Relative pose estimation is crucial for automating visual surveys for hull inspection, as it enables precise visual mapping of the hull surface. In recent research, sonar sensors have been used to estimate the relative pose to the hull surface. However, the requirement that cameras must maintain a close range to the hull surface for precise mapping can be a disadvantage for sonar sensors, since their precision degrades at close range owing to underwater acoustic characteristics.
This study addresses a stereo vision based relative pose estimation framework that can be utilized at close range while reducing system complexity and vehicle cost. Challenges in underwater stereo matching are dealt using series of computer vision techniques. To utilize the stereo vision system as a sensor to measure pose relative to the hull surface, we systematically derive the resulting relative pose and its uncertainty.
To demonstrate and evaluate the effectiveness of the proposed stereo-vision-based relative pose estimation methodology, experimental validation with stereo images are shown. Synthetic stereo images and actual stereo images obtained by autonomous underwater vehicle (AUV) are used for experimental validation. The validity of proposed algorithm of estimating relative pose and its uncertainty is confirmed by analyzing the errors.