<![CDATA[The expansion of the application of drones has dispersed in wide range across military and civilian sectors. The application in such search and rescue missions are applicable with integration of computer vision and machine learning. A key feature of the drone for such applications is the capability to detect and locate objects and targets. Despite traditional methods perform excellently, deep-learning methods are the game changer in detection due to their better accuracy and robustness, rendering them ideal for real-time applications. The methods, including the YOLO series, are in continuous development to further enhance their performance. however, the regular issuance of updated and newer versions has intrigued curiosity regarding the potential superiority of the newer version over the previous versions in drone application. Hence, this paper has chosen the YOLOv8, YOLOv5u and YOLOv11 models for implementation on a DJI Tello drone to detect a custom target. A dataset for the target as a single class to be trained and validated is generated through images annotation. The target is required to be captured in the position of middle of the frame. However, the analysis upon performance metrics found that every model recorded high rates of precision, accuracy and recall. Yet, the simulations and experimentations displayed the ability of the model to accurately recognize the target. Thus, in order to evaluate the performance of each model thoroughly, it is advisable to ensure the data is sufficient and unbiased, while properly choosing the right setting parameters to the YOLO models.]]>
Copyrights © 2025
