Robust Sensor Registration based on the Least Quantile of Squares under Inaccurate Data Association

Sensor registration, an inherent problem in multi-sensor data fusion systems, usually requires correctly associated data. However, the results of track-to-track association are usually inaccurate especially in the presence of sensor biases. In this paper, we focus on the problem of sensor registration under nonideal data association. Regarding misassociations as outliers, the least median of squares (LMedS) registration method was extended into the least quantile of squares (LQS), which can deal with even more than 50% percent of misassociations with a proper quantile parameter. Simulation results demonstrate the superior performance of the proposed method.