Real-Time Violence Detection Using Audio and Video: A Review

In response to the urgent demand for increased public safety and security measures, this study presents a novel method for real-time violence identification in surveillance footage. This paper studies audio and video modalities, leveraging deep learning algorithms for comprehensive analysis. Discusses YOLOv4, a cutting-edge object recognition model, to accurately detect pertinent objects like individuals and weapons within video frames. Subsequently, employment of MobileNet for feature extraction and classification. By integrating audio analysis with video, the system achieves enhanced accuracy and robustness. Discuss audio features from accompanying audio streams and fuse them with visual characteristics to provide a holistic understanding of the scenario and different approaches. With its real-time capability, the system holds significant promise for deployment in various security and surveillance applications, thereby assisting efforts related to public safety. This integrated approach marks a significant advancement in violence detection technology, offering valuable support in addressing contemporary security challenges.