非視覺型姿勢與動作辨識之研究常應用在身體復健、運動訓練以及影音娛樂等領域;然而這些應用將考量即時性與可穿戴性等需求。因此,本論文提出一套具即時性人體姿勢及動作辨識功能的系統,並強調可穿戴系統需具備之圖形介面功能與人性化設計。因此,本系統可作為一種運動訓練的輔具工具。本系統具有以下特點: (1) 可穿戴性:使用由布料製成的開關感測器與三軸加速度計兩種感測器。因為它們分別具有柔軟與體積小的特性,因此使用者穿戴時感覺具有舒適與輕巧性。此外透過ZigBee無線感測網路,本系統能即時支援5人同時使用以及遠距離無線操作。 (2) 即時辨識:姿勢與動作辨識部分透過查表方式,資訊包含比對加速度計與開關感測器狀態,能即時辨識出手部28種姿勢以及腳部14種靜態姿勢。更進一步,本系統也能完成動作辨識,如走路、抬腿、舉手等。 (3) 輔助運動: 將辨識結果與自行設計之互動遊戲結合,並藉由3D圖形使用者介面來呈現,讓使用者以遊戲的心態來從事各種運動活動。透過網際網路,協助人員藉由辨識結果能即時取得受訓者的活動狀態並給予意見。此外,我們的系統能夠透過動作相似度檢定能取得訓練動作執行的正確度並進一步取得動作執行次數。而次數、時間等資訊會自動透過資料庫儲存記錄供日後進度參考。
Research of the non-vision posture and motion recognition is often applied to such fields, rehabilitation, exercise, and video-audio entertainment. It will be considered the real-time and wearable characteristics. Therefore, in this study, we proposed the real-time human body posture and motion recognition system, which emphasizes the GUI function and the user-friendly design of the wearable system. It can be as one of the movement training auxiliary tools. Our system has the following characteristics as follows. (1) Wearable: Apply clip type sensor, which is made of the cotton material, and 3D-accelerometer in our design. Since these two sensors own soft and small size, users feel comfortable and light while wearing. Our system can simultaneously support 5 users with the real-time and tele-operator requirements by the ZigBee wireless sensor network. (2) Real-Time Recognition: In the posture and motion recognition part, our system can immediately recognize 28 hand postures and 14 leg postures by looking the table method, which contains the information of the clip type sensor and 3D-accelerometer. Moreover, our system can distinguish various motions, walks, raises legs, raise hands, and so on. (3) The auxiliary exercise: The recognition results and self design games can be integrated and presented by user's interface of the 3D graphics such that users can engage in the sport activities replaced by the game mode. Auxiliary experts can get users’ motions from our supported recognition system, and then give some suggestions to the wearable users by the internet network. Moreover, our system can examine and get the amount of the correct motion which is high degree similar to the standard designed one. Hence, the information, amount of the correct motion and activity time, can be automatically recorded to the database for progress in the future's reference.