A Distinctive Binary Descriptor and Two-Point RANSACWC for Point Cloud Registration

Abstract

Point cloud registration is a fundamental problem in many applications. The point cloud registration based on local shape descriptor has been widely researched. In order to further improve the performance of registration, a novel registration method is proposed in this article. First, a binary descriptor is designed to establish correspondences between two point clouds. The descriptor has high descriptiveness. Thus, more correct correspondences are established. Then, a 3-D transformation estimation technique is developed, in which multiple constraints are used to accelerate the computation. When the randomly selected correspondences do not satisfy the constraints, the iteration is skipped. Finally, the experiments are performed to analyze the descriptor and 3-D transformation estimation technique. The comparison with the existing descriptors is implemented on three datasets. The results demonstrate that our descriptor has better matching performance. As for the 3-D transformation estimation technique, the combinations of the constraints are first analyzed. The performance of different constraints is presented and the best combination is chose. The comparative results with the existing techniques demonstrate that the proposed 3-D transformation estimation technique can obtain better registration accuracy and computation efficiency.

Keywords

• Constraint
• local shape descriptor (LSD)
• point cloud registration
• 3-D transformation estimation technique