本論文研製一個基於手勢輸入之即時通系統。此系統利用撓性感測器(flex sensor)與加速度計，搭配一組穿戴式機構，量測各手指的彎曲角度和手腕的傾斜方向，並將量測所得資訊以RS232傳輸協定傳至電腦端，經濾波後進行手勢辨識演算以產生相對應的文字。最後搭配TCP/IP協定，以自製之即時通介面與其他電腦進行互動。 在手勢辨識演算方面，本論文先採用一決策樹架構，將不同手勢進行初步分類，再由一改良式的機率類神經網路進行最後的手勢辨識。實驗顯示此方法降低了辨識時的計算負擔，提高了整體的辨識效率。 This thesis develops a messenger system based on hand gesture recognition technique. Flex sensors and an accelerometer combined with a wearable gesture sensing device are used to measure the bending angle of each finger and the wrist. The measurements are transmitted to PC by RS232 transmission protocol. After filtering to smooth the data, these data are processed through the gesture recognition algorithm to generate the corresponding text. Finally, the users can interact with the others by the homemade messenger system interface through TCP / IP protocol. The proposed gesture recognition algorithm consists of two major steps, which first utilizes the concept of decision tree to classify the filtered measured data of different gestures, and then performs more delicate gesture recognition by a probability neural network to reach a verification result. Experiment shows that an improved verification rate with reduced computational burden of the problem is obtained by the proposed method.
