Tangerine Volume Estimation by RANSAC on Point Cloud Data

This work is focused on the development of a method of tangerine volume estimation based on the point cloud data. While monocular vision-based methods are often easier to implement and deploy in real world both in terms of algorithms and the sensors required, depth data is necessary to obtain precise volume estimate. In this work tangerines are approximated by ellipsoids, that are fitted to the point clouds in 3D. A dataset of real tangerines was collected and marked. For each tangerine the mass and volume were measured. RANdom SAmple Consensus (RANSAC) was used for the ellipsoid identification. A number of experiments were conducted with varying algorithm hyperparameters, including iteration number and the inlier threshold value. The output quality was assessed via comparing the prediction of the method with the real volume of the fruit. Experimental results suggest that the good enough performance could be achieved in real time with a portable computer.