A Trajectory Collaboration Based Map Matching Approach for Low-Sampling-Rate GPS Trajectories.

Wentao Bian,Xin Wang,Ge Cui

doi:10.3390/s20072057

Abstract

GPS (Global Positioning System) trajectories with low sampling rates are prevalent in many applications. However, current map matching methods do not perform well for low-sampling-rate GPS trajectories due to the large uncertainty between consecutive GPS points. In this paper, a collaborative map matching method (CMM) is proposed for low-sampling-rate GPS trajectories. CMM processes GPS trajectories in batches. First, it groups similar GPS trajectories into clusters and then supplements the missing information by resampling. A collaborative GPS trajectory is then extracted for each cluster and matched to the road network, based on longest common subsequence (LCSS) distance. Experiments are conducted on a real GPS trajectory dataset and a simulated GPS trajectory dataset. The results show that the proposed CMM outperforms the baseline methods in both, effectiveness and efficiency.

Highlights

With the development of positioning technology, massive GPS trajectory data has been continuously generated from vehicles such as cars, taxis and buses
The current map matching algorithms are developed to determine the correct path of low-sampling-rate GPS trajectories, based on various features, such as spatial [1,2], temporal [2], speed constraint [3,4] and turning information [5]
This paper proposes a collaborative map matching algorithm called collaborative map matching method (CMM) to address low-sampling-rate GPS trajectories based on trajectory clustering and resampling

Summary

Introduction

With the development of positioning technology, massive GPS trajectory data has been continuously generated from vehicles such as cars, taxis and buses. The current map matching algorithms are developed to determine the correct path of low-sampling-rate GPS trajectories, based on various features, such as spatial [1,2], temporal [2], speed constraint [3,4] and turning information [5]. These algorithms do not perform well when sampling rates are very low, i.e., the sampling rate exceeds three or four minutes, especially in a dense road network. Spatial features (e.g., distance similarity) and temporal features (e.g., speed similarity) [2,6] are sometimes ineffective in identifying the correct path between consecutive

Objectives

Methods

Results

Conclusion