Real-time Human Behavior Recognition Based On Articulated Model

摘要

In order to identify human behavior classification inIntelligent Security Monitoring System, an articulated modelto extracting human body and classifying the behaviors of themoving objects is presented in this paper. An improvedstatistical Gaussian model is used as adaptive backgroundupdating method. After silhouettes of objects are extractedfrom the video images, we propose an articulated model ofhuman, using the variety of body's trunk and limbs contourangles. The angles that can represent the pose of the skeletonmodel and length-width ratio of the human are used as featurevector. Finally Bayesian Networks is used for human posturetraining, modeling and activity matching to recognize thehuman motion. Experiment results have shown that thismethod gives stable performances and good robustness.