Efficient and robust 3D perception of seafloor targets using a single acoustic camera for low-visibility marine environments

Accurate and reliable three-dimensional (3-D) perception in low-visibility seafloor environments is essential for advancing underwater investigations. This study introduces a novel and efficient approach for rapid 3D seafloor target perception using a single acoustic camera (a forward-looking sonar). Unlike previous methods, the proposed approach eliminates the reliance on complex 3-D reconstruction theories and assumptions about target geometry, making it more robust and rapid. The information required for 3-D geometric solving is directly derived from the interaction between sonar motion and the acoustic imaging mechanisms. By integrating visual clues such as highlights and shadows in sonar images with sensor motion information, the 3-D bounding box of each seafloor target can be effectively inferred. The effectiveness of the proposed approach was validated through both simulation and real water tank experiments. The results demonstrate its robustness and accuracy in perceiving both structured and unstructured targets, highlighting the practicality of this approach in real-world underwater environments. This study achieves a 3-D perception of seafloor targets through rapid estimation of their bounding boxes, significantly improving robustness and efficiency while meeting the requirements of most marine engineering tasks. Potential applications include marine environment monitoring in dark conditions and autonomous navigation of marine robots in turbid water environments.