Triplet-set feature proximity learning for video anomaly detection

Kuldeep Marotirao Biradar, Murari Mandal, Sachin Dube, Santosh Kumar Vipparthi, Dinesh Kumar Tyagi

Abstract: Highlights•Our method uses triplet sets for video feature encoding to enhance model resilience.•Aggregating normal video features creates a meta-feature vector for behavior quantification.•AnoVIL, a new dataset, includes diverse normal and ab-normal human-centric activities for evaluation.•Our method outperforms existing methods on UCF-Crime, IIT-H, and AnoVIL dataset validated by ablation study.