Summary

Radio Tomographic Imaging (RTI) is an emerging technology that locates moving objects in areas surrounded by simple and inexpensive radios. RTI is useful in emergencies, rescue operations, and security breaches, since the objects being tracked need not carry an electronic device. Tracking humans moving through a building, for example, could help firefighters save lives by locating victims quickly.

RTI works by placing many small and inexpensive radios around an area of interest. Each radio is capable of transmitting and receiving wireless signals, creating a dense network of "links" that pass through the area. Objects that move within the area reflect and/or absorb the wireless signal, preventing some of the power from reaching its destination. An image of where the power is being absorbed can be formed using all the link power loss measurements, thus allowing one to know where objects within the area are located. This research uses theory found in imaging, estimation/detection theory, inverse-problems, regularization, signal processing, communications, electromagnetics, and wireless networking.

Intuitive Explanation

Radio tomographic imaging (or any tomographic imaging computation, for that matter) is similar to the numbers game, Magical Squares. In fact, we have developed an activity to teach tomography to fourth-grade elementary school students. In magical squares, you are given the column sums and row sums of a grid of numbers. Your job is to determine which numbers fill the spaces in the grid in order to achieve the given row and column sums.

In radio tomographic imaging, we also have these row and column sums. The change in RSS on the horizontal links crossing the medium are, effectively, row sums of the attenuation caused by each pixel in that "row". Similarly, the change in RSS on the vertical links crossing the medium are column sums of the attenuation caused by each pixel in that "column". There are two main differences between the magical squares and RTI. First, we have the benefit of many more measurements, across many different diagonal lines, since we measure the sum between any pair of sensors. Second, we have the problem that the sums we measure are not exact; there is random "noise" added in, that makes our measurements not exactly equal to the sum of the attenuations in each pixel. In fact, sometimes they are far off! So our assignment of numbers for each pixel will only give us sums that are "close" to the measured information.

Videos

RTI demo

The following video is an RTI experiment conducted by Joey Wilson and Neal Patwari at the Warnock Engineering Building at the University of Utah. The attenuation image is shown above the actual footage, with the red spots indicating heaviest attenuation.

Variance-based RTI (VRTI)

In the following video, Joey Wilson and Neal Patwari of the University of Utah describe and demonstrate their through-wall wireless network tracking system. Joey enters a home surrounded by simple radio devices. He does not carry any kind of electronic device on his body, nor are any devices deployed inside the tested part of the home. The system estimates his position using radio tomographic imaging.

Robust estimators for VRTI

In this video, a through-wall experiment is performed on a windy day in late spring. Leaves and branches swaying in wind cause a significant amount of RSS variance. In another experiment, several rotating fans are used to create some motion as noise ( Youtube Link ). However, our robust estimators: subspace variance-based radio tomography (SubVRT) and least squares variance-based radio tomography (LSVRT) can still track a person with sub meter accuracy.

Applications

Some applications of this technology include:

• Track people and vehicles in next-generation security systems
• Locate victims in fire and rescue operations
• Locate and count people in the rooms of "smart homes"
• Measure accurate usage statistics and trending in buildings and stores
• Automate control of other devices (HVAC, cameras, spotlights, etc)

Ongoing Research

Current research in this area is being performed by Joey Wilson, Yang Zhao and Neal Patwari.

Acknowledgements

This material is based upon work supported by the National Science Foundation under the Early Career Faculty Development (CAREER) Grant No. ECCS-0748206. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the researchers in the SPAN lab and do not necessarily reflect the views of the National Science Foundation.