An Effective Binocular Vision-based Sensing Approach for IAUV

IAUV (Intervention Class AUV) is a challenging field for AUV's new development, suitable for underwater manipulation with certain target in limited environment. Visual sensor makes vivid description of the scene, so could be helpful in solving key issues in IAUV autonomous manipulation. Existing visual sensing approaches for IAUV are mostly applicable for target with prior knowledge and may be lack of robustness and flexibility. Here a binocular sensing approach is proposed using texture-control based pyramid NCC (normalized cross correlation) to extract target region or outline, and target range. Making use of the texture difference between foreground and background, the target was extracted by texture control. Regions with low texture level would be discarded, so time was saved and matching efficiency improved. A series of other methods were also used to further quicken the approach and improve its accuracy. Its overall capabilities, especially the real-time performance, were quantificationally analysed through repeated experiments in lab. Finally the approach was validated in a water tank with both clear and turbid water.