Violence Detection Using Convlstm and LRCN
N Benjamin Studd, Sebastian Terence, Jude Immaculate, Selvarathi Selvarathi, Vinodh Ewards S.E., Titus · 2024
The rise in surveillance video footages depicting violence is on the rise throughout the internet which highlights the urgent need for fast and efficient real-time violence detection systems. In this paper, we use two models that are based on convolutional neural networks (CNNs) and long short-term memory networks (LSTM) for detecting violence efficiently. Frames from videos are extracted and preprocessed and given as input to the models, usage of CNNs and LSTM helps in learning temporal and spatial information simultaneously from each frame. The models, fine-tuned Convolutional LSTM (ConvLSTM) and fine-tuned Long-term Recurrent Convolutional Networks (LRCN) outperform existing models in terms of accuracy and speed, by achieving an accuracy of 92% and 94% respectively.