Axial image restoration of a single dome window view for stereo vision in unmanned underwater vehicles (UUV)

Abstract Unmanned Underwater Vehicles (UUVs) are typically used for monitoring and inspection tasks where it’s critical to locate objects and create 3D models of the surroundings. One of the most dependable techniques for accomplishing 3D localization and modeling jobs is stereo vision. This isn’t the case in underwater conditions, though, as cameras will record distorted views, which results in inaccurate 3D estimation. In real-world applications, it is more practicable and a more difficult task to place the two cameras of the stereo vision system inside a single dome glass. The goal of this study is to recover the axial image views of the two cameras within a single dome window from the refracted views that were recorded. The cameras’ position (along the Z and X axes) and orientation (pan and tilt) within a single dome window—of which there are 448 possible combinations—are taken into account by the kinematics model. The approach has been improved and validated using around 18 million experiments. Lastly, when compared to the original axial images, our approach may report axial images with an average overlapped similarity of 96%.