Authors: K. Vigneshwar, G.Bharath Simha Reddy, G.Shashidhar, A.Uday Kiran
Abstract: Public safety in public areas has become a significant concern for governments and businesses globally. Video surveillance systems are being increasingly integrated to ensure public safety, with deep learning techniques enhancing their ability to detect potential threats. Traditional video surveillance often relies on passive monitoring, but with advancements in AI, surveillance systems can now actively detect risks such as weapons (guns and knives) in real- time. This paper presents a deep learning-based system for weapon detection using MobileNet- V2, a CNN model known for its computational efficiency. MobileNet-V2 has shown an improvement of approximately 35% in processing speed compared to its predecessor, MobileNet-V1, while maintaining similar accuracy levels. This increase in speed is crucial for real-time weapon detection, where quick identification and response are vital to preventing threats. The study compares two approaches to weapon detection using CNNs, evaluating MobileNet-V1 and MobileNet-V2. The results indicate that MobileNet-V2 outperforms MobileNet-V1 not only in terms of speed but also in its ability to maintain high accuracy, marking a significant advancement in the field of weapon detection through deep learning. These improvements are vital in practical applications, such as public spaces, where large amounts of video data must be processed rapidly. The proposed system demonstrates a clear enhancement over prior methods in detecting guns and knives, offering a reliable, fast solution for real-time surveillance. This research highlights the effectiveness of MobileNet-V2 in improving public safety through advanced AI technology, providing a scalable solution for detecting threats in urban environments.
