This paper introduces an advanced method for weapon detection in surveillance systems, integrating wavelet-based preprocessing with the YOLOv8 deep learning framework. Addressing critical challenges such as occlusions, lighting variations, and complex backgrounds, the approach ensures robust detection in real-world scenarios. Experimental evaluations validate the effectiveness of YOLOv8, achieving an impressive accuracy and precision of 87.5%, recall of 81.2%, an F1-score of 84.2%, and an inference time of just 25 ms per frame. These metrics surpass traditional models like YOLOv3, Faster-RCNN, and SSD, which demonstrate slower inference times and lower performance across precision, recall, and F1-scores. The system's high accuracy and speed make it a viable solution for real-time applications in critical environments, such as schools, airports, and public transportation hubs. By leveraging YOLOv8's optimized architecture, this research contributes to the development of scalable, efficient, and adaptive surveillance solutions aimed at ensuring public safety and mitigating risks in diverse operational settings.