The necessity for automated content analysis has grown due to the quick expansion of online multimedia platforms and video surveillance systems, especially for identifying violent activity. When working with large-scale datasets, manual video data monitoring is inefficient, time-consuming, and prone to errors. In order to improve security and surveillance systems, this study suggests an automatic video violence detection method based on deep learning. The suggested approach finds violent behaviour patterns in video sequences by combining temporal feature extraction methods with convolutional neural networks (CNNs). While temporal dependencies are recorded using models like 3D CNN topologies or Long Short-Term Memory (LSTM) networks, spatial characteristics are retrieved from each frame. The model can effectively learn distinguishing traits because it is trained on labelled datasets that include both violent and non-violent video clips. Preprocessing methods like frame extraction, normalisation, and data augmentation are used to enhance performance. Metrics including accuracy, precision, recall, and F1-score are used to assess the model. The suggested strategy outperforms conventional machine learning techniques and achieves high accuracy in identifying violent acts, according to experimental results. This system can be used for content filtering on digital platforms, public safety monitoring, and real-time surveillance. Through automated violence detection, the research helps create intelligent systems that can guarantee safer surroundings.
Ekambaram et al. (Thu,) studied this question.