Summary: Multimodal Sensor Networks
We investigate the benefits that multiple modalities bring to the problems of target and
self localization in sensor networks. Specifically,our work focuses on coupling motes
equipped with magnetometers with cameras.
We first looked at how feedback from cameras can be used to calibrate the
magnetometers . In our model the magnetic field signal strength follows
a power law with unknown exponent beta. The goal of the calibration process is to estimate the
value of beta. The process works as follows: Motes use trilateration to estimate the position of
the target. This information is then passed to the cameras which locate the actual position of the
target. We then solve a non-linear optimization problem to minimize the distance between the
target's actual location and the location estimated using the magnetometer measurements. Our
simulation results show that the estimation process converges in a small number of iterations.
Next, we designed an integrated system for target tracking using cameras and
magnetometers . The proposed system initially uses two pre-calibrated, PTZ cameras
to localize the network's motes. To do so, each mote is equipped with a LED and the cameras
scan the scene until they both have the LED at their center of view. Then, the camera angles
are used to estimate the location of the mote through a generalized triangulation formulation.
Note that we do not require that the locations of the cameras are known. Once the mote