Romero Morais, Vuong Le, Truyen Tran, Budhaditya Saha, Moussa Mansour, Svetha Venkatesh
Appearance features have been widely used in video anomaly detection even though they contain complex entangled factors. We propose a new method to model the normal patterns of human movements in surveillance video for anomaly detection using dynamic skeleton features. We decompose the skeletal movements into two sub-components: global body movement and local body posture. We model the dynamics and interaction of the coupled features in our novel Message-Passing Encoder-Decoder Recurrent Network. We observed that the decoupled features collaboratively interact in our spatio-temporal model to accurately identify human-related irregular events from surveillance video sequences. Compared to traditional appearance-based models, our method achieves superior outlier detection performance. Our model also offers "open-box" examination and decision explanation made possible by the semantically understandable features and a network architecture supporting interpretability.
| Task | Dataset | Metric | Value | Model |
|---|---|---|---|---|
| Anomaly Detection | HR-ShanghaiTech | AUC | 75.4 | MPED-RNN |
| Anomaly Detection | HR-Avenue | AUC | 86.3 | MPED-RNN |
| Anomaly Detection | HR-UBnormal | AUC | 61.2 | MPED-RNN |
| 3D Anomaly Detection | HR-ShanghaiTech | AUC | 75.4 | MPED-RNN |
| 3D Anomaly Detection | HR-Avenue | AUC | 86.3 | MPED-RNN |
| 3D Anomaly Detection | HR-UBnormal | AUC | 61.2 | MPED-RNN |
| Video Anomaly Detection | HR-ShanghaiTech | AUC | 75.4 | MPED-RNN |
| Video Anomaly Detection | HR-Avenue | AUC | 86.3 | MPED-RNN |
| Video Anomaly Detection | HR-UBnormal | AUC | 61.2 | MPED-RNN |