P2-VGAN: a lightweight velocity-weighted graph attention network for privacy-conscious edge anomaly detection

To resolve these trade-offs, we present the Privacy-Conscious Velocity-Weighted Graph Attention Network (P2-VGAN; retaining “P2” as a legacy designator indicating a privacy-conscious representation rather than a cryptographic claim).

2.4 Weakly supervised video anomaly detection (WSVAD) Frame-level anomaly annotation is labor-intensive and difficult to scale.

3.3 Stage 3: Velocity-weighted attention mechanism (VW-AM) The VW-AM module embeds kinematic priors directly into the graph attention kernel.

7 Conclusion This paper presented P2-VGAN, a lightweight skeleton-based graph attention network for privacy-conscious video anomaly detection on edge CPU gateways.

doi: 10.1109/CVPR.2019.00133 Summary Keywords edge computing , graph attention networks , kinematic features , lightweight deep learning , privacy-conscious surveillance , real-time inference , skeleton-based action analysis , video anomaly detection Citation P. M and C. M. V (2026) P2-VGAN: a lightweight velocity-weighted graph attention network for privacy-conscious edge anomaly detection .