A Novel Sky-Aware Dehazing Approach for Outdoor Autonomous Systems

Clear outdoor images are essential for autonomous driving and accurate target detection, especially in haze. The majority of algorithms are unable to adequately address the issue of dehazing, resulting in a range of distortions, particularly in the sky area. This paper proposes an advanced dehazing algorithm for enhancing sky-area visuals (ESV). We segment the image into sky and non-sky areas, with atmospheric light levels being determined within the sky area. To enhance the recovery of the sky region, we suggest fine-tuning the sky's transmission to a predetermined constant value. Ultimately, the dehazed image is retrieved utilizing the atmospheric scattering model. Extensive experiments have shown that our proposed algorithm outperforms alternative methods, increasing PSNR by up to 1.3733%, 1.6360%, 2.4169%, 0.9512%, SSIM by up to 4.8995%, 0.6281%, 6.5335%, 8.7165%, enhancing the visuals of sky-area, compared to DCP, CAP, HC-CEP and AOD-Net.