State-dependent self-adaptive sampling (SAS) method for vehicle trajectory data

Mobile sensors provide new opportunities for traffic monitoring and data collection, which at the same time also impose challenges on data storage and transmission from the perspectives of the sensors (users). In this paper, we present a self-adaptive sampling (SAS) method for mobile sensing data collection. It uses the estimated vehicle flow states from individual trajectory to adjust the sampling rate. The methodology uses a Hidden Markov Model (HMM) classifier to classify individual data points into one of four basic vehicle flow states (Free Flow, Stopped, Acceleration, and Deceleration) and a Support Vector Machine (SVM) classifier to identify 'stop and go' segments in the trajectory. The identification of vehicle flow state is used for varying the sampling rate. The proposed SAS method can reduce the total amount of data by 67-77%, while keeping the most critical data points. The reduced data using SAS also show promising result in traffic modeling applications.