The realm of advanced deep learning architectures for violence detection is burgeoning and has the potential to revolutionize proposed approaches to preventing and identifying instances of violence. In this rapidly evolving field, deep learning architectures exhibit the capability to discern intricate patterns and make predictions based on vast datasets. Their suitability for violence detection is particularly pronounced, given their capacity to identify subtle indicators that may foreshadow aggressive behavior. This research advocates for an autonomous violence detection system leveraging a combined CNN + LSTM architecture. Each model adeptly extracts features at the frame level, progressively enhancing accuracy in the process. The proposed CNN + LSTM model, integrating spatial and temporal analyses, achieves an impressive accuracy rate of 90%. The synergy of Convolutional Neural Networks (CNNs) for spatial feature extraction and Long Short-Term Memory (LSTM) networks for temporal aggregation facilitates a nuanced analysis of local motion patterns. Rigorous evaluations conducted on publicly available datasets attest to the models’ efficacy in violence detection. The amalgamated solution, utilizing the CNN + LSTM architecture, is tailored for real-time surveillance, ensuring swift identification and notification of potential violence to relevant authorities for prompt remote intervention.