A fruit recognition method for automatic harvesting

This paper presents a method to detect and recognize mature tomato fruit clusters on a complex-structured tomato plant containing clutter and occlusion in a tomato greenhouse for automatic harvesting purpose. A color stereo vision camera (PGR BumbleBee2) is applied as the vision sensor. The proposed method performs a 3D reconstruction with the data collected by the stereo camera to create a 3D environment for further processing. The color layer growing (CLG) method is introduced to segment the mature fruits from the leaves, stalks, background and noise. Target fruit clusters can then be located by depth segmentation. The experimental data was collected from a tomato greenhouse and the method is justified by the experimental results. Our experiments included severe self and stereo occlusion, which wasn't both included in the previous work.