Feature extraction of near-spherical fruit with partial occlusion for robotic harvesting.

摘要

For a fruit-picking robot in natural scenes, feature extraction of fruits occluded by leaves and branches based on machine vision is a key problem. In this study, the cluster barycentre (CB), edge barycentre (EB), circular Hough transform (CHT) and least square circle fitting (LSCF) are used to extract the features of fruit. The results indicate that the first two methods cannot accurately determine the circle in the presence of partial occlusion. The objects extracted by the CHT method include false targets in addition to longer time and larger memory required. The LSCF method, on the other hand, can accurately extract the features in a real-time mode. When the occluded area ratio is less than 52%, or the occlusion angle is less than 216°, the accuracy of feature extraction using LSCF can meet the requirements of the robot operation.