Localization involves the tracking of objects, and there are currently a number of different ways to accomplish this. Currently, much of the technology used to track the objects requires the person or object to be holding some sort of tag that can interact with the network being utilized. Device-free localization is the process of locating people/objects without the need for the person or object to be carrying or wearing any sort of tag.
One promising method of device free localization involves using the Received Signal Strength (RSS) measured by devices that communicate over radio signal. With the price of small radio frequency emitting/receiving devices being much lower than other transmitting counterparts, the opportunity of using these devices to create a network capable of leveraging the coordinates of users through the use of RSS has become increasingly possible. By using RSS among enough devices, we can generate a general location for the object in question.
Some uses for device-free localization include tracking people that may be stuck inside buildings that are on fire. Being able to see where they are located can help firefighters choose what is the best plan of action and get to victims faster than having to search the entire building blindly. Another possible use for this technology could be for security purposes. Using this technology to keep track of movement in a building and keep track of where they may have been throughout the location tracked.
Device-free localization has become a process that has stemmed work across the country and world in order to find more efficient and accurate algorithms to use this information and transform it into a practical application. In response to this interest in device-free localization, competitions are held in the style of a hackathon to have students come and create test their own algorithms and methods against other groups. One such competition is the Bosch Device-Free Localization Competition held during Cyber-Physical Systems and Internet-of-Things Week.
The current design of the algorithm in use by Dr. Patwari operates as follows:
In fig. 1 5 radio transmitters were used as an example. The transmitters can only be in a receiving or transmitting state at any given moment, but not both at the same time. They can store the RSS received by each device, so given 5 nodes, they would store 5 RSS values. Since the transmitters cannot receive their own transmitted signal, a default value of 127 (the value used to denote ‘no measurement’ in the system) is kept where that transmitter number is expected.
In addition to the transmitters, there is a ‘listener’ node that is a Beaglebone Black (BBB), an open source radio receiver, which can pick up the transmitted signal and transfer the values to the computer, allowing for us to process the signal and then output in the form of a file. The system operates by utilizing 8 channels of transmission to counteract random destructive interference from variable environment conditions.
There are many different device-free localization methods that could be used to infer a person’s position from our RSS values. The first are fingerprint-based methods. Fingerprint-based methods compares live measurements to database of training measurements. It can use a histogram of measurements for each link to obtain the location of maximum likelihood. However, fingerprint based methods are exponentially more difficult to train for multiple people and training data can degrade over time as the environment changes. The histogram method also requires many measurements of the person standing in one location to obtain a location. Due to its requirement for extensive training data for each environment,, we are not planning on using this method.
Imaging is another method for device-free localization. Imaging creates images of changes to the environment, which can be used to estimate a person’s location. When using RF sensors, this method is referred to as radio-tomographic imaging (RTI). The image is divided into P voxels in a vector x = [x(1) … x(P)]^T. Each link between nodes will have an RSS measurement in y = [y(1) … y(L)]^T that is the difference from its value with an empty environment. We can create a linear model y=Ax+n, where n is additive noise. Using matrix inverse math, we can create an estimate for x at each voxel. RTI can also be based on the variance of each measurement link to view movement. Imaging methods can be very noisy, so smoothing is required to produce a clean image.
Another method for device-free localization is RSS tracking. By using the changes in each RSS link, a cross-point can be estimated. This point can be used to estimate position and velocity of the person. This velocity can be used to predict future position values. A Kalman filter could be used to track the location of highest likelihood over time.
There are many other methods for device-free localization algorithms, but these three are the most viable to complete within our time-constraints. Out of these three methods, we are planning on using imaging. This is because we thought of several methods to improve the current imaging algorithms, and because creating a live image should make presenting on our results easier.