Incoherent motion detection using a time-series Gram matrix feature

This paper p'VlJOses a /lew meThod for incoherent lIlotion recognitioll from video sequences. We use time-series ~pariQ-lempora/illlensity gradients within a space-rime patch. Using a global space-time patch, we fOl/lld ,hor the gradiem feature allows liS 10 distinguish an incoherent lIlotion from a coherem motion wilhout segmentalion. Furthermore 'he algorithm can rim in relll rime cI'en on an ell/belMetl device. III this paper. we I'erify /Ilorion recognilion peiformallce for actions which we consider coherent (walk/run) and incoherent (rurnlsquorlim'erse walk). To identify the //lulliple motion classes. we use lineardiscriminant analysis andthe KNN method. As u reslllt. Ollr method L'un distinguish multiple-class motion putrerns with a detection rate of ubol/t 80%. Also the detection rute of incolJerem mo? tions is JOO% with afalse positil'e rare of less tlWl1 JO %.

• Type: Conference
• Year: 2008
• Where: ICPR
• Authors: Hironobu Fujiyoshi, Masanori Miyoshi, Masato Kazui, Shoji Muramatsu
• Keywords: Computer Vision, Distinguish Multiple-class Motion, Global Space-time Patch, ICPR 2008, Incoherent LIlotion Recognitioll