Enhanced SDRSAC Algorithm for Robust Registration on Non-Cooperative Targets
Zhuoran Wang, Jianjun Yi, Lin Su, Yihan Pan · 2024
Efficiently registering point clouds of spatial targets or defunct debris is a formidable challenge. SDRSAC is a robust point cloud registration method based on correspondence-free approaches. In this paper, we propose an enhanced SDRSAC algorithm for rigid alignment of non-cooperative space targets. Our improvements to SDRSAC are twofold. First, we incorporate the difference in surface normal angles into the penalty matrix, allowing it to account for both distance and angle invariance under rigid transformations. This significantly reduces the possibility of ambiguous registrations in spatial targets. Second, we compute the planarity of local surfaces using eigenvalues from the neighborhood covariance matrix, focusing random sampling on flatter surface points. This approach lessens the impact of noise and outliers in real point clouds. Experimental results demonstrate that our method outperforms others in terms of registration accuracy and speed on non-cooperative space targets.