A Comprehensive Survey of Transformer-Based Models for Video Anomaly Detection in Surveillance Systems
Bibliographic record
Abstract
Video anomaly detection constitutes a pivotal component for ensuring public safety, regulating traffic networks, supervising industrial workflows, and enabling smart city ecosystems.Recent advances in deep learning -particularly transformer-based architectures -have markedly improved the modeling of high-order spatio-temporal dependencies within surveillance video streams.This study presents a systematic comparative evaluation of state-of-the-art frameworks, encompassing CNN-Transformer hybrids, dual-stream motion-appearance encoders, pure vision transformer architectures, weakly supervised paradigms, and class-incremental learning strategies.Experiments conducted on benchmark datasets including UCF-Crime, ShanghaiTech, CUHK Avenue, UCSD Ped2, RWF-2000, and Drone-Anomaly highlight domain-specific advantages: BiMT achieves superior accuracy on UCF-Crime; TDS-Net demonstrates robustness on motion-intensive corpora such as ShanghaiTech and Avenue; and unsupervised transformer models excel in aerial anomaly detection.Furthermore, SwinIoT provides edge-optimized inference for IoT-enabled smart environments, while CILAR-Net supports dynamic integration of emergent anomalous classes.The analysis underscores critical trade-offs-labeling cost reduction via ST-HTAM, real-time efficiency through ViT+SRU++, and anomaly localization achieved by SwinAnomaly.The findings indicate that no single architecture universally dominates across tasks; instead, optimal model selection is context-sensitive, determined by accuracy-efficiency requirements, annotation costs, adaptability, and deployment constraints.The contribution is a decisionsupport framework for selecting transformer-based anomaly detection models across heterogeneous video surveillance domains.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.004 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.002 |
| Bibliometrics | 0.001 | 0.002 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.002 | 0.003 |
| Open science | 0.003 | 0.001 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".