Abstract: In order to deal with point sets with sharp features or non-sufficient sampled points,this paper presents a segment driven feature-sensitive triangulation algorithm as an extension to general advancing front technique,to reconstruct high quality triangle meshes from unorganized,un-oriented point sets.With a smoothness threshold,the triangulation process is divided into two stages as segment reconstruction and stitching reconstruction.The segment reconstruction triangulates smooth regions starting from smooth seed triangulations,and by classifying elements and detecting features,filers out boundary outliers and relocate feature vertices in order to further advance fronts of these smooth segments.The stitching reconstruction stitches the gaps among segments or smooth regions of segments while advancing fronts of all segments,and relocates or inserts new feature vertices by intersections of local smooth surfaces.With this two-stage triangulation technique,the algorithm can robustly treat point sets with sharp features or bad sampling condition such as insufficient sampling,and especially avoids the expensive computation such as covariance analysis and intersection tests used in the general advancing front technique.Experiment results show that feature-sensitive quality triangulations can be achieved.