Enhancing Violence Detection in Surveillance Videos using Deep Learning Techniques

The limitations of manual monitoring and conventional techniques that rely on pre-constructed datasets are addressed by this study's innovative approach to real-time anomaly detection in security camera feeds. Using real-time data from various environments, the system uses CNN and RNN architectures for analysis, improves input quality with Gamma Correction, and uses the VGG16 model for feature extraction. Specialized Support Vector Machines (SVMs) are employed to identify irregularities and generate warnings for questionable activity. According to the results, CNN models perform better than RNNs in terms of accuracy and efficiency, especially when it comes to identifying violent incidents. This shows how effective the system is and how it surpasses more conventional methods.