Volume measurement of tulip bulbs

Samenvatting

As part of the Bollen 4.0 project, this study researched the possibility to measure volume of tulip bulbs using RGB-D images. Bulbs were segmented using a deep learning model. Given these segmentations, several volume estimation methods were developed and evaluated. Surprisingly, the method with the highest correlation to the ground truth volume did not make use of the depth information (R2 =0.65). This could be an indication that there is too much noise to get sufficient predictive values from the (depth)-images, of which poor stitching, badly peeled bulbs and incorrect segmentation by the deep learning model have been identified as possible causes.