Journal of Engineering and Applied Sciences
ISSN: Online 1818-7803ISSN: Print 1816-949x
Abstract
Detection of vehicles from aerial images is a challenging subject due to the large image resolution with small targets and variant orientations. Unfortunately, there isn’t any dataset large enough to be suitable for training deep models. Therefore, we recognize COWC, large aerial image dataset to use in vehicle detection. In this project, the third version of popular YOLO is modified to vastly improve its performance on aerial data. We trained on a large amount of aerial images from COWC dataset. The proposed detector was able to achieve mAP = 95% on VEDAI dataset. It outperformed SSD and R-CNN. For the OIRDS dataset, we achieved mAP = 67% without any previous training.
How to cite this article:
Shighaf Abdallah, Omar Hamdoun and Assef Jafar. Detecting Vehicles using YOLO from Aerial Images.
DOI: https://doi.org/10.36478/jeasci.2020.3586.3592
URL: https://www.makhillpublications.co/view-article/1816-949x/jeasci.2020.3586.3592