Discrete Wavelet Transform for CNN-BiLSTM-Based Violence Detection

摘要

In this paper, our approach aims to enhance the classification of violent and non-violent activities in public areas. Violent activities lead to the destruction of loss of life and general properties. These anti-social activities have been increasing at an alarming rate over the past years. Our approach, when merged with the camera surveillance system, can bring about real-time automation, in the detection of criminal activities. DWT-based convolutional bidirectional LSTM has been used to detect violent actions, and the results have been compared with the other existing approaches. Our proposed plan gives 94.06% classification accuracy for the widely used standard Hockey dataset.