Travel Time Estimation Using Probe Data on Signalized Arterial

Abstract

With the increased market penetration of mobile devices, probe-based travel time collection techniques that use wireless technologies such as Bluetooth, Wi-Fi, and GPS are attracting more interest in developed nations. Compared to conventional point detectors, the probe-based systems have merits in that they not only obtain direct link travel times but also require a low budget. However, outliers and short-term biases are the two main issues to be addressed, especially when a probe-based system is deployed on signalized arterials. Outliers are inevitable, mainly due to access points (gas stations, stores, etc.) and exit/entry routes on the corridor. Short-term biases mostly occur from traffic signals on the route that make some vehicles stop and allow others to move. This phenomenon can be more significant under the conditions of a low probe sample, short signal spacing, many signals, and so on. In this paper, techniques for resolving the outlier and bias problems on a signalized arterial are proposed. To address the outlier issue, the Ferguson statistical test, which is designed to iteratively remove outliers until no more deviated values exist, was applied; to tackle the bias issue, the Loess smoothing technique, also known as locally weighted polynomial regression, was applied. Data used for this study were manually obtained by a license plate matching technique on a signalized suburban arterial near Seoul in Korea. Two-day-long data with a span of six hours on each day including the morning peak contained many outliers and were filtered using the Ferguson test. Subsequently, the Loess smoothing technique was applied to the filtered data with different sampling plans: 25, 15, and 5%. As a result, travel time accuracies were markedly enhanced in the lowest sampling rate of 5% and the accuracy differences were proven to be statistically significant, while the opposite was found for other sampling plans. The proposed methods are expected to be applied for generating reliable travel times in probe-based systems in practice.