Evaluating Underwater Image Mosaicking Through Unreal Engine-Based Virtual Environment Simulation and Lapmv2

Abstract

Underwater image mosaicking is essential for marine research, seafloor mapping, and underwater archaeological documentation. It creates comprehensive visual maps by stitching image sequences from autonomous underwater vehicles (AUVs) or remotely operated vehicles (ROVs). To advance this technique, we are developing an automated mosaicking software for a multiAUV fleet. However, obtaining real underwater datasets with high-quality images and precise navigational metadata for performance validation of this software remains costly and challenging. To address this, we created virtual underwater environments in Unreal Engine 5, simulating 16 m×16 m and 40 m×40 m scenes with realistic lighting, visibility, and terrain features. Using the Large-Area Photo-Mosaicking (LAPM) tool for initial validation, we found that our synthetic data effectively replicates key challenges of underwater mosaicking, such as reduced visibility, color distortion, and vignetting. LAPM successfully processed our simulated imagery, validating the use of virtual environments for underwater algorithm development and highlighting further improvements needed in our custom software