Harnessing Deep Learning Models for Enhanced Violence Recognition in Modern Surveillance Systems

Authors

Anand, Prof. Vikas Kamle

Abstract

In the realm of video content analysis, distinguishing between 'Violence' and 'NonViolence' poses a significant challenge due to the dynamic and complex nature of video data. Traditional models like ResNet50 and MobileNetV2 have shown substantial capabilities in image classification tasks but struggle with the temporal aspects inherent in video. To address this, we introduce a hybrid deep learning approach that combines the spatial feature extraction strengths of InceptionV3 with the temporal pattern recognition prowess of LSTM networks. Our proposed method involves rigorous preprocessing of data, including noise reduction and feature extraction, followed by a systematic model training process. The performance evaluation of our approach has revealed a remarkable accuracy of 99.86% and a validation accuracy of 92.48%, outperforming the other evaluated models by a significant margin. These results not only validate the efficacy of the hybrid model in video classification tasks but also suggest its potential for broader applications in real-world scenarios that require nuanced content discernment.

Keywords

Violence Recognition , Deep Learning Models , Automated Surveillance , Real-Time Detection , Convolutional Neural Networks (CNNs).

Harnessing Deep Learning Models for Enhanced Violence Recognition in Modern Surveillance Systems. Anand, Prof. Vikas Kamle. 2024. IJIRCT, Volume 10, Issue 4. Pages 1-10. https://www.ijirct.org/viewPaper.php?paperId=2408069