Reviewed by FilmVZ.Com on October 9

K-interactions

Video Description :
Tagging Human Motion Using Skeletal Tracking With Kinect: Here recognize human interactions set -defined classes. We build K-10 Interaction Dataset ten classes -person interactions performed agents captured Kinect Xbox 360. We construct interaction representations terms local space-time features integrate representations SVM classification ( MILboost classification) schemes recognition. We aligned clips dataset Canonical Time Warping algorithm led improvement interaction classification results. The accuracy reported : ~84% (MILboost) ~82% (SVM) The video captured prototype final feature tuning classification. In final experiments, waving class discarded. Credits: Debaleena Chattopadhyay, School Informatics, IUPUI Tamara Berg, Computer Science Dept., Stony Brook University, NY Vision & MoCap Lab, Stony Brook University, NY