Device-Free Indoor Localization Based on Kernel Sparse Representation

Device-free localization (DFL) is now a much-needed direction in localization technology. When applied in practice, a large amount of localization data is received from radio sensors. In this study, we consider localization as a classification problem using sparse representation, which speeds up localization and makes real-time localization possible. Because of the linear indistinguishability of localization data, different classes of localization data vectors may be distributed in the same direction. Thus the common sparse representation classification problem cannot cope with this situation well. To further improve localization accuracy, we introduce the kernel sparse representation (KSR) method, i.e., the kernel function is combined with the sparse representation. First, the data is kernel transformed and projected to a reduced dimension, making the computational less difficult. Then, in order to obtain the sparse vectors, we apply orthogonal matching pursuit (OMP). According to the experiment results, KOMP, which has good noise immunity compared to OMP and other sparse coding algorithms, can achieve 100
^%
localization accuracy in a 0.5m×0.5m grid localization scenario.