Authors

K. Tran, D. Nguyen, H. Nguyen

Abstract

Violent and abnormal behaviors detecting has been an interesting field recently. Many types of research have been done to accurately predict the behaviors of human whether it’s violent or not. However, it requires a lot of modern and strong infrastructure to complete the task. To fill the gap, in this research, we propose a skeleton-based method that doesn’t require much of infrastructure but very fast and accurate to detect violent behaviors. Our method contains two stages: extracting deep features from image frames to estimate human pose, and then we define four handcraft features to classify whether the act is violent or not. We tested our method on UT-Interaction [1] dataset which result is very tempting and promising