Loop Detector Sampling Frequency Errors Affecting Length Based Classifications

Abstract

The classification of vehicles has multiple purposes, including, but not limited to, roadway planning, quantifying environmental impacts, and determining roadway deterioration. Length based classifications from loop detectors is a common way to classify the vehicles, where a loop detector is effectively a metal detector embedded in the pavement that can measure when a vehicle is over the loop. Loop detectors can operate from 60 Hz to over 1000 Hz. As this work demonstrates, when a loop detector is operating at a lower frequency, the sampling resolution can lead to large errors due to the large, discrete steps between observable measurements. This problem has been elusive since large quantities of ground truth speeds and lengths for vehicles are prohibitively difficult to collect. To bypass this problem a simulation model is developed to calculate the length measurement errors arising from a given detector's sampling period. The derivation of the sampling error can explain some of the length errors seen in vehicle classification. Finally, to empirically validate the model, this work uses a low sampling frequency LIDAR sensor (40 Hz) that is deployed concurrent with a conventional loop detector based classification station (300 Hz). The 40 Hz data are evaluated in the context of the 300 Hz data, since the relative impact of the sampling errors should be very small in the latter case. However, an additional complication arose because the spacing between detectors differs and introduces a secondary source of errors. As such, the LIDAR data analysis was inconclusive.