http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
Human Action Recognition Based on 3D Human Modeling and Cyclic HMMs
Shian-Ru Ke,Hoang Le Uyen Thuc,황젝넹,유장희,최경호 한국전자통신연구원 2014 ETRI Journal Vol.36 No.4
Human action recognition is used in areas such assurveillance, entertainment, and healthcare. This paperproposes a system to recognize both single and continuoushuman actions from monocular video sequences, based on3D human modeling and cyclic hidden Markov models(CHMMs). First, for each frame in a monocular videosequence, the 3D coordinates of joints belonging to ahuman object, through actions of multiple cycles, areextracted using 3D human modeling techniques. The 3Dcoordinates are then converted into a set of geometricalrelational features (GRFs) for dimensionality reductionand discrimination increase. For further dimensionalityreduction, k-means clustering is applied to the GRFs togenerate clustered feature vectors. These vectors are usedto train CHMMs separately for different types of actions,based on the Baum–Welch re-estimation algorithm. Forrecognition of continuous actions that are concatenatedfrom several distinct types of actions, a designed graphicalmodel is used to systematically concatenate differentseparately trained CHMMs. The experimental resultsshow the effective performance of our proposed system inboth single and continuous action recognition problems.