Moving Object Detection Using Hybrid Yolo-v8 with Improved GAN

In this paper, we propose a hybrid model combining YOLOv8 with improved Generative Adversarial Networks (GANs) to further improve performance detecting moving objects under challenging conditions. The improved GAN component generates realistic synthetic data, enriching the training dataset with varied and complex scenarios. This augmentation allows YOLOv8 to learn more effectively from the enhanced dataset, improving its accuracy and robustness in diverse environments. The GAN also aids in post-processing by refining the resolution and clarity of detected objects, reducing false positives and negatives. We evaluate the proposed hybrid model on the Visdrone, Tinyperson, and PASCAL VOC2007 datasets. The experimental results reveal that the hybrid YOLOv8 and GAN model significantly outperforms state-of-the-art models, achieving higher accuracy, precision, recall, and mean Average Precision (mAP) across various evaluation metrics. This demonstrates the effectiveness of integrating GANs with YOLOv8, providing a powerful and reliable solution for object detection in challenging conditions.