EOD-Net: Enhancing Object Detection in Challenging Weather Conditions Using an Innovative End-to-End Dehazing Network

With the progress of deep learning, significant advancements have been made in solving computer vision tasks such as object detection and semantic segmentation. However, in real-world scenarios like autonomous driving, accurate object predictions become crucial and challenging under adverse weather conditions, including haze and fog. These conditions severely hinder the performance of computer vision algorithms, leading to potential risks and errors. This work proposes an original end-to-end Enhancement for an object detection network (EOD-Net), specifically designed to address dehazing. Our network employs multi-scale feature fusion and residual refinement modules to enhance the quality of dehazed images. We extensively evaluate our method on well-established dehazing benchmark datasets and demonstrate its superiority over several state-of-the-art approaches in terms of quantitative metrics. Additionally, through comprehensive computer simulations, we illustrate that our method significantly improves the accuracy of object detection models in dense haze conditions. This research contributes to the advancement of image dehazing algorithms and their practical applications in computer vision.