Journal of Engineering Research and Reports

The creation of AI-powered drone-based security models has been fueled by the growing need for intelligent surveillance systems. In order to detect security risks in real time, this project focuses on developing an AI-driven security drone system that combines Deep Learning models with unmanned aerial vehicles (UAVs). A wide range of data, including video frames and aerial photos, was gathered and preprocessed using augmentation, normalisation, and resizing. Three sophisticated object detection models, YOLOv5, YOLOv6, and YOLOv7, were then trained using this dataset. The models were evaluated using key performance metrics, including Precision, Recall, F1-score, and mean Average Precision (mAP). Results revealed that YOLOv5 achieved a Precision of 0.88, Recall of 0.86, F1-score of 0.85, and mAP of 0.805; YOLOv6 achieved a Precision of 0.95, Recall of 0.94, F1-score of 0.92, and mAP of 0.876; while YOLOv7 demonstrated superior performance, achieving a Precision of 0.98, Recall of 0.98, F1-score of 0.96, and mAP of 0.917. Furthermore, the developed models were tested in a simulated environment using Mission Planner with Software-In-The-Loop (SITL), enabling realistic flight path monitoring and real-time event detection without requiring physical deployment. Although the system was validated in a simulated environment, it lays the groundwork for future real-world applications across critical domains. The study concludes that the YOLOv7 model offers the highest accuracy and reliability for drone-based real-time surveillance and threat detection, setting a strong foundation for future deployment in security operations, disaster management, agriculture, and other sectors.