Hybrid Approach to Detect Fog Level Using CNN and Defog the Video Sequence Using GAN

Fog significantly reduces video clarity by lowering contrast, blurring objects, and distorting colors hampers scene understanding in critical applications such as autonomous driving, traffic surveillance, and remote sensing.Existing defogging methods like Retinex-based algorithms and dark channel prior often fail under varying fog densities and lack real-time adaptability, leading to detail loss or visual artifacts.To address these challenges, this study introduces a hybrid deep learning approach that integrates a Convolutional Neural Network (CNN) for fog level detection and a Generative Adversarial Network (GAN) for adaptive video defogging.The CNN accurately estimates fog density per frame, enabling the GAN to adjust its processing and preserve essential visual features.Preprocessing steps such as grayscale conversion and histogram equalization, enhance feature extraction and improve defogging performance.The system is designed for real-time deployment and adaptability across different atmospheric conditions.Evaluation using metrics such as PSNR, SSIM, FADE, NIQE, MAE, and RMSE demonstrates superior performance compared to existing and state-of-the-art methods like MSBDN and FFA-Net.Notably, under heavy fog, the proposed model achieved a PSNR of 26.0 dB, SSIM of 0.86, FADE score of 0.41, and runtime of 0.31 s/frame, confirming its efficiency and suitability for safety-critical, lowvisibility environments.