Detection of feature lines in a point cloud by combination of first order segmentation and graph theory

We present a method to find closed feature lines which indicate sharp edges in a point cloud in the context of reverse engineering. We start with a first order segmentation which results in different point clusters. A weighted graph structure is built where vertices correspond to the resulting point clusters and edges connect neighboring clusters. By choosing the weights carefully, the minimum spanning tree gives, after removal of some particular edges, a first approximation of the feature lines. This approximation has many short branches, which we remove with a pruning algorithm. Since the resulting graph consists of many unconnected pieces of feature lines, we introduce an algorithm that grows a part of a feature line and connects it to another part of the same feature line. A final clean up results in a good polygonal approximation of the feature lines.