臺灣大學: 電機工程學研究所傅立成杜明翰Tu, Ming-HanMing-HanTu2013-03-272018-07-062013-03-272018-07-062011http://ntur.lib.ntu.edu.tw//handle/246246/253954這篇論文描述了一個利用單相機人體三維手臂追蹤系統。為了使系統能夠應用於人機互動平台，本論文目的在於提出一個可達到系統即時性要求的演算法。 這裡利用了多重權重取樣粒子濾波器(MIS particle filter)結合影像中的多種線索，以追蹤手臂的任意動作。由於單相機影像缺少了深度資訊，我們基於從運動恢復結構(Structure-from-motion)，提出了一個連續姿態估計演算法，並作為粒子濾波器的三維線索。人體手臂在此定義為一個圓柱體組成的體積模型，使結構恢復演算法能得到更可靠的姿態重建結果。透視投影所造成的大小變化，以及影像中的二維資訊則作為觀測模型對粒子濾波器的粒子進行機率評估。最後，透過實驗來驗證此系統的整體效能及可靠性。This research presents a 3-D human arm tracking method with a monocular camera. In order to integrate the tracking system with the robot interface, the ability to achieve a real time performance is desired. Multiple clues are integrated by the multiple importance sampling (MIS) particle filter to track the arms with arbitrary motion. Due to the lack of depth information on a monocular camera, a sequential pose estimation based on structure-from-motion (SFM) is proposed to online estimate 3D posture of the arms. The estimation result is then designed as one of the proposal functions of MIS particle filter to generate the 3D posture hypotheses. In order to obtain a more reliable recovery of the motion from online resolution of the SFM, the structure of each body parts is priory assumed to be volumetric model. The feature points, which are spread according to the volumetric model of each body parts, are tracked to measure the image displacement for online solving the 3D motion. Additionally, while the objects in view appear larger when they are closer to the camera, the size effect from perspective projection is considered to disambiguate the arm posture. Finally, the hypotheses from the multiple importance sampling are verified by 2D visual features and the 3D posture information augmented with the size effect. Experiments show the robustness and efficiency of the proposed algorithm.1809276 bytesapplication/pdfen-US粒子濾波器影像追蹤從運動恢復結構手臂追蹤Particle FilterVisual TrackingStructure-from-motionArms Trackingthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/253954/1/ntu-100-R98921010-1.pdf