Hand Posture Recognition Using Adaptive Skin Color Detection and Shape Matching

手勢辨識為目前人機介面的熱門研究主題，由於手勢的方便性及其直覺性，近年來更有車廠想將手勢應用於操控車上的資訊娛樂系統，讓駕駛者可以專心的將視線放於路面上，能夠安全地一邊開車一邊操控車上的手機通話、CD播放器、調整電台等等。 本論文提出以自適應性膚色偵測自動調整使用者的膚色值域，並由此得到使用者的手部影像，接著計算其輪廓匹配程度。我們定義了10種台灣數字手勢，能夠即時的辨識，可用於一般的靜態手勢辨識應用，或是作為動態手勢的基礎，並且整體平均辨識率高達9成，我們發現相較於先前文獻的方法大大提升其便利性，但效能和辨識率卻未受明顯地影響，並且辨識的手勢集更為完整，以及手掌的正反面入鏡、甚至是使用左手皆可準確的辨識。Hand gesture recognition has become popular in the field of human-computer interaction research. Because of the convenience and intuitiveness of hand gestures, automobile companies have applied hand gestures in their infotainment systems in order to eliminate driver distractions while on the road. This thesis proposes a hand gesture recognition method using adaptive skin-color detection for automatic skin color thresholds in order to obtain accurate segmentations of the hand contours, from which similarity of the contour shapes can be calculated. Hand gestures for 10 numbers in Taiwanese sign language are defined and are detected in real-time. Our system can be used for generally static hand gesture recognition applications, or can be extended for dynamic hand gesture recognitions. The overall average recognition rate for our system is 90%, and our system is more efficient in comparison to previous work while accuracy is unaffected. In addition, the dictionary of our hand gestures is relatively more complete, and users can even use their left hands to get the same satisfactory recognition results