Study of Analyzing Dynamical Behaviors and 3D Reconstruction from Video Sequences

点击翻译按钮获取中文摘要

Among many research topics related to video sequence analysis, this research focuses on the following topics: (1) Human posture estimation, (2) Tracking and recognizing humans, (3) Pose estimation of a moving camera and 3D reconstruction, and (4) Analysis of botanical trees’ behaviors. In (1), we study a method that utilizes results of detedting the user’s postures of the face, hand and fingers for human interface between a large screen, showing effectiveness of the proposed method. In (2), we explore human tracking and action recognition by a tensor analysis based method, recognition of customers’ behaviors in shops by a Bayesian Network based method, and Condenstation algorithm based recognition of sign langues’ vocabularies that use facial expressions and hand gestures. In (3), we research how to detect moving objects from dynamic video sequences and to reconstruct 3D structure of the detected objects, showing the validity of the proposed method. In (4), we study how to track a leaf in front of other leaves using a Particle filter, showing promising results.