Robust regression applied to ultrasound location systems

Local positioning systems (LPS), specially those using ultrasound, are able to accurately estimate the location of persons or objects indoors. However, under certain circumstances, its accuracy can be strongly deteriorated by outlying noise. This paper analyzes and compares several strategies for robust trilateration, such as high-breakdown-point robust methodologies, as well as the parity space outlier detection procedure, which is commonly used in GPS. This analysis is performed by simulation in a typical ultrasound location system scenario based on the actual location of nodes in the 3D-LOCUS system [1]. It is shown how the traditional parity space outlier detection method overcomes robust methodologies when only one ranging error is present, whereas it is not able to detect two simultaneous faults. It is proposed a modification of the LTS robust estimation methodology that offers a good performance even when several range measurements are erroneous, due to multipath and occlusions effects. The complexity of the robust algorithms studied is low enough for being implemented in the 3D-LOCUS system without affecting its current 10 Hz update rate.