Motion History and Shape Orientation Based Human Action Analysis

Swe Nwe Nwe Htun, Thi Thi Zin

Abstract: In recent decades, many research works focused on the considerations of falling and post-falling event analysis in the aspect of image processing technology to meet the consumer perspective of users. In this paper, the more informative considerations of human action analysis are developed for the prior state of fall or normal actions. In doing so, the effective background subtraction namely the Mixture of Gaussian with Low-Rank Matrix Factorization model is used to obtain robust and adaptive foreground from the practical video sequences. Then, the spatial-temporal bilateral grid for video sequences is constructed by using a standard graph cut theory to improve the cleaned foreground in order to detect the moving/motionless object region. Then, human actions are analyzed by employing motion history and shape orientation using the approximated ellipse method. The experiments are conducted on publicly available video sequences and our simulated video sequences.