Proposal for volume estimating of cherry tomatoes based on 3D point cloud

This paper proposes a method for estimating the volume of cherry tomato fruits using an RGB-D camera for the quantitative evaluation of their growth. In the proposed method, the fruit region is extracted by applying SegNet to the orthoimage generated from a 3D point cloud, enabling the reconstruction of the fruit’s 3D shape. The volume is then estimated by approximating the 3D point cloud with an ellipsoid using the least-squares method. Experimental results show that the volume estimation error for cherry tomatoes not occluded by leaves was -0.01±0.53 ml, whereas for those occluded by leaves, the error was 18.66 ± 14.33ml. For cherry tomatoes occluded by leaves, additional shape reconstruction was required for the occluded region.