A modified single image dehazing method for autonomous driving vision system

Managing unforeseen situations, particularly in low-visibility environments caused by weather degradation, continues to be a significant challenge for the autonomous driving vision system. This paper aims to enhance the visibility of degraded images captured by the system's sensors by removing haze. To achieve this goal, we propose an algorithm that predicts transmission from a regression model using random forest and atmospheric light using a quad-tree decomposition method. We evaluate the performance of the haze removal algorithm on three benchmark datasets (FRIDA2, D-HAZY, and RESIDE) using both quantitative and qualitative analyses. Our proposed method yields the lowest count of saturated pixels (∑) in blind contrast enhancement assessment, with ∑ = 0.0001. The implications of our approach are significant. By utilizing the RF-transmission estimation and quad-based atmospheric light prediction, the proposed haze removal algorithm demonstrates greater robustness in preventing unintended black or white color pixels in the dehazed image. This improvement can contribute to safer autonomous driving, particularly in low-visibility conditions, where the reliability of image processing systems is paramount.