本論文主要開發一具人機介面、硬體架構及智慧型控制器的智慧型輪椅。本輪椅提供使用者一個即時的人臉辨識及人臉控制輪椅之系統,並提出一個以模糊和影像為基底的自動充電方法。首先,基於電動輪椅本身硬體架構上,利用FPGA對輪椅控制器以控制輪椅的電壓以達到輪椅控制的目的,並架設各種感測器,藉以實現能夠進行多方面控制之系統。 在使用者辨識方面,利用特徵點尺度不變的特點作為人臉辨識的依據,以臉部特徵點進行臉部辨識,達到辨識輪椅使用者身分的目的。在人臉控制方面,本文使用色彩分析的方法,因此,統計樣本膚色的色彩範圍,並以此範圍內的膚色做為人臉偵測及定位的基準。為避免在不同光線下易造成誤判,本文以自適應閥值過濾眼睛特徵,並利用眼睛及嘴唇位置的幾何關係確定是否為正確的人臉特徵,最後定義出一條人臉位置基準線,判斷基準線的左右傾斜率以達到人臉控制的目的。 在輪椅的自動充電方面,我們提出了一個基於影像處理的新式自動充電方法,利用隨機圓偵測及橢圓偵測,我們可以從圓的重心得到充電位置及輪椅位置,經由這些資訊,輪椅可以自動校正角度及速度。除此之外,電子羅盤和超音波感測器被架設在輪椅上,以提供輪椅和充電位置的相關角度及距離。另外,我們採用模糊控制器控制輪椅,讓在調整輪椅角度及速度方面可以更精確,最後實驗的結果展現出我們提出的方法和機構是有效及可行的。
In this thesis, an intelligent wheelchair including the human-machine interface, the hardware architecture and the intelligent controller is implemented. The wheelchair provides the user a real-time face recognition and control system. Besides, a fuzzy based automatic charging method is proposed. Firstly, based on the primal hardware architecture, the FPGA technology is adopted to control the voltage of the wheelchair for reaching the control object and some sensors are installed for achieving the multiple control methods. For user recognition, by utilizing the scale invariant character of the feature point for the face recognition basis and through the character point of the face, the face recognition is processed for achieving the object of user recognition. In the face control, the color analysis method is adopted. Therefore, the color ranges of statistical samples for skin color are gathered and the range is defined as the skin color range for the bases of the face detection and face localization. In order to avoid the misjudging under the different brightness, the self-adaptive threshold is propounded to filter the eyes and lips characters. In addition, the geometric position relation is used to examine the correct face features. Finally a baseline of the face position is defined to check the tilt rate of the baseline for achieving the purpose of the face control. For automatic charge of wheelchair, we propose a novel automatic charging method based on the image detection. By utilizing the random circle detection (RCD) and ellipse detection method, we can obtain the relative position of charging position and wheel robot from the center of a circle. From the information, the wheel robot can adjust the angle and speed automatically. Besides, the electrical compass and the ultrasonic sensor are installed in the wheel robot for obtaining the relative angle and distance of wheel robot and the charging position. Moreover, we adopt the fuzzy controller to control the wheelchair for adjusting the angle and speed of wheel robot more precisely. Finally, the experiment results show that the proposed approach and the mechanism are effective and feasibility.