Security Alert Initiative for Violence Undertaking – A Violence Detection System in College Environment
S Shreenithi, K.G. Srinidhi, Gurusamy Jeyakumar, T Senthilkumar, S Vasan, Md. Ashraful Babu · 2025
Automated video surveillance systems are important for spotting violence, like fights, to keep places safe. Considering a college environment it needs a lot of security personnel to check on the footage constantly and to react. This results in higher costs in terms of both time and finances. Implementing automated violence detection in surveillance cameras within a college setting is essential to address these challenges. Violent behavior is identified through the analysis of camera footage, allowing for prompt alerts to be sent to officials for immediate action. Previous research has largely relied on handcrafted features or traditional learning models to detect violence, but these methods often struggle with diverse video formats. The current systems encounter issues such as frequent false alerts and significant computational demands when monitoring and analyzing videos in real time. This paper proposes a novel approach to violence detection using ConvLSTM, a deep learning model that combines the spatial feature extraction capabilities of Convolutional Neural Networks (CNN) with the temporal modeling strengths of Long Short-Term Memory (LSTM) networks. By fusing CNN and LSTM, ConvLSTM achieves state-of-the-art performance, with an accuracy of 91% and a Macro Average F1-Score of 90%. Experimental results demonstrate the effectiveness of ConvLSTM in detecting violent behavior, with precision and recall rates of 95% and 91%, respectively, for violent behavior classification.