本論文製作一個基於手勢辨識技術之穿戴式裝置，並將其應用於互動式的手語學習。在穿戴式裝置上，將彎曲感測器(flex sensor)與三軸加速度計裝設在手套上，用來量測手指彎曲角度與手部傾斜方向，然後運用微控制器DE0-Nano進行手勢偵測與辨識。在手勢處於動態變化或靜止的偵測上，本文提出一通道邊界機制，更新適應性卡爾曼濾波器參數，過程雜訊共變異矩陣與觀測雜訊共變異矩陣，的數值以達到穩定的狀態追蹤與濾波效果。然後利用所研擬轉折點與變化幅度的判斷方法，偵測出手勢由動態變化轉換為靜止的時機點，以此作為後續手勢辨識的起始點。在手勢辨識部分，首先利用決策樹將不同手勢進行初步分類，接著再由機率類神經網路運算得到辨識結果，此結果經由藍牙傳送至智慧型行動裝置以手語遊戲的方式呈現。實驗結果顯示所研擬演算法的可行性，並且降低了辨識時的計算負擔。 This thesis develops a wearable devices on the basis of a hand gesture recognition technology, and applies it to an interactive sign language learning. Both flex sensors and a three-axis accelerometer of which mounted on the glove as an important part of the wearable device are used to measure the angles of the fingers and the direction of hand. The microcontroller DE0-Nano operates with the measurements to perform state detection and gesture recognition of the hand. For accurate tracking and detection of the state of the hand, the adaptive Kalman filter plays an important role. We develop a new method to update the parameters process noise covariance and measurement noise covariance to ensure good tracking and noise filtering. A criterion operating on this data that utilizes the concepts of turning points and amplitude variation is developed to determine the time instant after which it is better suited for hand gesture recognition. As the first step the proposed hand gesture recognition method utilizes the concept of decision tree to give a preliminary classification of the filtered measurements of different gestures. Then a probability neural network processes the data to produce a final result. The result is transmitted via Bluetooth to a smart mobile device and displayed in a proposed sign language game. Experiment shows the validity of the proposed algorithm; furthermore, the computational burden during recognition is reduced.