Sub-pixel Underwater Object Size Measurement Algorithm Based on Improved Otsu Binarization and Edge Curvature Filtering

Abstract

In the field of target recognition, shape size is an important attribute of object. There have been many research bases on vision-based size measurement in air, but few studies have been done on underwater target size measurement. This paper mainly proposes solutions to the problems of external lighting conditions, the reflective properties of object materials, and the disturbance of robot motion. Firstly, based on the improvement of Otsu, an Otsu method suitable for bimodal or multimodal object images is proposed. Secondly, a linear interpolation denoising method based on edge contour curvature fluctuation is proposed, which combines polynomial interpolation sub-pixel technology to achieve contour edge smoothing. Finally, using polynomial interpolation sub-pixel technology, taking the image threshold segmentation effect and the measurement accuracy of the object size as the measurement standard, the underwater ball and pipeline are tested, and the measurement accuracy is improved by more than 2%, which verifies the robustness of the method.