Human Action Recognition Based on Discrete Fourier Transform and LK Optical Flow

Abstract

With the advancement of technology the security issue has become more important In order to prevent illegal behavior and to lower down the accident ratio in home care or medical care it is an important topic for using surveillance systems to recognize illegal actions In this thesis the method which is based on the Discrete Fourier Transform (DFT) and optical flow was proposed In the proposed method first of all the optical flow image of sequences is captured and the foreground image is extracted by background subtraction Moreover the overlapping image of optical flow and background subtraction are taken as the important features Second the DFT is used to overcome the problem of focal distance and object offsets Finally Hierarchical k-Nearest Neighbors (H-kNN) is used to classify correctly According to experiment results it shows that proposed method is robust and efficient

Date of Award	2014 Jul 22
Original language	English
Supervisor	Shen-Chuan Tai (Supervisor)