Precision underwater localization using pyramidal arranged AR markers and real-time image enhancement

Precise visual localization using augmented reality (AR) markers is crucial in underwater operations. However, conventional planar markers rapidly degrade under turbidity, refraction, and limited viewing angles. This paper presents a task-optimized vector-based multi-marker configuration, strategically inclined (0°–20°) around a central target, integrated with a real-time enhancement pipeline combining Gray-world balancing, Retinex, CLAHE, un-sharp masking, and guided filtering. Our novelty is a task-oriented system-level coupling of three elements: (i) an inward-inclined pyramidal multi-marker array that stabilizes depth sensitivity, (ii) an ROI-selective six-stage enhancement pipeline that preserves marker edges under turbidity, and (iii) a refraction-aware PnP model for metric accuracy. This co-designed integration is distinct from planar-marker or restoration-only pipelines and is systematically validated across terrestrial trials, physically grounded Jaffe–McGlamery turbidity simulation, and real tank experiments. Experiments were conducted in three stages—terrestrial trials, simulated turbid conditions, and real underwater environments. On land, the proposed marker inclination and fusion reduced detection errors by 28% compared with a baseline single marker (0°, no enhancement). In simulated turbidity, the region-selective enhancement raised recognition rates from 78.35% to 99.8%. In real underwater trials, the mean detection error decreased from 28.49 mm to 7.90 mm, corresponding to a 72.28% reduction ( p < 0 . 001 ). These results demonstrate that the proposed system substantially improves robustness and accuracy, offering practical advantages for underwater docking, inspection, and construction tasks. • Conical multi-marker arrangement and image enhancement reduced the distance error from 28.5 mm to 7.9 mm in underwater experiments. • AR marker recognition rate improved from 78% to 99.8% in simulated turbid water. • ROI-based enhancement achieved real-time processing of 4K underwater images at 6.3 frames per second. • Low-cost system enables robust underwater localization for diverse marine tasks.