人機互動、智能生活是現行電腦科學研究領域中十分熱門的議題，而人體動作辨識更是被認為達到這些目標的重要關鍵。而在微軟推出Kinect之後，有越來越多的研究單位選擇使用較為低成本且普及的Kinect來做姿態辨識。本論文透過結合無線感測器改善Kinect 先天性的偵測弱點，針對當部分肢體平貼於地面或有肢體重疊的姿態進行姿態辨識。 本論文將對使用Kinect、無線感測器、以及Kinect結合感測器等三套系統進行比較當使用不同裝置於姿態辨識的差異，其姿態辨識率、任務完成時間、使用者感受等。 使用Kinect搭配OpenNI結合無線感測器所產生的人體骨架的三維座標(x，y，z)值，投影到Unity 3D 遊戲場景中，利用3D骨架呈現人體姿態，而其骨架資訊經過正規化後，使用自我組織特徵映射網路(SOM)進行分群，再與正確樣本進行即時性(real-time)的比對。 本論文提出的Kinect體感技術結合無線感測器模組輔助系統在姿態辨識上有平均94.6%的平均辨識率，明顯比其他兩套系統來的高，可以得出其系統整合了兩種體感技術於姿態辨識的優點。本系統在辨識複雜度高的姿勢動作時，也能有平均94.6%的成功辨識率。 In this study, we use wireless sensors combined with Kinect to improve the detection weakness of Kinect which is some parts of human bodies flat on the ground or gesture overlapping. In human gesture recognition, we compare Kinect system, sensor system, and Kinect combined sensor system those three systems in four ways: devices recognize difference, gesture recognition rate, task completion time, user’s satisfaction.
Three-dimensional coordinate (x, y, z) values, which is 3D human skeletons’ data using Kinect with OpenNI combining with wireless sensors, projected to Unity 3D game scene and it shows result and compares with the correct sample by using self-organizing map network (SOM) in real time. This study proposes a combination of wireless sensor and Kinect system in human gesture recognition with an average of 94.6% on the average recognition rate.
