Time Series Action Recognition based on SVM with Fisher Kernel

Accession number;04A0392173
Title;Time Series Action Recognition based on SVM with Fisher Kernel
Author;SHIMOSAKA MASAMICHI(Univ. of Tokyo)   MORI TAKETOSHI(Univ. of Tokyo)   SEGAWA YUUSHI(Univ. of Tokyo)   HARADA TATSUYA(Univ. of Tokyo)   SATO TOMOMASA(Univ. of Tokyo)   
Journal Title;Nippon Robotto Gakkai Gakujutsu Koenkai Yokoshu (CD-ROM)
Journal Code:L4867A
ISSN:
VOL.21st;NO.;PAGE.2J24(2003)
Figure&Table&Reference;
Pub. Country;Japan
Language;Japanese
Abstract;This paper proposes the recognition algorithm for time-series human motion based on kernel machines. The kernel computation in the proposed methods utilizes Fisher score from Hidden Markov Model as representation of time-series human motion, and the system discriminate the motion by Support Vector Machine. Experimental result shows that the recognition accuracy of the proposed algorithm is superior to that of the conventional approach with Hidden Markov Model, which is based on log likelihood value. (author abst.)