Abstract: Variations in lighting across tunnel sections alter drivers′ fields of view, causing misperception of information and compromising safety. Additionally, dynamic changes in field contrast induce visual fatigue and reduce perceptual ability. To investigate how field contrast and its gradients influence visual performance under dynamic tunnel lighting, the authors develop a "dynamic gradient-field contrast versus visual performance" analytical framework using a multi-factor controlled experiment, in which target position, color, luminance, and contrast are systematically varied. A random forest-Shapley additive explanation (RF-SHAP) interpretable machine learning algorithm elucidates nonlinear interactions among these parameters. The results indicate the following: (1) A visual field contrast inflection effect exists: in the threshold zone, contrast of 0.2－0.4 reduces reaction time by 20.5%, whereas increasing contrast by 0.2 beyond 0.4 reduces it by only 2.3%. (2) Brightness and contrast show a strong interaction: in the interior zone, increasing background luminance to 8.75 cd/m² compensates for low-contrast signals, raising recognition success to 70% (a 30% improvement over 4.5 cd/m²). (3) Sensitivity at the visual field periphery (30°) exceeds that of the central area (5°); when low contrast (0.2) combines with high luminance (6 000 cd/m²) at 30°, reaction time increases by up to 53.1%. (4) Recognition efficiency for warm-colored targets (red) is 14.2% higher than for cool colors (blue), with chromatic differences exerting the greatest effect under low contrast (0.1). (5) RF-SHAP analysis identifies visual field contrast and luminance as dominant factors, displaying U-shaped influence curves and confirming the practical applicability of dynamic lighting thresholds in engineering design.

Key words: tunnel lighting, dynamic light environment, visual field contrast, visual performance, random forest, Shapley additive explanation value