A benchmark for point clouds registration algorithms

Point clouds registration is a fundamental step of many point processing pipelines; however, most algorithms are tested on data that collected ad-hoc and not shared with the research community. These often cover only very limited set use cases; therefore, results cannot be generalised. Public datasets proposed until now, taken individually, few kinds environment mostly single sensor. For these reasons, we developed benchmark, for localization mapping applications, using multiple publicly available datasets. In this way, able to sensor can produce clouds. Furthermore, ground truth has been thoroughly inspected evaluated ensure its quality. some datasets, accuracy measuring system was reported by original authors, therefore estimated it our own novel method, based an iterative algorithm. Along data, provide broad problems, chosen different types initial misalignment, various degrees overlap, problems. Lastly, propose metric measure performances algorithms: combines commonly used rotation translation errors together, allow objective comparison alignments. This work aims at encouraging authors public instead ad-hoc, objectivity repeatability, two characteristics in any scientific field.