隨著電子化產品的普及，數位家庭生活正逐漸改變我們傳統的生活習慣，而其中各種人機介面技術不斷地被推陳出新，不但提供了使用者與系統之間進行溝通的管道，也讓使用者能夠更簡單、更迅速的完成操作。近年來，以人類自然肢體動作型態為基礎的輸入裝置，由於其符合人性化、直覺操控等特性，已經成為人機互動技術發展的主流；一些以人體動作或手部姿勢動作辨識為主的產品也廣受消費者的歡迎，如任天堂公司的Wii遊戲機、Apple公司推出的iPhone及Microsoft的Kinect等，不同類型的感測器與應用方式無所不在的滲透到我們的生活中，為我們創造出更多元、更便利的生活環境。本研究利用加速度感測器偵測手勢動作在三維空間中的加速度變化，首先以牛頓第二定律計算其速度、位移等資訊，再將之投影至低維度空間以得到手勢特徵向量；當特徵向量經過正規化與內插處理後，便可輸入PNN機率神經網路進行手勢辨識的處理。本研究所述之辨識架構不但不需要大量的訓練樣本數，而且學習速度快速，對於誤差的資訊也具有相當的容忍性與高度的辨識率。為了達到微型化與實用化的目的，我們也在實驗中將系統實現於以ARM 32-bit Cortex-M3 核心晶片為基礎的嵌入式平台上進行測試與驗證，實驗結果顯示我們所提的方法確實能符合即時處理的功能要求，而且架構簡單有效，適合用於開發具直覺特性的智慧型人機互動系統。 In past few decades, electronic products have become a part of the building blocks of modern society; lifestyles of humans are changing progressively along with the development of digitized everyday objects. New types of human-machine interaction technologies are constantly introduced by employing different kinds of sensor technologies. The evolvement of human-machine interaction technologies makes us to be able to communicate with machines in natural and simple manners. In recently years, input devices which are based on motion sensing technologies have become the main stream of the development of human-machine interaction area. Products based on body motion or hand gesture, such as Wii (Nintendo), iPhone (Apple Computer) and Kinect (Microsoft), have achieved great success on the consumer market. By utilizing various kinds of sensing technologies in everyday objects, a new, convenient, and creative lifestyle is achievable in the near future. In this thesis, we use accelerometers to determine the variation of the acceleration of hand gestures in three-dimensional space. We use Newton's second law of motion to process the velocity and displacement data of a gesture and projected it to a reduced dimensional space to get the feature vectors of a gesture. And, we use normalize and interpolated feature vectors as input of a PNN to recognize the input gesture. The recognition framework we used in this thesis needs only few samples and time for learning; and with high toleration of data deviation, the identification rate of this recognition framework is high. For practical use and small size purpose, we use an embedded system platform based on ARM 32bit Cortex-M3 to implement the recognition framework. Experimental result shows that the method we proposed is simple, effective, capable of doing the real-time process of the gesture recognition, and suitable for the development of intuitional intelligent human-machine interaction systems.
