Abstract
Video anomaly detection has emerged as an important research topic in computer vision and analysis. This paper presents an improved spatio-temporal color mechanism for video anomaly detection, incorporating adaptive color modules and multi-scale feature enhancement techniques. The proposed system uses a two-stream architecture that processes spatial and temporal data in a parallel way while maintaining efficient interactions. A novel hierarchical attention structure captures multi-scale appearance features, enabling the detection of anomalies at various spatial resolutions. The temporal attention component introduces an adaptive temporal sampling strategy that efficiently processes long video sequences while preserving critical temporal dependencies. The framework includes a feature enhancement mechanism that dynamically adjusts the importance of different spatio-temporal features based on their relevance to anomaly detection. The analysis of several benchmarks, including UCF-Crime, CUHK Avenue, and ShanghaiTech, demonstrates the superiority of the plan. The framework achieves a significant improvement in detection accuracy, with AUC scores of 89.7%, 87.6%, and 88.2% respectively, while maintaining computational efficiency. Suitable for real-time use. The results showed an average improvement of 5.4% in the detection accuracy compared to the state-of-the-art method, establishing the value of the request for submission in the inspection application.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call