Bridging the gap between image restoration and navigational safety in hazy conditions: A new visibility estimation metric for maritime surveillance
Wentao Feng, Guobei Peng, Wengang Mao, Ryan Wen Liu · Ocean Engineering · 2026
Visibility distance is a critical factor for maritime navigational safety, as it determines the effective observation range of shipborne and shore-based monitoring systems. Under hazy conditions, the degradation of visual information significantly reduces the observable distance, leading to increased navigational risk and economic loss. Although numerous image dehazing methods have been developed to enhance visual quality, existing image quality assessment (IQA) metrics (e.g., PSNR, SSIM, FSIM, FADE, NIQE, etc.) fail to provide a physically interpretable link between image restoration quality and actual navigational safety thresholds, even though the dehazing performance could be undoubtedly obtained in maritime practice. This limitation arises because these objective metrics do not reliably reflect absolute visibility levels or effective observation range. Motivated by the strong correlation between navigational safety and visible distance, this work proposes a visibility-oriented evaluation framework that links image enhancement performance with practical visibility estimation. In particular, to alleviate the issue of lacking hazy and clear image pairs for dehazing networks, a Maritime Simulated Visibility Dataset (MSVD) is constructed using the Unity3D physics engine to simulate maritime traffic scenes under graded visibility conditions. The dataset provides paired hazy–clear images together with precise visibility annotations, enabling quantitative analysis of visibility restoration. Besides, a new dehazing visibility evaluation metric is proposed by leveraging object detection performance as an intermediate indicator. By establishing the relationship between visibility distance and detection accuracy, the proposed metric translates improvements in image restoration into measurable visibility gains. Six different dehazing methods are then utilized to perform visibility restoration. The dehazing performance and visible distance estimation are quantitatively compared using traditional image quality assessment metrics and our visibility evaluation metric. Experimental results in different imaging conditions demonstrate that (1) our dataset MSVD provides a reliable benchmark for evaluating dehazing performance across graded visibility levels; (2) our visibility evaluation metric contributes to highly-reliable estimation of visible distance, thereby supporting a balanced trade-off between navigational safety and operational efficiency.