Distributed High Accuracy Peer-to-Peer Localization in Mobile Non Line of Sight Environments
Kannan Ramchandran and Venkatesan N E
National Science Foundation
Location based services are gaining popularity with the availability of position information predominantly through GPS and other modalities. However, GPS does not work well in closed spaces such as indoors and urban canyon environments. Further standard GPS receivers have errors of over five to fifty meters that is unacceptable for applications such as vehicle safety, autonomous robotic systems , Unmanned Air Vehicle (UAV) systems etc, that mandate sub-meter accuracies. In a multipath-rich environment, the received signals are no longer gaussian in nature challenging the use of standard estimation techniques like the well-known Kalman Filtering framework and its extensions. The goal of this project is to develop an inexpensive localization system by fusing multiple modalities and exploiting cooperation between a large number of imprecise cheap nodes to achieve highly precise location estimates. We exploit the theory of graphical models and pose the problem of localization as a problem of inference over a large graph and design computationally efficient message passing algorithms for inference. We also use tools from optimization theory to provide approximate solutions to the problem of non line of sight localization with guarantees under which exact localization can be achieved.
Figure 1: Vehicular Network
Figure 2: Coupled Hidden Markov Model Representation for collaborative localization