A Literature Review on Real-Time Crime Detection Using Deep Learning and Edge Computing

The growing need for security in urban and commercial environments has driven the development of intelligent surveillance systems capable of detecting criminal activities in real time. While traditional cloud-based solutions offer advanced capabilities, they face limitations in terms of latency, privacy, and bandwidth usage. In this context, Edge Computing has emerged as a promising alternative, enabling local and fast processing of video data through embedded artificial intelligence models. This review article presents a comprehensive analysis of recent advances in real-time detection of thefts and weapons using Edge Artificial Intelligence (Edge AI). A total of 30 scientific articles published between 2018 and 2025 were selected and categorized, taking into account detection models, computing platforms, evaluation metrics, datasets, and real-world applications. The results highlight the predominant use of lightweight convolutional neural networks, especially YOLO-based models, implemented on devices such as Jetson Nano, Raspberry Pi, and Google Coral. Key challenges addressed include detection under lowlight conditions, identification of small or partially concealed weapons, and the reduction of false positives. The review identifies gaps in the current literature, such as the lack of annotated real-world datasets and the need for behavior-based models in retail contexts. Finally, emerging trends and future research directions are discussed, aiming at the development of efficient, accurate, and privacy-respecting Edge AI systems for real-time security surveillance. © 2025 IEEE.