Description of Invention
Natural atmospheric phenomena like fog, haze, and smog can degrade road image quality, affecting road marker visibility crucial for Advanced Driver Assistance Systems (ADAS). Current dehazing methods like contrast stretching, guided filtering, and dark channel priors are limited in handling diverse haze patterns. Deep learning-based approaches have shown promise with Residual Neural Networks (ResNet) being efficient for tasks like object detection and classification. We propose a ResNet-based dehazing method using Atmospheric Scattering without explicit transmission matrix, improving PSNR and SSIM over state-of-the-art techniques on various haze patterns while preserving scene details. Introducing a novel ResNet-based dehazing approach using the Atmospheric Scattering Model (ASM) that eliminates the need for separate computation of transmission matrix and atmospheric light. Through validation and fine-tuning, this methodology substantially improves the Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM) compared to current state-of-the-art dehazing techniques. In summary, the proposed ResNet-based dehazing technique exhibited strong performance on multiple datasets, including NYU2, RESIDE, and FRIDA. Its validation on real-world road images from Cityscapes showcased notable results, particularly in enhancing marker-road contrast for potential improvements in marker classification algorithms in future work.