Monocular Vision-Based Coastal Distance Estimation for GNSS-Denied AUV Operations: A Feasibility Study

The issue discussed in this article concerns navigation support, including that for autonomous underwater vehicles, which are subject to a systematic increase in positioning error when using dead-reckoning methods in GNSS-denied coastal areas. In such situations, the ability to capture images of the vehicle’s surface surroundings and use information about objects within the field of view of the optical system becomes useful. The proposed method uses a monocular camera and image segmentation to estimate distances to selected coastal reference objects. Determining the number of pixels occupied in the vertical plane by an object of known height enables the distance to be calculated using well-known mathematical formulas. The method was evaluated as a semi-automatic feasibility study using images acquired from a motorboat under maritime conditions. Forest regions of interest and the sea–land reference line used for building measurements were selected manually. The resulting estimates were evaluated against adopted reference distances derived from camera GPS coordinates and manually selected mapped reference locations. The reported differences describe agreement with these references, whose uncertainty was not independently quantified, and do not establish metrologically traceable absolute measurement accuracy. Image quality and segmentation errors affect the resulting distance estimates. The results indicate the potential of the method to provide auxiliary distance observations; however, deployment on an AUV and integration with its navigation system were not experimentally validated. This article describes the method and presents the results.