This thesis presents the development of a method for automatic apple size estimation and orientation, needed in the context of automated harvesting. The reason for the increasing investment of time and resources in the design and creation of machines for autonomous agriculture is that their use reduces food and resources waste, improves productivity, lowers production costs. State of the art solutions examined during the first steps of the research focus on creating an accurate estimation of the shape of the original object with the use of deep learning models. However, these approaches are not suitable for real-time use in the fields, since their latency would not be helpful in increasing productivity.