3D Convolutional Neural Network for Crowd Behavior Classification in Surveillance Videos

초록

Classifying crowd behavior from videos is an ongoing challenge in computer vision because of its complexity. In particular, crowd behavior classification in a video surveillance system is necessary because it enables effective monitoring of various situations, such as dangerous situations. Due to this demand, recently, various studies are being conducted to perform such classification using 3D Convolutional Neural Networks (CNN). In this paper,for developing a new deep learning model, we will summarize the existing 3D CNN-based studies on crowd behavior classification and define the baseline performance through experiments.