Vehicle Detection in Unmanned Aerial Imagery Through Advance You Only Look Once Architectures
U Nishitha, V Lokesh, Tata Kaushik, Rimjhim Padam Singh · 2024
The detection of vehicles by unmanned aerial vehicles (UAVs) is gaining substantial attention in the field of traffic control. In aerial images, vehicles have a unique perspective and take up less pixels than other objects in the broader data set.. This study uses the YOLOv8 (You Only Look Once version 8) deep learning architecture to propose an enhanced method of vehicle recognition in aerial photos. This work goes beyond the conventional YOLOv8 implementation by introducing new methods for data augmentation to enhance the efficiency and the model’s resiliency. To build a diverse and enriched dataset, the suggested methodology makes use of cutting-edge techniques such image cropping, contrast correction, and brightness improvement. Training the YOLOv8 model on this enhanced dataset produces encouraging results; on a different test set with ground truth annotations, a noteworthy achievement of ${7 0 \%}$ accuracy in vehicle detection was made. This accomplishment highlights the effectiveness of our strategy and shows a notable increase in accuracy over earlier approaches. The study demonstrates the potential of YOLOv8 in tackling the difficulties of aerial vehicle detection and highlights the crucial role that data augmentation plays in improving model performance and resilience.