Removing Outliers from 3D Macrotexture Data by Controlling False Discovery Rate

Abstract

Measurement of pavement macrotexture by noncontacting means is often contaminated by erroneous readings made by the instrument. These errors can be caused by extreme diffusion or refraction of the light source by aggregate or bitumen and are manifest in the data as outliers in both the positive and negative directions. In three dimensions, these errors can manifest themselves along the width and length of the measured profile as singularities or packets of continuous data. The problem is confounded by the constantly changing nature of pavement surfaces and large quantity of data gathered in three-dimensional (3D) applications. The identification and treatment of outliers proposed in this work is a new method that effectively treats outliers while continuously adapting to the surface measured. This is done by controlling the rate of false discoveries (measurements incorrectly identified as outliers) without affecting adjacent, correct, measurements.