Density Adaptive Point Set Registration

Probabilistic methods for point set registration havedemonstrated competitive results in recent years. Thesetechniques estimate a probability distribution model of thepoint clouds. While such a representation has shownpromise, it is highly sensitive to variations in the den-sity of 3D points. This fundamental problem is primar-ily caused by changes in the sensor location across pointsets. We revisit the foundations of the probabilistic regis-tration paradigm. Contrary to previous works, we modelthe underlying structure of the scene as a latent probabil-ity distribution, and thereby induce invariance to point setdensity changes. Both the probabilistic model of the sceneand the registration parameters are inferred by minimiz-ing the Kullback-Leibler divergence in an Expectation Max-imization based framework. Our density-adaptive regis-tration successfully handles severe density variations com-monly encountered in terrestrial Lidar applications. Weperform extensive experiments on several challenging real-world Lidar datasets. The results demonstrate that our ap-proach outperforms state-of-the-art probabilistic methodsfor multi-view registration, without the need of re-sampling.