Multi-sensor data fusion framework and validation of algorithms with reference datasets

Abstract

Spatial data fusion algorithms are widely applied in dimensional metrology for quality assessment or surface reconstruction. Multi-sensor point cloud fusion combines the advantages of multiple sensors by merging their measurements into a single coordinate system and reducing the prediction uncertainty and systematic errors. Algorithms designed for these tasks employ many methods that require thorough evaluations through a common framework. To address this need, this paper proposes a framework for simultaneous registration and approximation, and introduces a reference data generator for unbiased evaluations of data fusion algorithms with heterogeneous and anisotropic noise assumptions for applications involving multiple sensors. The bias for the generated reference data is evaluated close to floating point accuracy, which validates the generation method, and uncertainty evaluation on ICP variants reveals that reference data is more suitable to evaluate point cloud fusion algorithms. The proposed framework and data generator allows developing and validating more accurate data fusion algorithms.