An Overview of Deep Learning Methods for Violence Detection

ABSTRACT: Violence detection in video has gained significant attention due to its critical applications in public safety, surveillance systems, and automated monitoring. Traditional methods relying on handcrafted features and manual observation have been surpassed by deep learning techniques, which offer improved accuracy and the ability to automatically extract complex patterns from video data. This paper provides an overview of state-of-the-art deep learning methods for violence detection, focusing on Convolutional Neural Networks, hybrid CNN+LSTM models, skeleton-based approaches, and Transformer architectures. Additionally, we explore the challenges faced in vision-based violence detection, including the impact of lighting conditions, model complexity, dynamic environments, and the diversity of violent behaviors. Finally, we examine popular datasets used in this domain and compare the performance of different deep learning models to offer insights for future research.