Research on surface reconstruction of turbine blades based on 4PCS registration based on edge key point extraction

Point cloud data registration is a key technology in the fields of three-dimensional reconstruction in reverse engineering and pattern recognition in artificial intelligence. Although it has undergone continuous updates of point cloud acquisition equipment and reverse reconstruction algorithms, there is still a problem of inaccurate alignment of point cloud data from different perspectives after spatial transformation. This paper proposes a point cloud registration method based on edge feature point extraction, using turbine blades with multi-surface gradients and small inflection points as the measured object, and improving the registration time and low accuracy during the registration process as the evaluation criteria. The optimal projection plane was calculated using principal component analysis and projected to this plane. The vertices of the polygon were extracted as edge key points by using the minimum convex polygon method, which significantly reduced the search space of the 4PCS algorithm, and completed the final registration of the turbine blades with GICP fine registration. Through experimental comparison, this method has significantly improved the registration time and registration accuracy compared with Super4PCS coarse registration plus GICP fine registration and SAC-IA coarse registration plus GICP fine registration, which has improved the registration efficiency of point cloud data reverse fitting.