Human-centric Sensing
By Mani Srivastava†,Tarek Abdelzaher‡,and Boleslaw Szymanski¶洁白的反义词
美颜拍摄Abstract
Thefirst decade of the century witnesd a proliferation of devices with nsing and communication capabilities in the posssion of the average individual.Examples range from camera phones and wireless GPS units to nsor-equipped,networked fitness devices and entertainment platforms(such as Wii).Social networking plat-forms emerged,such as Twitter,that allow sharing information in real time.The unprecedented deployment scale of such nsors and connectivity options usher in an era of novel data-driven applications that rely on inputs collected by networks of humans or measured by nsors acting on their behalf.The applications will impact domains as diver as health,transportation,energy,disaster recovery,in-telligence,and irregular warfare.This paper surveys the important opportunities in human-centric nsing,identifies challenges brought about by such opportunities, and describes emerging solutions to the challenges.
1.Introduction
Our work is motivated by the recent surge in nsing applications characterized by distributed collection of data by either lf-lected or recruited participants for the purpo of sharing local conditions,increasing global awareness of issues of interest, computing community statistics,or mapping physical and social phenomena.This type of applications has recently been called participatory,opportunistic,or human-centric nsing[1].Examples of early applications include CarTel[2],BikeNet[3], MMM2[4],and ImageScape[5],among others.
A confluence of technology trends has precipitated the advent of such nsing applications,where the focus of nsing process is more personal or social.The first t of technology trends has to do with the proliferation of a wide variety of nsors in the posssion of the average individual.The cond t of trends lies in the proliferation of options for ubiquitous and real-time data sharing,as exempli-fied in the ubiquity of smart phones with network connectivity and the increasing popularity of social networking ,Twitter)for information publishing.
On the nsor front,active RFIDs,smart residential power meters(with a wire-less interface),camera cell-phones,in-vehicle GPS devices,accelerometer-enhanced entertainment ,Wii-fit),and activity monitoring , the Nike+iPod system)have all reached mature market penetration,offering un-precedented opportunities for data collection.Major industry initiatives,such a
s †Department of Electrical Engineering,University of California,Los Angeles,Los Angeles, CA90095,USA
‡Department of Computer Science,University of Illinois at Urbana-Champaign,Urbana,IL 61801,USA
¶Department of Computer Science,Renslaer Polytechnic Institute,Troy,NY12180,USA Article submitted to Royal Society T E X Paperwps删除空白行
2Mani Srivastava†,Tarek Abdelzaher‡,and Boleslaw Szymanski¶HealthVault,automate collection of and access to information.A significant num-ber of vendors announced wearable health and biometric monitoring nsors since 2008that automatically upload ur data to HealthVault.
On the social and networking fronts,the ubiquitous proliferation of cell-phones and social network cites offers ample opportunities for real-time data sharing.Mod-ern cell phones are equipped with a non-trivial collection of nsors,in addition to Bluetooth,WiFi,4G,and near-field communication options that turn the device into a portal for connecting physical world instrumentation to the Internet.Ve-hicular Internet access,offered in some new car ,Chrysler’s Uconnect Web,and BMW’s ConnectDrive)enables new applications that exploit network connectivity to export nsory information on the move.For example,rvices such as OnStar have long since exported on-
board diagnostics(OBD-II)measurements to offer remote access to a large number of vehicle nsors and gauges.
The availability of nsing devices,Internet connectivity options,and social fo-rums for information sharing open up an important new category of distributed applications in energy,health,environmental,and military domains that rely on individual and community nsing.New rearch challenges emerge from the in-volvement of human populations in a nsory data collection and decision-making loop.They include incentives,recruitment,privacy,data accuracy,system modeling, and interpretation of social nsing dynamics.This paper categorizes human-centric nsing applications,surveys the aforementioned challenges,and discuss emerging solutions.
2.Dimensions of Human-centric Sensing
疫情防控背景>离婚协议怎么写In traditional nsor networks,the emphasis has been on unattended and au-tonomous system operation,with the run-time role of humans limited to being end-consumers of information products(using information products from the n-sor network to make decisions and take actions).By contrast,the distinguishing aspect of human-centric nsing systems is a larger involvem
情侣之间的祝福语
ent of humans along other points in the data-to-decision path.Human involvement is particularly uful in nsing various process in complex personal,social,and urban spaces where traditional embedded nsor networks suffer from gaps in spatiotemporal cover-age,limitations in making complex inferences,inability to adapt to dynamic and cluttered spaces,and aesthetic and ergonomic problems.By taking advantage of people who already live,work,and travel in the spaces,and their adaptability and intelligence,human-centric nsing makes it possible to get information that otherwi is not possible.While human-centric systems are quite diver,one can classify them in terms of the extent and role of human participation in nsing, which falls under one or more of the following categories:
•Information source:Humans are often versatile and unique sources of infor-mation about process and relationships that exists in our spaces.Indeed,in the defen and curity arena there is a long history of information gathered via HUMINT(HUMan INTelligence)as oppod to electronic nsors.More recently social media outlets,such as Twitter,have emerged as sources of real-time information about society.In social,behavioral,and medical sciences, Ecological Momentary Asssments(EMAs)of human subjects are commonly ud for information that is hard to get from physical nsor sources.
Article submitted to Royal Society
Human-centric Sensing3•Measurement collection:A cond role for humans is to participate in operating a nsor to collect raw measurement data.The advent of powerful consumer-grade mobile smartphones equipped with embedded or wirelessly connected nsors has suddenly enabled billions of individuals to collect geo-tagged nsor measurements about their immediate spaces,such as an image or a sound clip or a temperature reading.Not only does such nsing naturally provide nsor coverage where interesting process are happening,but also the human experti in intelligently operating the nsor is uful in gath-ering higher quality ,capturing high quality images in a cluttered space with poor lighting).
•Sensor data processing:For raw nsor measurements to be uful,infor-mation features must be extracted from them and annotated with metadata describing the context.The tasks are often difficult to do algorithmically and autonomously,particularly for complex phenomena and rich nsor data types,and human participation can simplify them considerably.For example, having an individual who captured an image with a smartphone camera to also classify and tag the image with information identifying the scene can aid subquent analysis of that image.Crowd-sourcing platforms such as Amazon Turk can also be leveraged to engage a large number of human participa
nts in processing raw nsor data[6].
•Information sharing:Yet another form of human participation is in proac-tively distributing and sharing the nsor data time ries and derived infor-mation with interested end urs.Unlike traditional nsor networks where data is disminated automatically,human participation during sharing gives the data owner more control over the information being shared(for example, to remove personally nsitive information).Internet connectivity that com-mon on smartphones together with s social networks and media rvices such as Facebook,YouTube,and Twitter provide the tools for such dismination.
Even more powerful is the advent of cloud-hosted data repository rvices capable of managing nsor data time ries,such as Pachube,Google Pow-erMeter,and Microsoft HealthVault.
•Information fusion and analysis:Lastly,humans often play a crucial role in analyzing nsory information,fusing it from diver sources,and extract-ing actionable inferences.In some cas,nsor-sourced information has im-perfections,such as information gaps,errors,uncertainty,and bias,making automated inference hard.In other cas,such as natural language and im-ages,the nsory information may be too complex for current machine learn-ing alg
orithms.Human analysts can alleviate the shortcomings by bringing knowledge about the social,political,economic,and cultural context in which the nsory information was obtained.
瓦斯爆炸的条件In addition to where,in the data-to-decision pipeline,the human participation occurs,another dimension of human-centric nsing is the nature and purpo of human participation.The nature of participation can span a range of possibilities that include voluntary,opportunistic,incentivized,directed,and organized,while the purpo may include collecting nsory information for lf-analysis,a top-down directed nsing campaign for a director’s purpo,or bottom-up data collection that emerges naturally from the participants’cau.In the ctions below,wefirst explore single ur applications,where individuals collect data for lf-analysis.We Article submitted to Royal Society
4Mani Srivastava†,Tarek Abdelzaher‡,and Boleslaw Szymanski¶
then describe coordinated community nsing,where a community of lf-interested or incentivized parties explicitly join a nsing compaign or an otherwi coordi-nated effort to collect information.Finally,we cover human-centric nsing that emerges naturally without central coordination,for example,when communities propagate information for a shared cau in social spaces such as Twitter.
3.Single-ur Applications and Challenges
One important form of human-centric nsing is where an individual collects nsory information for their own u.Owing their inspiration to“lifelogging”applications (capturing and archiving memories of ones life in the form of a continuous time-ries of data[7])the nsing systems provide individuals with information about their activity,health,and lifestyle,and enable them to introspect about the choices they made,analyze their conquences,and take actions.A good example of such an application is the PEIR system[8]that enables individuals with mobile phones to learn the impact of their transportation choices on the environment due to ve-hicular emissions,as well as the exposure they get to environmental pollution.In addition to letting individuals introspect about their data,the PEIR system lets them lectively share it with others,and compare it against group statistics.
While the nsors providing data for such applications may be embedded in the spaces we live in,more common is to u nsors that are always carried on one’s person,either built into one’s mobile phone or in parate wearable nsors.For example,measurement traces from accelerometers,gyros,and GPS embedded in the mobile phones can be ud to obtain a geo-stamped time ries of ones activity and transportation state(such walking,running,sitting,sleeping,biking,and driving), and make inferences such as computing one’s p
hysical energy expenditure.Separate nsors,embedded in personal and social spaces,are often necessary for a variety of reasons.Proper placement of nsor on the body may not be possible in some cas with a mobile phone that is carried in the pocket or held in the hand.Sensing modalities such as an ECG and SPO2nsors are not typically embedded in mobile phones.Size and battery life optimization considerations might further dictate the choice and location of nsors.Finally,other items of frequent personal u,such an individual’s car,may be instrumeted.The nsors may be wirelessly connected (with Bluetooth and ANT radios being the common technologies)to a mobile phone for real-time retrieval of nsory information over the nsor area network,or may log the data in a local memory for later retrieval when connected to a personal computer.At the back end,the applications typically u software running on the ur’s mobile phone or personal computer,or increasingly more commonly as a cloud rvice,for archiving,visualization,analysis,and sharing of the nsory information with social contacts.While human-centric nsing,as a tool for capturing and reflecting on one’s life,is becoming increasingly commonplace,veral technical challenges prent hurdles to wide adoption,some of which we discuss here.
中国古代十大美男(a)The Energy Challenge
The applications described above ek to u the smartphone either as a ns-ing device,or as a co
mmunication gateway for wearable wireless nsors.However, modern smartphones are designed primarily as devices for sporadic u of personal Article submitted to Royal Society
Human-centric Sensing5 communication,mobile applications,and web rvices,and not for continual ns-ing.Sensors such as light nsors,accelerometers,gyros,and magnetic compass were incorporated primarily for the purpos of offering richer ur interfaces,such as display adaptation to screen orientation and lighting condition,and gesture bad control.As human-centric nsing applications have begun to u the nsors to make continual measurements and inferences about the ur’s context,the limita-tions of the platform in terms of battery life become evident.Even emingly simple modalities such as the accelerometer turn out to be quite energy constrained be-cau of the high sampling rates needed for inferring physical context and the lack of architectural support in the I/O subsystems for handling nsor data streams. Complex modalities such as the GPS and imager are even more energy hungry,as are the wireless radios needed for the phone to communicate with wearable nsors. The end result is that smart phones which may last for a day or two when ud in their intended role as personal communication and computing devices,barely last for3-4hours when ud for continual nsing.
In the short term,the key to at least partially meeting this energy challenge lies in smarter lection,a
ctivation and duty cycling of the nsors,making u of information such as the current contextual state,model of expected behavior,and external constraints.While measurements from multiple nsors may contribute to inferring a contextual variable of interest,the nsitivity of the inference to the nsors may vary over time,and can be exploited algorithmically to lectively shutdown or lower the sampling rate of nsors.An example is the SensLoc system described in[9]that actively controls a GPS receiver,a WiFi scanner,and an accelerometer,and fus their measurements to detect commonly visited places and commonly traverd paths.Additionally,prior knowledge of a road map or building layout may be ud to constrain possible evolution of future location,and thus further limit the nsor samples needed.
In the long run,however,the smartphone platform architecture may need to evolve to support more energy efficient sampling of nsors.For example,dedi-cated hardware that can deposit nsor data samples to the main memory and perform simple processing on the data without waking up the main processor can significantly reduce the energy overhead.Wearable nsors,ud external to the smartphones,suffer from their own energy challenge,primarily due to their small size and weight that verely limits the battery size.This is particularly true for nsors designed for high rate nsing modalities,such as an ECG signal,where there is little opportunity to duty cycle naively.Instead,comp
ute intensive local processing that would predict the occurrence of an event of interest,and specialized circuits that would detect their start at an early stage,may be ud to activate and shutdown the nsors smartly.Additionally,a major source of power consumption in wearable nsors for physiological signals is the analog frontend that is ud to amplify andfilter the tiny signals,and,wor,the circuits are hard to duty cycle becau of long time constants associated with thefilters.More optimized analog-to-digital pathways together with the u of emerging compressive and event-driven sampling mechanisms,instead of the more typical Nyquist sampling,would be cru-cial to meeting the energy challenge.
Article submitted to Royal Society
