Point Cloud Segmentation to Approximately Convex Surfaces for Fruit Recognition

A fruit recognition approach based on segmenting the point cloud acquired by a 3D camera into approximately convex surfaces is considered. A segmentation approach which transforms a depth image into a triangular mesh and then segments this mesh into approximately convex segments is applied to depth images of fruits on trees. An analysis of the results obtained by this approach is performed with the intention to determine how successful the studied method is in detecting fruit as separate objects in a point cloud. The reported analysis gives a valuable insight into the potential applicability of the tested methodology in the preprocessing stage of a fruit recognition system as well as its drawbacks.