Hệ thống giám sát thông minh phát hiện bất thường dựa trên trí tuệ nhân tạo

Abstract

This study, titled Intelligent Monitoring System for Danger Prediction and Alert, proposes a comprehensive system to protect children and women from dangerous situations such as violence, abuse, and kidnapping. Based on trajectory data and audio analysis, the system applies machine learning models such as Seq2Seq, LSTM combined with K-means clustering algorithm to detect abnormal movement patterns, while using XLM-RoBERTa and GPT models to identify harmful content in conversations. The research has developed a mobile application integrating features like geofencing, schedule monitoring, and instant alerts when anomalies are detected. Test results show the XLM-RoBERTa model achieved an F1-score of 0.88 in detecting harmful content, demonstrating the system's practical applicability. The study concludes that combining advanced artificial intelligence techniques can create effective solutions to protect vulnerable populations, while proposing future development directions such as IoT integration, expanding the target audience, and cooperation with relevant authorities.