Monocular Metric Distance Estimation in Maritime Scenes via Reference-Based Scale Recovery

Nazario Pizzicoli, Matej Fabijanić, Guido Lazzerini, Alessandro Ridolfi, Fausto Ferreira

Keywords: Monocular depth estimation, Metric scale recovery, Maritime perception, Reference-based calibration

TL;DR: Metric scale recovery from maritime images up to 100m using DepthAnythingV2 and known reference points.

Abstract: Ranging in maritime environments is essential for collision avoidance and situational awareness. Radars and the Automatic Identification System (AIS) provide long-range positional awareness, whereas monocular cameras could offer a low-cost alternative for short to mid range perception if paired with a method to estimate distance to objects. This work investigates whether pretrained foundation monocular depth models preserve a geometrically proportional relationship between predicted relative depth and true metric distance that enables simple scale recovery. We apply a single-frame, reference point based inverse-depth calibration using least squares regression to convert relative depth predictions into metric distance estimates. The approach is evaluated in two settings: a controlled indoor pool experiment with static reference buoys and an outdoor lake experiment using a surface vehicle equipped with a GPS. The indoor experiment achieves approximately 0.38m RMSE and relative error of 5.65\%, while the outdoor experiment yields 6.79m RMSE with a relative error of 5.95\% with distances up to 100m. Results indicate that pretrained relative depth models exhibit sufficient geometric consistency to support coarse metric distance estimation when suitable reference anchors are available.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 10