Activity recognition method of multiple view angles from monocular videos
单目视频中的多视角行为识别方法

An activity recognition method of multiple view angles from monocular videos was introduced. To aim at recognition in un-calibrated videos, this method could solve the problem choosing a uniform activity characteristic sector of different view angles by utilizing 3-D information included in 2-D image sequence effectively and classifying human skeleton. The experiments show that it has good recognition rate for activity of different view angles from monocular videos.