Toward Fast and Accurate Violence Detection for Automated Video Surveillance Applications

Toward Fast and Accurate Violence Detection for Automated Video Surveillance Applications

{BE Students}* Mhaske Pragati Sambhaji, Patil Sakshi Balraje, Sable Aishwarya

Ramhari, Sawant Dipali Bhagwat

{Faculty}* Tanuja Shrikant Dhage

Department of Computer Engineering, DattaKala Group Of Institution Faculty Of

Engineering, Bhigwan-413130, University of Pune, Maharashtra, INDIA.

pragatimhaske2@gmail.com

dipalibsawant015@gmail.com

aishwaryasable59@gmail.com

sakshibalrajepatil123@gmail.com

Abstract

Surveillance cameras are increasingly utilized worldwide, but the large volume of generated video data poses challenges for real-time monitoring. This paper presents a deep learning-based approach for automatic violence detection in surveillance footage. By leveraging 3D convolutional neural networks and transfer learning from pre-trained action recognition models, we propose an efficient and accurate system for identifying violent activities. Our experimental evaluation on multiple datasets demonstrates improved classification accuracy with fewer model parameters compared to state-of-the-art methods.

Keywords: Violence Detection, Video Surveillance, Deep Learning, Anomaly Detection, Human Activity Recognition, Security, Smart Cities, 3D Convolutions, Action Recognition