模糊推論系統 於行動裝置上體感辨識之研究

Abstract

體感操作的應用於近年來開始蓬勃發展，藉由體感操作的直覺式應用是加速其發展 的主要原因。特別是在遊戲產業的發展中，Microsoft 推出的Xbox Kinect 更將體感操作 推向新的境界。透過Kinect 的3D 深度感應器利用光學跟圖像感應技術可以直接偵測使 用者活動，不需要藉由其他感應器輔助。雖然Kinect 辨識能力已經相當完善。但在推 廣與應用上能需要使用關鍵元件，例如3D 深度感應器。要在其他領域使用體感操作則 必須準備Xbox 主機與Kinect。如果使用目前持有率持續增加的智慧型手機作為體感操 作的感應器，可將應用範圍推廣。要在效能有限的行動裝置上實現體感辨識所需要的準 確度與流暢度，必須選擇使用較低資源消耗的方式，因此本研究在體感辨識上使用光子 流系統與模糊推論系統實作。 本研究以人體手勢偵測為例，使用模糊推論系統於行動裝置上實現體感辨識的方法。 在規劃體感辨識時，系統兩項主要核心功能的實現方式。第一項為人體特徵的判定，於 系統設計時選擇以降低系統資源需求的方式，以及因應隨時執行偵測的條件。第二項為 建立動作辨識之模糊系統，透過學習影像的方式建立適合的模糊規則庫達到可辨識手勢 的效果。

The application of motion control is increasing these years, and intuitive control based on body operation is the main reason for accelerating its development. Especially in development of game industry, Xbox Kinect launched by Microsoft further brings motion control to a new state, which can directly detect users’ activities through 3D depth sensor of Kinect combining with optical and image sensing technology, without other auxiliary sensors. Although recognition capability of Kinect has been quite perfect, it still needs key elements, such as 3D depth sensor. Therefore, Xbox and Kinect are essential if motion control is applied in other fields. Application scope can be expanded if smart phones with increasing holding rate are used as sensors of motion control. Low resource consumption shall be selected to realize accuracy and fluency of motion control in mobile device with limited efficiency. Therefore, this paper uses optical flow and fuzzy inference system for implementation. This paper takes body gesture as an example and uses the method that fuzzy inference system is applied to mobile device for motion recognition. There are two implementation approaches of main core functions of the system in motion recognition. The first one is adjustment of characteristics of human body, which shall be selected in system design by means of decreasing system resource requirements and available to performing detection at any time. The second one is to establish a fuzzy system of motion recognition, and establish proper fuzzy rule base through the mode of learning image so as to recognize gestures.