擴增實境技術應用於麻將辨識系統

麻將是一種源自中國的策略遊戲，一般分為13張(廣東)、16張(台灣)兩種玩法，也有其他較少見的變種玩法，但是麻將牌型複雜，各種胡牌的組合對於老手來說可以簡單辨認是否胡牌、聽什麼牌，但對於新手來說或許過於複雜，本研究結合影像辨識、人工智慧與演算法將麻將辨識結果顯示於智慧眼鏡上，輔助麻將玩家迅速判斷是否聽牌或胡牌。首先，本研究設計一Android APP運行於智慧眼鏡BT-300，此APP將麻將牌拍照存檔，並上傳至以Python撰寫之伺服器，伺服器以感興趣區域(region of interest)以及透視轉換、目標函式最小化裁切出各個麻將牌，再以TensorFlow運算辨識各個麻將牌，再使用演算法運算此組牌是否聽、胡牌，並顯示於APP。從實作結果可以看出使用者只需戴著智慧眼鏡即可得知自己的牌型是否聽、胡牌，減少人腦計算出錯以及耗時的問題，相當方便，且準確度可達80.9%，讓新手玩家快速上手，享受麻將的樂趣。

關鍵字

麻將； TensorFlow ； Python ；智慧眼鏡

並列摘要

Mahjong is a strategy game originating from China. It is generally divided into 13 (Guangdong) and 16 (Taiwan) gameplays. There are also other rare variants of the gameplay, but the mahjong tiles are complex, and the combination of various cards is for the veteran. In this case, you can easily identify whether you are playing cards or listening to cards, but it may be too complicated for newcomers. This study combines image recognition, artificial intelligence and algorithms to display mahjong identification results on smart glasses, and assists mahjong players to quickly determine whether they are listening or playing cards. First of all, this study designed an Android APP to run on the smart glasses BT-300. This app archives the mahjong tiles and uploads them to the server written in Python. The server uses ROI (region of interest) and perspective transformation, target function. Minimize the cutting of each mahjong card, then identify each mahjong card with TensorFlow operation, and then use the algorithm to calculate whether the group card is listening, Hu card, and displayed in APP. Users only need to wear smart glasses to know whether their cards are listening or not. It is very convenient for the novice players to get started quickly and enjoy the fun of mahjong.