Perception Enhancement for Autonomous Vehicles in Foggy Conditions | ||
International Journal of Applied Intelligent Computing and Informatics | ||
Article 1, Volume 1, Issue 2, September 2025, Pages 1-10 PDF (597 K) | ||
Document Type: Original Article | ||
DOI: 10.21608/ijaici.2025.348797.1005 | ||
Authors | ||
shimaa ragab hamdy* ; Moustafa Elgendy; Ahmed Shalaby | ||
computer science,benha university | ||
Abstract | ||
Autonomous vehicles face significant challenges in adverse weather conditions, such as fog, which reduces visibility and degrades the quality of camera-captured images, making it difficult to detect targets and obstacles. Developing an effective image defogging algorithm is crucial to enhancing the optical system’s ability to adapt to varying environmental conditions. In this research, we apply an existing autoencoder-based approach to address the fog removal problem, as it has demonstrated fast and efficient results. Unlike traditional methods that train the model on the original data, this approach trains the model on the noise present in the input data, as noise is simpler to regenerate than the original data. Noise reduction is then performed by subtracting the extracted noise from the noisy input. To further enhance the model's performance, we focus on improving the feature extraction process by incorporating architectural optimizations, such as the use of skip connections and advanced data augmentation techniques. Various optimizers are also compared to achieve better accuracy. The model's performance is evaluated using SSIM and PSNR metrics, with results showing PSNR and SSIM values of 26.03 and 0.939, respectively, for the outdoor dataset, and 29.87 and 0.966 for the indoor dataset when applied to the RESIDE dataset. | ||
Keywords | ||
Autonomous vehicles; Adverse weather conditions; Perception; sensing; Deep learning | ||
Statistics Article View: 29 PDF Download: 18 |