Vehicular sensing applications, such as the popular StreetView service by Google, offer tremendous opportunities for cost effectively monitoring urban and suburban environments. With participatory sensing, where individuals are able to contribute their own data, this potential grows even further, possibly to a point where StreetView type imagery could be recorded and distributed in near real-time, as opposed to collected once every few years.
However, the aggregate bandwidth requirements of such an application by far surpass even the fastest cellular data services still on the drawing board. This project investigates alternative strategies for dealing with this coming data deluge, including in-network storage, distributed query processing, and vehicle-to-vehicle communication using extremely high-speed local wireless links.
The anticipated results of the project include:
- network and link layer mechanisms for efficient, GPS-assisted, vehicle to vehicle communication and in-network storage,
- experimental demonstration and evaluation of said mechanisms in urban environments, and
- analysis and simulation of the storage and communication capacity of mobile vehicular sensor networks,
- open-source releases of software artifacts produced.
Participatory sensing is the idea of leveraging equipment carried by ordinary people as they move around in their environment, to create large-scale sensor networks capable of observing phenomena that would otherwise be cost-prohibitive or infeasible to observe. The participants are sometimes called volunteers, but depending on the application and sensor modality, the participants do not need to be actively involved in, or sometimes even aware of, the task at hand. In this project, we have studied the use of cellphones as participatory sensors. Modern smartphones come with a GPS receiver, and frequently report their location to third-party services. We have studied how large collections of such reports may be used to automatically generate accurate and up-to-date road maps, as an alternative to today's labor intensive, hand-drawn maps. The state-of-the-art map inference engine created as part of this project may one day be used to create customized maps for different types of travelers and vehicles - such as trucks, bicycles, tourists, strollers or wheel-chairs. Another outcome in along the lines of GPS tracking is our "thrifty tracking" system. We have identified the equation governing the trade-off between accuracy, cost (in terms of cellular plan data usage), and delay for GPS tracking apps, and designed the first tracking system that lets the user specify two out of the following three parameters: (a) data budget, (b) maximum location error, or (c) delay. The system then adjusts the third parameter to meet the user's specified constraints. For example, you may be willing to use only 10 kilobytes of data per day tracking your teen driver, but you might tolerate a 15 minute delay in reporting. Our system produces a tracking schedule that minimizes the location error given your specified constraints. Finally, while several smaller results have come out of this work, the last one worth special mention is our work on tracking unmodified smartphones using their Wi-Fi transmissions. When Wi-Fi is enabled on a smartphone, it periodically transmits information, including a unique identifier. We described a system for producing accurate movement traces from such transmissions, recorded by Wi-Fi receivers near the transmitting phone. In our experiments, we were able to track thousands of unique phones passing by road-side monitors mere hours, illustrating the potential of this technology both for important traffic measurements, and for nefarious privacy intrusions on unsuspecting smartphone users.
