Integrated Violence and Weapon Detection Using Deep Learning
Vaibhav Yadav, Sanskar Kumar, Atul Goyal, Sidharth Bhatla, Geeta Sikka, Amandeep Kaur · 2024
Detecting violence in public spaces is crucial for ensuring safety and security. While current methods primarily rely on body movement cues, they often struggle to identify instances like single-gunshot violence. In this study, an innovative approach is proposed to overcome this limitation by integrating weapon detection to enhance accuracy and reduce false positives. Drawing inspiration from recent advancements in deep learning, techniques such as feature extraction using pre-trained networks are leveraged. Our methodology involves the combination of convolutional neural networks (CNNs) for spatial feature extraction and long short-term memory (LSTM) networks for temporal relation learning. Additionally, the efficacy of integrating transformer-based models, such as the Vision Transformer, for enhanced feature representation is explored. Through comprehensive evaluation on diverse datasets, our approach demonstrates high accuracy while maintaining real-time processing speed. Comparative analysis against state-of-the-art methods reveals superior performance, achieving an accuracy of up to 98% and a processing speed of 131 frames/sec. This research significantly contributes to the enhancement of violence detection systems by addressing challenges related to accuracy, speed, and applicability across various video sources.