We present a robust localization algorithm for multiple radiation sources using a network of sensors under random underlying physical processes and measurement errors. The proposed solution uses a hybrid formulation of particle filter and mean-shift techniques to achieve several important features that address major challenges faced by existing localization algorithms. First, our algorithm is able to maintain a constant number of estimation (source) parameters even as the number of radiation sources K increases. In existing algorithms, the number of estimation parameters is proportional to K and thus the algorithm complexity grows exponentially with K. Second, to decide the number of sources K, existing algorithms either require the information to be known in advance or rely on expensive statistical estimations that do not scale well with K. Instead, our algorithm efficiently learns the number of sources from the estimated source parameters. Third, when obstacles are present, our algorithm can exploit the obstacles to achieve better isolation between the source signatures, thereby increasing the localization accuracy in complex deployment environments. In contrast, incompletely specified obstacles will significantly degrade the accuracy of existing algorithms due to their unpredictable effects on the source signatures. We present extensive simulation results to demonstrate that our algorithm has robust performance in complex deployment environments, and its efficiency is scalable to many radiation sources in these environments.
展开▼
