, University of Passau, Germany
, Cornell University
, Lancaster University
Pages: pp. 15-17
Abstract—Much of the past pervasive computing research has been devoted to systems composed of a small number of devices that interact with a single individual or a small group of users. Body-worn sensors and sensors embedded in the user's environment were used to infer user location, activity, and information about the user's immediate surroundings, shaping the concept of context awareness. This issue is motivated by the fact that as technology becomes truly pervasive, we must proceed from the "single user, single system" perspective to "large-scale heterogeneous systems" involving many devices and many individuals collaborating over different spatial and temporal scales.
Keywords—large-scale opportunistic sensing, sensor networks, people flow, social networks, traffic prediction
Much of the past pervasive computing research has been devoted to systems composed of a small number of devices that interact with a single individual or a small group of users. Body-worn sensors and sensors embedded in the user's environment were used to infer user location, activity, and information about the user's immediate surroundings, shaping the concept of context awareness.
This special issue is motivated by the fact that as technology becomes truly pervasive, we must proceed from the "single user, single system" perspective to "large-scale heterogeneous systems" involving many devices and many individuals collaborating over different spatial and temporal scales. The technological drivers that facilitate such a perspective change are well known:
Thus, millions of sophisticated, networked, mobile-sensing systems are distributed across the world. Mechanisms for user motivation, community-based collection, evaluation, and utilization of vast amounts of sensor data can benefit from a continuously online, networked user base. People are also increasingly willing to contribute data for causes they consider worthwhile, provided their privacy concerns are respected.
The pervasive computing research community has leveraged the above developments to create novel large-scale sensing systems and applications. Well-known projects include MIT's reality mining, 1 the Sociometer project, 2 the MetroSense project, 3 with its concept of people-centric sensing, and the recent European FuturICT initiative ( www.futurict.eu). 4 On an abstract level, such large-scale sensing applications can be motivated by three considerations.
First is leveraging the geographical spread and mobility of users to collect data over large areas and time scales, as the articles in this special issue demonstrate. A well-known example is the OpenStreetMap project, which has built a detailed map of almost the entire world from user-contributed GPS traces. 5 Other examples range from the collection of data on seismic tremors to traffic prediction to road quality monitoring.
The second consideration is the use of collaborative sensing to improve accuracy and reliability. Combining data from many users can improve sensing and reasoning results. Thus, the OpenStreetMap's accuracy is much better in places where data from many users is available. However, collaborative sensing can also help with more traditional context-related reasoning. It has, for example, been shown that collaboration between physically adjacent users can dramatically improve the accuracy of pedestrian dead reckoning (PDR) for indoor navigation. 6 In this issue, the article by Nicholas D. Lane and his colleagues demonstrates the benefit of collaboration through social networks to improve inference results at a lower cost to users.
The third consideration is the sensing of collaborative, community-level phenomena. The ability to collect, compare, and combine data from many users opens up the possibility of studying and recognizing collective behavior, social structures, and community-level context. This research area, pioneered by MIT's Reality Mining and Sociometer projects, 1,2 is becoming increasingly popular. In this issue, the article by Richard A. Becker and his colleagues at AT&T exemplifies this type of work.
The idea of large-scale sensing has long been propagated by the wireless sensor networks (WSN) community. Although common issues clearly exist between WSN research and large-scale pervasive sensing, there are also fundamental differences. Most important is the transition from systems developed and deployed specifically for a given sensing task to "opportunistic" sensing piggy-backing on devices that are primarily being used for a different purpose. In general, this purpose is being a human user's personal appliance (mostly a phone). This means that system and application design must not just account for technical issues such as resource management, service discovery, and interoperability, but must also pay attention to various human factors—in particular, incentives for participation in the sensing, user interuptibility, and privacy concerns.
The articles assembled in this issue represent a broad range of the applications and challenges of large-scale pervasive sensing.
The first article, "A Tale of One City: Using Cellular Network Data for Urban Planning," by Becker and his colleagues, considers opportunistic use of mobile phone call detail records to derive maps of urban activity. The authors use mobile phone data collected over three months in a suburban city to show how geographical distributions of work activity and nightlife can be derived and visualized in novel ways, for instance as a tool for urban planning.
The next two articles are likewise concerned with large-scale data collection in urban environments, but focus on the problem of estimating traffic routes. In "PFlow: Reconstructing People Flow Recycling Large-Scale Social Survey Data," Yoshihide Sekimoto and his colleagues present a study based on travel survey data collected in Tokyo and Hanoi, and show that they can reconstruct the flow of people from location disclosures that are fragmented in time and space. Although not concerned with sensing in the narrower sense, their work demonstrates opportunistic use of data collected from individuals to derive mobility patterns.
The work described in "Estimating Origin-Destination Flows Using Mobile Phone Location Data," by Francesco Calabrese and his colleagues, is closely related, but uses mobile phone data to estimate the flow of people in terms of origins and destinations of individual trips. The authors show that their method correlates well with census survey data while also capturing temporal variations—for instance weekday and weekend patterns.
The final article in this collection, "Exploiting Social Networks for Large-Scale Human Behavior Modeling," by Lane and his colleagues, considers opportunistic use of social network data for sensor-based modeling of human behavior. The idea underlying this work is the combination of data from individuals with close social ties, as identified by friendship or colocation. By leveraging a community of users in this way, behavior can be learned from larger data sets while reducing the burden on individuals for labeling training data.