Automatic City Region Analysis for Urban Routing

TL;DR

This paper analyzes urban human mobility patterns using large taxi GPS datasets to identify functional city regions, and demonstrates how this information can improve network data delivery efficiency in urban Delay Tolerant Networks.

Contribution

It introduces a novel quad-tree region division method and applies association rules to infer functional regions from mobility data in three cities.

Findings

Identified four types of city regions based on mobility patterns.

Achieved up to 183% improvement in data delivery ratio in DTNs.

Validated the approach across three different cities.

Abstract

There are different functional regions in cities such as tourist attractions, shopping centers, workplaces and residential places. The human mobility patterns for different functional regions are different, e.g., people usually go to work during daytime on weekdays, and visit shopping centers after work. In this paper, we analyse urban human mobility patterns and infer the functions of the regions in three cities. The analysis is based on three large taxi GPS datasets in Rome, San Francisco and Beijing containing 21 million, 11 million and 17 million GPS points respectively. We categorized the city regions into four kinds of places, workplaces, entertainment places, residential places and other places. First, we provide a new quad-tree region division method based on the taxi visits. Second, we use the association rule to infer the functional regions in these three cities according to temporal human mobility patterns. Third, we show that these identified functional regions can help us deliver data in network applications, such as urban Delay Tolerant Networks (DTNs), more efficiently. The new functional-regions-based DTNs algorithm achieves up to 183% improvement in terms of delivery ratio.