One video is sufficient? Human activity recognition using active video composition | IEEE Conference Publication | IEEE Xplore