Robust Registration under Inaccurate Track Association in Multi-sensor Fusion

Sensor registration, which requires correctly associated data, is a significant problem in multi-sensor fusion. However, the result of track association is inevitably inaccurate in the presence of random errors, sensor biases, track loss and false tracks. In this paper, we address the problem of sensor registration under inaccurate track association. Regarding misassociations as outliers in the registration process, a robust registration method based on the least median of squares (LMedS) is presented. Furthermore, we analyze the special properties of misassociations, and propose a more robust registration approach on the basis of the least quantile of squares (LQS). Simulation results demonstrate the superior performance of the proposed methods.