Enhancing Public Safety Through Advanced Video Analysis: A Conv-LSTM-SVM Model for Violence Detection in Surveillance Footage

This study pioneers a new method for detecting violence in surveillance videos, addressing a major challenge in public safety and video analysis. The study presents a hybrid model that uses Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM) networks, and Support Vector Machines (SVM) to detect violent incidents in video data. The Convolutional Long-Short-Term Memory and Support Vector Machines (Conv-LSTM-SVM) model combines CNN spatial feature extraction, LSTM temporal dependency modelling, and SVM classification power. A pre-trained DenseNet121 model extracts spatial information efficiently via transfer learning from large datasets in the proposed architecture. An LSTM layer captures temporal dynamics needed to understand video sequence activity development, while an SVM with a Radial Basis Function (RBF) kernel creates complex decision boundaries in the feature space with many dimensions for the final categorization. The model was developed, trained and tested using the Keras library running on TensorFlow, using an experimental research design. The model is tested using two well-known datasets: the UCF Crime dataset, which contains 1900 surveillance clips of 13 classes of violent situations, and the RWF-2000 dataset, which analyses real-world fighting. The proposed model is the best in its class, outperforming CNN, LSTM, and Conv-LSTM models with 97.3% accuracy on the UCF Crime dataset. Cross-dataset validation yielded 92.5% accuracy on the RWF-2000 dataset without changes, demonstrating robust generalization. The study also considers how public safety could be improved by processing several video streams in real time and reducing false alarms. An examination of the ethical challenges and restrictions of automated surveillance systems, such as privacy, biases, and human supervision was also done. This research uses advanced video analysis to improve public safety by creating more efficient and adaptive surveillance systems.