Grape berry calibration by computer vision using elliptical model fitting

In the context of vineyard and wine process management, producers are looking for novel tools to estimate the growth stage and potential quality of vine grapes. Several studies show the importance of the berry size as a growing and quality indicator. This paper proposes an image processing method to estimate the berry size of vine grapes by image acquisition in the field. The goal is to recover the profile of each visible berry in the grape, including partially occluded ones. In the present approach, an elliptical contour model is used to recover the complete shape of each berry from its visible edges. This method combines a primary edge detection and cleaning process based on a watershed algorithm, and a direct least square ellipse fitting algorithm.