<![CDATA[This study aims to design and develop a prototype surveillance camera system based on the You Only Look Once version 5 (YOLOv5) algorithm for real-time detection of sharp objects, namely knives and scissors, integrated with Telegram notifications. The dataset consists of 2000 images (1000 images per class), annotated via Roboflow and trained in Google Colab. The methodology includes data collection, preprocessing, model training, model conversion, and real-time detection implementation using Python in PyCharm. Evaluation results show a mean Average Precision (mAP@0.5) of 0.88 and mAP@0.5:0.95 of 0.577. The scissors class achieved higher precision and recall (0.934 and 0.88) compared to knives (0.808 and 0.795). Real-time testing produced an average confidence score of 0.445 and an average Frame Per Second (FPS) of 0.56, indicating hardware limitations. Confusion matrix analysis revealed a 58% misclassification rate of knives as background, higher than scissors (42%). This study confirms the effectiveness of YOLOv5 for sharp object detection in security applications, with potential for improvement through hardware optimization and dataset diversification]]>
Copyrights © 2026
