Adaptive Sampling Methods for Vehicle Trajectory Data

Thesis (Ph.D.)--University of Washington, 2019The ubiquitous use of smartphones and the emergence of new technologies such as ridesourcing and connected/automated vehicles provide new opportunities for mobile sensors in traffic monitoring and data collection. To make GPS based smartphones an effective and practical source of transportation data, one needs to address the multifaceted challenges related to mobile sensing. Since the mobile sensing technology requires individual sensors (often from end-users) sending location information periodically to the data collector (e.g., a server), one of such challenges is the storage and data transmission cost incurred to individual sensors/users, as well as the battery life of mobile sensors. This research aims to balance the data transmission cost and the needs to collect detailed mobile sensing data by developing a method to resample the smartphone-based GPS data at the user side. The work introduces the concept of Vehicle Flow State (VFS) to explain the implicit nature of the probe vehicle’s motion. Then the work proposes a methodology which first estimates the vehicle flow state (VFS) of the sensor/vehicle from its trajectory data and then uses the estimated VFS to adjust the sampling rate of the trajectory accordingly. The primary contributions of this work are as follows. First, this work develops the concept of vehicle flow state (VFS) and developed an HMM-based method to identify the VFS of an individual vehicle. Second, two self-adaptive sampling strategies for vehicle trajectory data are presented based on the identified VFS, which reduces the overall data size and transmission cost. Finally, this work presents comprehensive testing and validation of the proposed methods with real-world trajectory data. The methods and algorithms provided in this work will be of significant value to the server-side and the user/client side of a smartphone-based vehicle trajectory data collection system. The reduced data using proposed methods show a promising result in traffic modeling applications (such as queue length estimation) and the end user’s privacy protection