Crime Detection from Pre-crime Video Analysis with Augmented Pose Information

This study focuses on the task of pre-crime event detection in videos, specifically in the context of shoplifting. While video understanding and anomaly detection in videos have been widely studied, our work proposes a novel approach of utilizing human pose information to augment the pre-crime video data with the aim of predicting critical events such as shoplifting. We used pre-crime scenes from shoplifting videos and normal videos in a 3D CNN architecture, with the addition of pose information as augmented data. The contribution of our study lies in the use of pose information, which captures relevant behaviors of people (such as looking around, walking back and forth, and changing direction) immediately before committing a crime. Our experimental results demonstrate the effectiveness of the proposed method in improving pre-crime event detection accuracy.