A novel algorithm for segmenting fruit from unorganized point clouds

This paper proposes an efficient and robust four-step process to extracting fruit shape from background. At first, point clouds is divided into octree cells by an adaptive subdivision; Second, converting teach octree cell into a splat and approximating its local surface with MLS; Third, estimating geometric properties of each splat by Principal Component Analysis (PCA) and covariance analysis; Forth, a recursive incremental process is employed to finding geometric similarity splats and generating ultimate splats set of fruit. Paper provides, finally, eleven angles range scans of an apple and demonstrates segmentation of apple shape from leaves. The result proves the validity and practicability of this method in segmentation.