Simulation environments offer significant advantages in real-world drone applications by mitigating potential damage caused by accidents, collisions, or system failures while enabling safe and controlled testing conditions. In this study, object detection was conducted using images captured by a camera mounted on a drone within the Gazebo simulation environment. Object classification was performed using the YOLOv3 deep learning architecture. This study investigates how physical parameters such as drone altitude, gimbal angle, and ambient lighting affect object detection performance. The collected data were analyzed using the TOPSIS and Grid Search methods to identify the optimal parameter configurations. The results indicated that the most effective parameter values were a drone altitude of 5-6 meters, a gimbal angle of 15°, and an ambient light intensity ranging from 0.6 to 0.9. The influence of drone speed on detection accuracy was also assessed. The analysis revealed that the detection accuracy for specific object categories declined at higher speeds. Fine-tuning the physical parameters was crucial in improving detection performance and mitigating these accuracy reductions. The findings suggest optimizing the drone’s physical parameters and velocity enhances object detection accuracy.
Kelek et al. (Sun,) studied this question.