中文模板-基于深度和边缘信息的人体运动跟踪.pdfVIP

飥 鍩轰簬娣卞害鍜岃竟缂樹俊鎭殑浜轰綋杩愬姩璺熻釜飥 1 1 + 1 2 鍒橀暱寰 , 鍒樺織骞 , 鑳″織鎭 , 鍙剁懓鏄€ 1(鍝堝皵婊ㄧ悊宸ュぇ瀛 璁＄畻鏈虹瀛︿笌鎶€鏈闄? 榛戦緳姹 鍝堝皵婊 150080) 2 (榛戦緳姹熺鎶€瀛﹂櫌 鐢垫皵涓庝俊鎭伐绋嬪闄?榛戦緳姹 鍝堝皵婊 150027) * Motion Tracking by Depth and Silhouette Information 1 1+ 1 2 LIU Chang-Zheng , LIU Zhi-Guang , HU Zhi-Heng , YE Gui-Yun 1(College of Computer Science and Technology , Harbin University of Science and Technology , Harbin 150080, China) 2 (Electrical and Information Engineering College, Heilongjiang Institute of Science and Technology, Harbin 150027, China) + Corresponding author : E-mail: triggerliu@ Abstract : This paper presents a novel approach for human motion tracking without markers, which is different from the conventional marker based methods. The joints are estimation by adding depth and silhouette informat ion. Combining the information from both the dep th and the color frames, a new matte is generated. A depth mask is generated by background subtraction and it can be used to handle the edges of the foreground objects. The segmentation step does not make strong restrictions to the capture environment, e.g. homogenous clothing or background, constant lighting etc. Camshift algorithm utilizes color distribution to find target candidate and the background noise could be remove by combining motion detection and Camshift algorithm. Since the changes between different frames are so small Kalman filter is used to predict coordinates of each joint so as to minimize the searching cost. Curve Fitting is made to reconstruct the motion trajectories of the discrete jo