O-VAD: Industrial Video Anomaly Detection through Object-Centric Tracking and Reasoning
2026-07-20 • Computer Vision and Pattern Recognition
Computer Vision and Pattern RecognitionArtificial IntelligenceComputation and LanguageMultiagent Systems
AI summaryⓘ
The authors focus on detecting unusual objects or events in industrial videos, which is important for manufacturing quality control. They found that current video anomaly detection methods do not work well in complex industrial settings due to strict rules and object changes. To solve this, they created a new approach that does not need extra training or special knowledge and looks at how objects change over time, much like human inspectors do. Their method showed better results than existing techniques and also explains what type of anomaly it found.
Industrial Video Anomaly DetectionVideo Anomaly Detection (VAD)Vision-Language Models (VLMs)Spatial-Temporal DynamicsObject State EvolutionAnomaly ReasoningAgentic FrameworkQuality ControlTemporal State Trajectories
Authors
Mei Yuan, Qi Long, Qifeng Wu, Zhenyang Li, Yizhou Zhao, Lei Wang, Yang Liu, Min Xu
Abstract
Industrial Video Anomaly Detection (IVAD) aims to identify anomalous objects and events in an industrial process, which is crucial for modern manufacturing and quality control systems. Existing VLM-based anomaly reasoning methods are capable of detecting open-ended anomalies in general domains. However, their performance declines in industrial settings characterized by intricate object transformations, strict physics, and procedural constraints. To tackle the complexity of such interaction-intensive detection, we introduce a training-free agentic framework for anomaly detection free of domain-specific knowledge, emphasizing object state evolution like humans inspectors. It is designed to track spatial-temporal dynamics and underlying transformations of detected objects over time, and then reason over the object-wise temporal state trajectories to identify abnormal objects in grounded frames. Our method overcomes limitations of prior approaches that rely on retraining on normal clips or injecting domain knowledge as context for test-time inference. Extensive experiments on three IVAD datasets demonstrate that our method outperforms frontier VLMs, agentic frameworks, and traditional VAD methods fine-tuned on the respective datasets, while providing interpretable reports over anomaly processes and types.