Anomaly Detection based on Pose Estimation and GRU-FFN

Anomaly Detection has recently received significant attention as an efficient concept for surveillance. The main challenges of anomaly detection mainly come from three aspects:1) Anomaly behavior always occurs suddenly with unknown. 2) One class anomaly has a completely different feature from another. 3)Anomaly datasets are rare. Most current approaches learn features of normal samples for reconstructing video frames, from which anomaly is detected by reconstruction error. However, they ignored the representation capacity of CNNs, and abnormal behavior can still be reconstructed. Moreover, feature representation-based methods are difficult to distinguish between signal and noise. To solve these problems, we propose a novel framework for anomaly detection in this paper, which is equipped with human motion prediction ability. Special emphasis is placed on skeleton key points prediction which relies on deep learning. The experimental results show that the proposed approach outperforms current state-of-the-art techniques quantitatively and qualitatively.