Crowd anomaly behavior detection has emerged as a key research area within video surveillance. This paper explores strategies for analyzing crowd behavior through the lens of collective movement and examines how video surveillance technologies can be leveraged to address crowd safety challenges. Particular emphasis is placed on the role of deep learning in advancing anomaly detection. We review relevant literature published between 2019 and 2025, categorizing deep learning-based approaches into three main types: object detection, feature extraction, and data reconstruction/prediction. For object detection methods, we summarize network architectures, anomaly detection techniques, and Area Under the Curve (AUC) evaluation metrics across benchmark datasets. For feature extraction approaches, we detail the networks employed, the temporal and spatial features extracted, and their performance metrics. For reconstruction and prediction-based methods, we outline the data generators, generated objects, and associated discriminators. Finally, the paper offers a forward-looking perspective on the challenges and future directions in crowd anomaly detection research. • Comprehensive review categorizes anomaly detection methods, metrics, and architectures (2019–2024). • Deep learning transforms crowd analysis via object detection, feature extraction, and reconstruction/prediction. • Benchmark datasets validate AUC performance for spatial-temporal feature-based approaches. • Future challenges include scalability and real-time processing for surveillance systems.
Zhang et al. (Fri,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: