کنفرانس مهندسی برق ایران

صفحه اصلی / سی و دومین کنفرانس بین المللی مهندسی برق

Real-Time Object Detection and Depth Estimation in Quadcopters through Intelligent Image Processing with YOLOv8

نویسندگان :

Amir Mahdavi¹ Mojtaba Mohsen Haghighi² Saeed Khankalantary³

1- دانشگاه خواجه نصیر الدین طوسی 2- دانشگاه خواجه نصیر الدین طوسی 3- دانشگاه خواجه نصیر الدین طوسی

کلمات کلیدی :

Quadcopters،object detection،YOLOv8،Real-time،object tracking،depth estimation

چکیده :

Recognizing objects in images captured by quadcopters poses a significant challenge due to the dynamic nature of both the quadcopter and the objects. This study presents novel enhancements to the YOLOv8 model for object detection in quadcopter images. By systematically testing and selecting optimal values for each hyperparameter, the model's accuracy is significantly improved. Furthermore, this article utilizes specialized networks for object tracking and distance detection, in place of traditional methods such as using two cameras. The study also addresses the need for both speed and accuracy in object tracking by introducing a novel method that employs two networks to determine object position and distance. Extensive testing on the VisDrone2019 dataset demonstrates a 3.0% increase in mAP50 and a 1% increase in mAP50-95 compared to the original YOLOv8 algorithm. Additionally, the study identifies a high-performing model for distance estimation. These contributions significantly improve object detection and distance recognition techniques in quadcopter images, addressing the challenges posed by dynamic environments and small objects.