Bounding-Box Trajectories Matter for Video Anomaly Detection
Researchers have developed TrajVAD, a new framework for video anomaly detection that utilizes bounding-box trajectories. This approach models normal kinematic patterns using normalizing flows, outperforming existing pose-based methods. An extended version incorporating pose information further enhances performance on key datasets like ShanghaiTech. AI
IMPACT This research introduces a novel method for video anomaly detection, potentially improving security and public safety systems by more accurately identifying unusual events.