Home Projects Publications People Events Sponsors

Our projects


Top end mobile phones include a number of specialized (e.g., accelerometer, compass, GPS) and general purpose sensors (e.g., microphone, camera) that enable new people-centric sensing applications. Perhaps the most ubiquitous and unexploited sensor on mobile phones is the microphone -- a powerful sensor that is capable of making sophisticated inferences about human activity, location, and social events from sound. SoundSense, is a scalable framework for modeling sound events on mobile phones. SoundSense is implemented on the Apple iPhone and represents the first general purpose sound sensing system specifically designed to work on resource limited phones. The architecture and algorithms are designed for scalability and SoundSense uses a combination of supervised and unsupervised learning techniques to classify both general sound types (e.g., music, voice) and discover novel sound events specific to individual users. The system runs solely on the mobile phone with no back-end interactions.

Read more about the project in our paper published in MobiSys '09. [manuscript]


CenceMe is a personal sensing system that enables members of social networks to share their sensing presence with their buddies in a secure manner. Sensing presence captures a user’s status in terms of his activity (e.g., sitting, walking, meeting friends), disposition (e.g., happy, sad, doing OK), habits (e.g., at the gym, coffee shop today, at work) and surroundings (e.g., noisy, hot, bright, high ozone). CenceMe injects sensing presence into popular social networking applications such as Facebook, MySpace, and IM (Skype, Pidgin) allowing for new levels of "connection" and implicit communication (albeit non-verbal) between friends in social networks. The CenceMe system is implemented, in part, as a thin-client on a number of standard and sensor-enabled cell phones and offers a number of services, which can be activated on a per-buddy basis to expose different degrees of a user’s sensing presence; these services include, life patterns, my presence, friend feeds, social interaction, significant places, buddy search, buddy beacon, and "above average?"

Read more about the project in our white paper published in EUROSSC '07 and system evaluation pubished in Sensys '08. [EUROSSC '07 manuscript] [SenSys '08 manuscript]

Tracking Mobile Events with Mobile Sensors

We are interested in applying the people-centric sensing concept to the problem of detecting and tracking mobile events. For instance the mobile phones found in a crowd of people can be used to track a lost child or a teenager's disruptive car stereo. There are a number of challenges in building a mobile event tracking system using people-based mobile sensors such as those found on mobile phones. First, mobile sensors need to be tasked before sensing can begin and only those mobile sensors near the target event should be tasked for the system to scale effectively. Another issue that complicates the design of the system is that the mobility of people is uncontrolled. Finally, there is no guarantee that there will be sufficient density of mobile sensors around any given event of interest. This results in time-varying sensor coverage and disruptive tracking of events, i.e., targets will be lost and need to be efficiently recovered.

We have built MetroTrack, a fully distributed tracking system based on off-the-shelf mobile phones capable of tracking mobile targets through collaboration among local sensing devices that track and predict the future location of a target using an efficient distributed Kalman-Consensus filtering algorithm. MetroTrack is implemented on Nokia N80 and N95 phones and as a proof-of-conceptwe show that MetroTrack is effective at tracking a mobile noise source in an outdoor urban environment. Simulation results indicate that MetroTrack is robust in the presence of different mobility models and mobile density.

More information on this project online soon.

Leveraging Opportunism to Increase Sensing Fidelity

In a people-centric opportunistic sensor network (OSN), mobile sensors may receive queries both on behalf of applications running locally on the mobile device and applications running remotely on back end infrastructure (e.g., Internet). For instance, a sensor-enabled mobile phone may be requested to gather data about particulate matter in the air for a certain area of the city. These queries minimally specify a sensor type to be sampled and the rate and duration of the sampling, and may also include conditions under which it is preferable that the sampling take place. These sampling conditions, i.e., the sampling context, may include things like location, position on the body/clothing, and the activity of the person carrying the sensor. Given the dynamism of a mobile sensor’s context in the people-centric paradigm, it may often be the case that the mobile sensor can not itself meet the required sampling context, and in fact, due to device heterogeneity, may not even possess the requested sensor type on-board. This means either the query is dropped/rejected, or the fidelity of the captured data (with respect to what the application wants) is reduced. Further, a given query may explicitly require inputs from more than one sensor. To address these issues in mobile people-centric sensor networks we propose Quintet, a method for orchestrating neighborhood sensors to increase sensing fidelity through opportunistic in-situ sensor sharing. With Quintet, an underqualified node, in terms of sensor type or sampling context, can temporarily borrow the sensor of a qualified node in an effort to satisfy an application query. We describe a sensor sharing protocol focused on local interactions between mobile sensors in the field. The protocol uses lightweight solicitation/reply messaging exchanges between a primary queried node and potential sharing nodes it encounters as it moves in the field to feed a distributed context and resource aware matching algorithm that makes sensor sharing decisions. So far, we have implemented and evaluated our techniques in a human pedestrian sensor network testbed using both mote-class devices and mobile phones. We demonstrate that sensor sharing provides a viable method to improve sensing fidelity and delay in answering application queries, and also facilitates a new class of applications that require collaborative sensing.

More information on this project online soon.

Second Life Sensor

Virtual world simulators like Second Life represent the latest and most successful frontier of online services for entertainment and business. People have virtual lives in these worlds using personal avatars. Bridging real life and these virtual worlds together is challenging, but enables new application scenarios for these systems. Existing work focuses on the representation of objects inside virtual worlds. We believe that the overall experience will be improved if we can reproduce not only inanimate and passive objects like buildings and beer steins but also the subjects in these virtual environments.

In order to have a seamless integration between the physical and virtual worlds, and to maximize ease of integration and scale of adoption, the system should rely on devices that people already use every day. We argue that the sensors embedded in commercial mobile phones can be used to infer real-world activities, that in turn can be reproduced in virtual settings. We are investigating the challenges related to the implementation of such a system both from algorithmic and a systematic points of view. We have started by implementing Second Life with CenceMe, a platform to infer activity and more general context information from sensor-equipped phones.

Read more about the project in our short paper published in HotEmNets'08. [manuscript]


Opportunistic people-centric mobile-sensing model introduces a new security challenge in the design of mobile systems: protecting the privacy of participants while allowing their devices to reliably contribute high-quality data to these large-scale applications. AnonySense is a privacy-aware architecture for realizing pervasive applications based on collaborative, opportunistic sensing by personal mobile devices. AnonySense allows applications to submit sensing tasks that will be distributed across anonymous participating mobile devices, later receiving verified, yet anonymized, sensor data reports back from the field, thus providing the first secure implementation of this participatory sensing model. Our papers describe the underlying threat model and trust model, and show how AnonySense provides the desired security properties. We evaluated our prototype implementation through experiments that indicate the feasibility of this approach, and through demonstration applications. Our experiments show that our implementation is efficient, and how our location-blurring feature can provide statistical k-anonymity.

Read more about the project here.

Measuring How Our Bodies Impact Wireless Communication

Future mobile sensing systems are being designed using 802.15.4 low-power short-range radios for a diverse set of devices from embedded mobile motes to sensor-enabled cellphones in support of, for example, people-centric sensing applications. However, there is little known about the use of 802.15.4 in mobile sensor settings nor its impact on the performance of future communication architectures. We present a set of initial results from a simple yet systematic set of benchmark experiments that offer a number of important insights into the radio characteristics of mobile 802.15.4 person-to-person communication. Our results show that the body factor - that is to say, the human body and where sensors are located on the body (e.g., on the chest, foot, in the pocket) - has a significant effect on the performance of the communications system. While this phenomenon has been discussed in the context of other radios (e.g., cellular, WiFi, UWB) its impact on 802.15.4 based mobile sensor networks is not understood. Other findings that also serve to limit the communication performance include the effective contact times between mobile nodes, and, what we term the zero bandwidth crossing, which is a product of mobility and the body factor. which is a product of mobility and the body factor.

Read more in our project report published in EWSN '08. [manuscript]

BikeNet: A Mobile Sensing System for Cyclists

There is substantial interest in the cycling community in collecting data quantifying various aspects of the cycling experience. Existing commercial products have begun to integrate data from multiple local sensors (including biometric sensors and a GPS receiver) on a single user display, and even provide map software. A limitation of the currently available products is the inability to share data with other riders in real-time. Further, often real-time performance analysis of locally collected data is limited to local display of simple statistics like min, max and mean over the entire trip. When the road terrain is highly non-uniform and uphills can last a long time, comparing current speed against a trip-wide average loses significance.

With the BikeNet application, we work to address these limitations by allowing cyclists to share information about themselves and the paths they mutually traverse for real-time display. In BikeNet, information sharing occurs via short range radios, and can be direct (i.e., bike-to-bike) or indirect through neutral third-party entities called Checkpoints. Checkpoints are virtual storage and aggregation devices that are placed along roads and trails frequented by cyclists; they store location-specific performance data on per-cyclist and aggregate bases.

Read more in our project report published in SENSYS '07. [manuscript]

SkiScape: Sensing the Slopes

Skiers are interested in knowing current trail conditions (e.g., ice, bare spots, congestion) when at the base of the mountain in order to determine which lift to use to get to the desired trail head at the top of the mountain. Resort managers are interested in learning skier flow statistics to estimate wear on lift to use to get to the desired trail head at the top of the mountain. Resort managers are interested in learning skier flow statistics to estimate wear on the terrain in order to enact preventative maintenance (e.g., close trail, make artificial snow). Safety/emergency personnel are interested in tracking skiers' location and speed in case of accidents (e.g., fall off trail, avalanche), and also to prevent accidents by speed policing. Skiers may be interested in tracking their own location or the location of their friends on the mountain, as well as avoiding lifts with long queues. We are inspired by the resemblance of a ski resort trail map to a static sensor network data dissemination tree; many trails with heads at the top of the mountain funnel towards a small number of lift entry/collection points at the base of the mountain. In the SkiScape, ski lifts provide a continuous supply of data mules (skiers) to the trails at no cost to the sensing/communication infrastructure. Static sensors, mounted on light poles, sense data about the adjacent trail area; mobile sensors, mounted on skiers, can collect data in their locality as the skier traverses the mountain. Skiers opportunistically collect/carry data of interest as they travel along the trails to the data sinks at the base. In this way we leverage a sparse deployment of both static and mobile sensors to give a more complete picture over time of the field of interest at lower cost than would be required with a fully static deployment.

Read more about the idea in the poster abstract from SENSYS '06. [manuscript] [poster]

Dartmouth Campus Area Sensor Network

We have started to deploy Sensor Access Points (SAPs) - Aruba AP-70 WiFi APs and with one or more Tmote Invents - across Dartmouth Campus. Currently we have complete coverage of the Computer Science building and plan on covering Thayer School of Engineering next and then experiment with mobile sensors and their interaction with these SAPs to develop out MetroSense. Collectively this infrastructure presents a general purpose, open programming environment for people-centric applications. The idea is to "parasitically" use existing APs to build out our campus area SAP infrastructure. SAPs seamlessly support standard WiFi access, ground-truth sensing (since SAPs have sensors), and sensor comms (802.15.4) and run the Open-WRT Linux distribution on the AP and TinyOS on the mote.

Read our architecture white paper published in WICON '06. [manuscript]

last modified: June 2009
© Computer Science Department at Dartmouth College
The project research is sponsored by: