Stability Estimates of LIDAR Range Profile Feature Extraction Techniques under Random Time Shifts

Abstract

The problem of obtaining time-shift invariant features for the LIDAR range profiles is considered. Techniques under consideration are based on the Fourier transform, complex wavelet transform, and Karhunen-Loeve transform. For each feature extraction technique, a classifier is trained and the resulting recognition error rates are estimated. These error rates are obtained for various widths of the random delay window. The comparison of provided estimates allows selecting a technique adequate to the expected ranges of the random time shift.