Mobility Modes Awareness from Trajectories Based on Clustering and a Convolutional Neural Network

Rui Chen,Jianguang Wang,Xiang Yao,Wanli Li,Mingjian Chen

doi:10.3390/ijgi8050208

Rui Chen, Jianguang Wang + Show 3 more

Open Access

https://doi.org/10.3390/ijgi8050208

Copy DOI

Abstract

Massive trajectory data generated by ubiquitous position acquisition technology are valuable for knowledge discovery. The study of trajectory mining that converts knowledge into decision support becomes appealing. Mobility modes awareness is one of the most important aspects of trajectory mining. It contributes to land use planning, intelligent transportation, anomaly events prevention, etc. To achieve better comprehension of mobility modes, we propose a method to integrate the issues of mobility modes discovery and mobility modes identification together. Firstly, route patterns of trajectories were mined based on unsupervised origin and destination (OD) points clustering. After the combination of route patterns and travel activity information, different mobility modes existing in history trajectories were discovered. Then a convolutional neural network (CNN)-based method was proposed to identify the mobility modes of newly emerging trajectories. The labeled history trajectory data were utilized to train the identification model. Moreover, in this approach, we introduced a mobility-based trajectory structure as the input of the identification model. This method was evaluated with a real-world maritime trajectory dataset. The experiment results indicated the excellence of this method. The mobility modes discovered by our method were clearly distinguishable from each other and the identification accuracy was higher compared with other techniques.

Highlights

The mobility behavior of users is the key factor for understanding the spatiotemporal characteristics of human activity, transportation conditions, and environment
We propose to extract route patterns by means of origin and destination (OD) points clustering based on Ordering points to identify the clustering structure (OPTICS)
We propose a deep learning approach to identify mobility modes based on a convolutional neural network (CNN)

Summary

Introduction

The mobility behavior of users is the key factor for understanding the spatiotemporal characteristics of human activity, transportation conditions, and environment. Massive travels by users can be categorized into different modes, such as transportation modes (subway, bike, taxi, and bus, etc.) [1] and different frequent route patterns [2]. Distinguishing different mobility modes and understanding the character of them play a key role in location-based services such as destination and route prediction [3,4] and analysis of travel behavior [5], etc. The recent literature on trajectory mining mainly focused on significant locations discovery, anomaly detection, location-based activity recognition, and mobility modes identification [1,8]. Our work considers the information of both travel activity types and route patterns to categorize different mobility modes. A person travels from location A to location B by bus/taxi, or a vessel travels from port A to port B for tugging/carrying cargo

Methods

Results

Conclusion