Real Time Violence Detection System Using YOLOv7 and Deep Learning Techniques

In recent times, violence in public spaces such as schools, hospitals, and other community environments has increased a lot. Due to this alarming rise in violence CCTV cameras have been deployed everywhere to enhance safety and security. However, these systems depend on manual monitoring which is prone to human errors and cause delays in the identification of violent events and action which can lead to significant damage. To overcome these limitations, this study has proposed a violence detection and alarm system based on deep learning algorithms. The hybridization of YOLOv7 allows rapid recognition of possible threats by identifying human presence and movement, and the lightweight classifier provides a high accurate classification of the identified tasks while reducing the processing computational resources of the system. With this dual-model-based framework, it is guaranteed that both the speed and accuracy can be achieved, and that the system can be implemented on edge devices such as mobile devices. The function of the system is to identify violence while it is happening and to issue alerts to security or law enforcement authorities committed to rapid response. The system was tested in real-time scenarios and achieved 91% accuracy in detecting violent actions. The system incorporates a Telegram bot to send real-time alerts to the relevant authorities. By leveraging advanced deep learning techniques, this study seeks to close the gap between existing surveillance methods and the need for a more automated and accurate violence detection system.