本論文主要的目的是在設計一個易於使用且可大量生產的低成本手型辨識系統，以便於使用者在不需要配戴任何裝備的情況下仍然能對硬體下達簡易的命令以達到控制的目的。運用本系統可以識別出使用者以左手背指示出的手勢，目前已經可以辨識出0到9的單一數字，未來只要搭配相映的操作規則即可成為能辨識百位或更多位的數字辨識系統。本系統的應用範圍很廣，包括互動式玩具、電視選台器、各式家電的簡易控制等。 單純就圖形識別而言適用的演算法相當繁多，不管是在事前的特徵萃取或是其後的類別辨識上，都有相當多的解決方案。但是若在硬體資源限定在低成本(low cost)且必須做到即時(real time)辨識的目的時，能用的演算法就相對的稀少了。我們將提出的手型識別演算法概分為四大部分。首先，必須先設定好sensor的運作模式(此一部分的設定值必須燒錄於電子式可清除程式化唯讀記憶體可複寫的記憶體(EEPROM)中，於開機時自動設定)，接著將再經由事先設定好，適合本系統光感測器(sensor)的YUV範圍來過濾出膚色，取得二元的影像。接著使用一些自然手勢中的特性來找出手掌的掌心以及手掌半徑，並且將手指的資訊正規化到一16 x 8的矩陣中。最後，再使用校正矩陣將影像資料稍微修正後即可將最後的資訊送入模糊(fuzzy)推論系統中。以推論規則來判定是何種手勢，再將結果輸出至七段顯示器來告知使用者。 The main purpose of this thesis is the design of a low cost hand gesture system to allow users control over any given hardware through simple commands without the need of any additional input devices. The system is capable of recognizing gestures made by the user using the back of their left hand, currently supported gestures are numerals 0 through 9. The system can be used in a wide range of applications such as interactive games, television channel selectors and simple control over electrical home appliances. The methods available to the field of pattern recognition are numerous, there are many different ways of handling the initial sampling and the following classification stages. However, the range of methods shrinks considerably under the constraints of low cost hardware and real-time recognition. The hand gesture recognition system we propose here can be divided into four stages. First is setting up the operating mode of the image sensor (the relevant settings are saved into EEPROM and are written into the sensor on boot-up). Secondly these settings are used to generate a binary image from the raw image sensor data, using YUV thresholds to discriminate between skin and background colors. Now several properties of hand gestures are used to find the center and size of the palm, after which the hand image is normalized into a 16 by 8 grid. Finally, after making some last corrections to the grid data, it is send into a fuzzy logic system that determines the type of hand gesture made. The results are shown to the user on a seven segment display.
