Urban arterial real-time performance measurement using privacy preserving mobile sensors

Abstract

Arterial performance measures, including intersection performance measures that assess the signal plan and the service quality of the intersection and arterial performance measures that monitor the road facilities between intersections, are crucial and beneficial to travelers and traffic engineers in urban areas. Most existing arterial performance measurement studies are based on data from fixed-location sensors (e.g., loop detectors), such as flow, occupancy and speed. However, fixed-location sensors are restrained for application in wide area due to the limited coverage of current arterial detection systems and the high cost for installation and maintenance. The recent proliferation of Global Position System (GPS) equipped vehicles and devices have led to the emergence and rapid deployment of mobile sensors, those that move with the traffic flow they are monitoring. Mobile sensors provide an important alternative way for traffic data collection, as they can reveal detailed behaviors and provide spatially continuous trajectories of vehicle, but only for a sample of the entire traffic flow.

The new data format calls for novel modeling approaches to estimate arterial performance measures. The objective of this doctorial research is to (i) model the traffic of arterial signalized intersections and corridors based on privacy preserving mobile sensor data; and (ii) to estimate intersection performance measures, including delays, queue lengths, and signal timing information using mobile data. The research also aim to investigate and solve some critical issues in arterial traffic modeling using mobile data, such as over-saturation and low penetration issue.

The proposed methods were validated using field experiments and micro-simulation data. The results show that mobile sensors can be a very valuable supplement of fixed location sensors in evaluating the performance of arterial traffic, especially for the congested traffic conditions in which most vehicles are delayed. This research also investigates the feasibility of solving traffic problems by employing advanced machine learning methods, which would be helpful to the mobile sensor based traffic modeling research in the future.

In this research, an arterial Virtual Trip Line (VTL) system is constructed to collect intersection delays and short vehicle trajectories when vehicles pass the intersection. Based on the delay measurements from this system, a linear programming method is proposed to estimate the intersection delay pattern that is defined as intersection delay of any vehicle in term of arrival time. The real time intersection delay pattern turn out to be piece-wise linear, and contains discontinuities and non-smoothness. As the discontinuity in delay pattern implies the signal timing information, a Support Vector Machine based method is then developed to figure out the boundaries of cycles and further the cycle by cycle signal timing information such as cycle length, and splits. The non-smoothness indicates the interface of congested flow and free flow. This feature inspires the idea of real time queue length estimation by detecting critical points in the intersection delay patterns. To address the significant randomness and non-stationarity of the arterial traffic flow, a probabilistic graphic model is constructed with stochastic assumptions on the arrival and departure processes of a signalized intersection. The hidden traffic flow that is not directly measured from the mobile sensors could be reconstructed statistically based on the proposed Bayesian Network. We also study the formulation and dispersion of platoons between intersections, which is critical in profiling arrival time patterns and estimating arterial corridor travel times.