隨著近幾年來浮空手勢操作的發展,人們逐漸從傳統鍵盤與滑鼠 的操作介面,轉變為更符合人類的直覺操作模式,如:手勢操作。本 論文提出一個基於三維深度的手勢辨識與追蹤演算法,本系統使用低 成本的雙攝影機來計算深度影像,不但可以提供手勢的深度資訊,也 能在嚴峻複雜的背景中正常運作。 目前大部分的手勢偵測演算法採用膚色或運動量值做為前處理 步驟,但只透過膚色濾除和運動量無法在含有相近顏色背景下維持此 系統的功能性,本論文提出一個適應性的膚色深度過濾,此方法可以 有效分離出系統需要的手部區塊,也能改善追蹤演算法的成效。最後 透過深度資訊完成深度動態手勢辨識,經過多位使用者測試,手勢方 向移動功能準確率 93.7%,深度推拉功能準確率 95.6%,手勢旋轉功 能準確率 94.5%,動態手勢準確率 85.92%。;Accompany with mid-air control system have been developed in recent years, people gradually change their usage from tradition keyboard and mouse to the intuitive manner, like hand gesture control. This thesis proposed a hand recognition and tracking with depth information. We use stereo camera to capture stereo image and calculate depth map. The system not only can provide depth information but also can work under critical backgrounds. Most methods of hand detection apply skin filter or motion filter as one of pre-processing. However, only applying skin filter or motion filter as segmentation step can’t maintain system function correct while background pixels are close to skin color. In the proposed, we adopt adaptive depth filter which can separate foreground which improve performance on tracking algorithm. We also proposed dynamic gesture recognition by using depth data. Our accuracy of direction function is 93.7%, accuracy of push/pull function is 95.6%, accuracy of rotation function is 93.7%, accuracy of dynamic function is 85.92%.