為了準確記錄人體動作,近年來已發展出機械式、電磁式、光學式等動作捕捉技術,供醫療、娛樂、訓練、智慧家庭等多元用途。其中光學式深度感測產品價格低廉,但因過多雜訊與自我遮蔽等因素,動作精確度有明顯的不足。為了達到自然人機互動,本研究以改善深度感測動作捕捉技術為目標,目的是降低使用者因自我遮蔽、突然改變運動方向時造成的動作雜訊,產生流暢的骨架動作。此一研究採用Kinect感測裝置取得骨架資料,依據關節運動軌跡及人體關節活動度為基準,對各關節軌跡作雜訊偵測。接著,利用前後骨架資訊預測新的骨架位置,並對新骨架作平滑化處理。最後,將此新骨架與原骨架作比較,以驗證其成效。
In recent years, many human motion capture technologies, such as mechanical, electromagnetic and optical sensing technologies, have been developed for recording human motion accurately for medical, entertainment, training, smart family and other applications. The optical depth sensing devices are cheap but the accuracy of motion is obviously insufficient due to excessive noise and self-occlusion factors. The purpose in this thesis is to reduce the motion noise caused by self-occlusion and sudden change of motion direction to generate smooth skeleton motion. This work gets the skeleton data from Kinect and detects noise of all joints by evaluating of motion trajectories and joint angles. Then, predict the new skeleton from adjacent non-noise data and smooth all joint trajectories. Compare the new skeleton with the original skeleton to verify the effectiveness of proposed method.