, (3) path planning, and (4) mobile phone control. Additionally, we integrate these functions and develop two simple interfaces towards friendly human-robot interactions. All the systems, user interface, software frameworks and applications proposed in this thesis are implemented with native C/C++/Java programming language, and the integrated development environments are Visual Studio and Eclipse.