SDRSAC: Semidefinite-Based Randomized Approach for Robust Point Cloud Registration Without Correspondences

Abstract

This paper presents a novel randomized algorithm for robust point cloud registration without correspondences. Most existing registration approaches require a set of putative correspondences obtained by extracting invariant descriptors. However, such descriptors could become unreliable in noisy and contaminated settings. In these settings, methods that directly handle input point sets are preferable. Without correspondences, however, conventional randomized techniques require a very large amount of samples in order to reach satisfactory solutions. In this paper, we propose a novel approach to address this problem. In particular, our work enables the use of randomized methods for point cloud registration without the need of putative correspondences. By considering point cloud alignment as a special instance of graph matching and employing an efficient semi-definite relaxation, we propose a novel sampling mechanism, in which the size of the sampled subsets can be larger-than-minimal. Our tight relaxation scheme enables fast rejection of the outliers in the sampled sets, resulting in high quality hypotheses. We conduct extensive experiments to demonstrate that our approach outperforms other state-of-the-art methods. Importantly, our proposed method serves as a generic framework which can be extended to problems with known correspondences 1 .

Normal-Based Flower Pollination Algorithm (FPA) for Solving 3D Point Set Registration via Rotation Optimization

“…In the experiments, the deterministic methods that are used for comparison are ICP [10], ICPP, CPD [11], IRLS [38], and SDRSAC [16]. And the heuristic methods are ABC [39], bat-inspired algorithm (BA) [40], cuckoo search (CS) [41], DE [42], FPA [32], GA [43], HS [44], and particle swarm optimization (PSO) [45].…”

“…For all the heuristic methods, the number of function evaluations (FEs) is 40000, and each population is normalized in the range [-1, 1]. The maximum iterations of ICP, ICPP, and CPD are 60, and the parameter settings of IRLS and SDRSAC are referenced in the prior work [37,16].…”

“…To weaken the dependency on the initial position, a branch and bound (BnB) strategy was introduced to explore optimal Euclidean transformation [15]. Recently, a semi-definite and relaxation method was used in the stochastic sampling objective function to fulfill model alignment when initial position is complex [16]. A mutual matching based on dynamical adaptation between the color and the spatial information was used to estimate transformation parameters [17].…”

PHASER: A Robust and Correspondence-Free Global Pointcloud Registration

“…Global pointcloud registration approaches, such as the work of Rusu et al [3] and Lei et al [12] seek to efficiently find a unique solution by extracting point descriptors to find correspondences for the registration without prior knowledge. The work Le et al [4] proposes a randomized approach to overcome issues with computing descriptors and point correspondences. Yang et al [13] proposed a global variant of the ICP algorithm based on the branch-and-bound technique for global optimization.…”

“…Furthermore, the global alignment problem becomes significantly more challenging when the data contains different viewpoints yielding confined overlap between the pointclouds and many outlier correspondences. Alternatively, approaches which do not rely on correspondences are less affected by such issues and can potentially increase the overall system's robustness [4]. Also, it is evident that in many of the aforementioned cases, an assessment of the alignment quality is another critical aspect for several robotic applications, for instance, its uncertainty for incorporating loop-closure factors in graph-based SLAM.…”