Human detection with a four-engine unmanned aerial vehicle

In this thesis, it aims to compare the performance of YOLO (You Only Look Once) algorithms (YOLOv7, YOLOv8 and YOLOv10), which are among the latest technologies in the field of human detection, by using a data set obtained from photographs taken with a four-rotor unmanned aerial vehicle (UAV). The dataset consists of a rich collection of images containing various human positions, and the training process was carried out in the Google Colab environment. In this thesis, the training processes of each YOLO version are discussed in detail and comparisons are made. The performances of YOLOv7, YOLOv8 and YOLOv10 have been evaluated for human detection accuracy and overall efficiency. The performance of the algorithms was evaluated on criteria such as accuracy, sensitivity, sensitivity, F1 score and mAP50. YOLOv8 was found to be more effective than the other two algorithms with the highest accuracy (89.6%) and mAP50 (97.4%) values. YOLOv7 and YOLOv10, on the other hand, have accuracy rates of 82.8% and 88.0%, respectively. These findings provide an important basis for determining which algorithm is more effective in imaging and object recognition applications performed with unmanned aerial vehicles. The results of the study are a guide for future research and applications.