Maximum clique and maximum consensus are widely used formulations for robust point cloud registration. Maximum consensus seeks one transformation that aligns as many matches as possible, while maximum clique selects matches through pairwise distance tests. Maximum clique is often used as a substitute for or an approximation to maximum consensus, with empirical evidence suggesting a close relationship. However, how closely the two formulations agree is not fully understood. We study this relationship theoretically. We show that, at any fixed positive tolerance, a maximum clique can have arbitrarily large alignment error and be arbitrarily larger than a maximum consensus set. When rotation is fixed, we derive sharp error bounds and tests that exactly characterize maximum consensus for the translation subproblem. We also study group-$k$ consistency, which uses consistency of groups of $k$ matches, and determine the smallest group sizes that guarantee alignment error bounds independent of point cloud scale and geometry. These are $d+1$ for rotations and translations and $d+2$ when reflections are allowed, in dimensions $d\ge2$. Finally, our experiments on nearly 8,000 registration examples quantify the frequency of these limitations in practice and show how the derived bounds can be used to check whether a subset of matches can be aligned within a point tolerance.
翻译:暂无翻译