Automatic Registration of Point Clouds by Combining Local Shape Descriptor and G4PCS Algorithm

Registration is usually the first step for the usage of point cloud data. Registration of point clouds is crucial in numerous practical applications. In order to quickly and accurately align point clouds, we combine the local shape descriptor and generalized four-point congruent set (G4PCS) algorithm. A descriptor with high computation efficiency is chosen to establish the correspondences. Then, the constraints of the G4PCS algorithm are applied and three angular constraints are proposed. These constraints are divided into three groups according to their computation complexity. The three groups of constraints are used to accelerate the searching process of four correct correspondences. The experiments are performed to compare our method with several existing methods. The experimental results show that our registration method has the best registration precision and computation efficiency. The introduction of the local shape descriptor can enhance both registration precision and computation efficiency. The proposed three angular constraints can also further boost the search process.