Abstract: The effective sensing of non-line-of-sight occluded targets by multipath exploitation radar primarily relies on the reflection of electromagnetic waves from building surfaces and corner diffraction. This capability provides critical information support for diverse applications such as intelligent driving, disaster rescue and urban warfare. However, existing NLOS target localization methods typically perform localization based on imaging results from direct multipath fusion, but the overlapping of echoes from different propagation paths easily generates numerous ghost targets, severely interfering with the identification of true targets. To address this issue, this paper proposes a single-channel ultra-wideband radar NLOS target localization method based on multipath geometric hypotheses and association matching. First, a multipath propagation model is constructed for a typical L-shaped building corner scenario, and the effective sensing region is defined by incorporating radar beam directivity. On this basis, a trajectory hypothesis verification strategy is developed, in which candidate trajectories within the effective region are assumed to correspond to diffraction-to-right-wall reflection paths. Distance constraints are established using elliptical geometric properties to generate candidate target sets, which are then screened by introducing a time-delay matching error factor to determine the true target. Both simulation and experimental results demonstrate that the proposed method effectively suppresses ghost interference and significantly improves the reliability of NLOS occluded target localization.