Advanced Deep Learning Approaches for Accurate and Efficient Suspicious Behavior Detection in Surveillance Videos

Violence Artificial Intelligence (AI) and Deep Learning (DL) systems present a difficult research area for identifying violence in videos within urban security frameworks and video surveillance systems. The proposed model divides violence detection tasks in video into two stages to achieve both rapid processing and precise outcomes. The LeNet-۵ model operates at a speed of ۰.۸ frames per second to filter out non-violent videos during the first stage of operation. The second analysis stage employs the ResNet-۵۰ model to inspect videos for potential violence when their probability surpasses ۰.۴. The Real-Life Violence dataset consisting of ۱۹۵۱ videos with ۱۰۰۰ violent and ۹۵۱ non-violent videos was used for testing this system. The implementation produced ۹۷.۰۳% accuracy together with ۹۵.۷۰% recall and ۹۸.۴۶% precision and ۹۷.۰۶% F۱-Score and AUC of ۰.۹۹۰۲. Each frame requires only ۲۰ milliseconds of processing time which allows real-time application of this system. A comparative analysis with existing methods, such as ۳D-CNN, ViT, and YOLOv۵+TSN, highlights the superiority of the proposed model in terms of both accuracy and speed. The system achieves better violence detection capabilities and operational reliability in real-world applications because it decreases detection errors.

Keywords:

Anomaly Detection , Deep learning , Pattern Recognition , Feature Fusion , Video Analysis