Abstract
Accurate localization of trapped victims in postdisaster environments is challenging due to the failure of normal operational systems, severe Non-Line-of-Sight (NLOS) propagation, multipath effects, and an unknown number of active devices. This paper proposes an information-guided unmanned aerial vehicle (UAV)-assisted localization framework for multi-victim detection in post-disaster collapsed-building scenarios. The approach is particularly relevant to Vehicle-to-Everything (V2X)-enabled disaster settings, where connected vehicles and emergency units act as distributed signal sources in the absence of fixed infrastructure. A UAV follows a spiral trajectory and selects informative sensing locations using a Reference Signal Received Power (RSRP)-based utility function to improve geometric diversity. Hybrid Time-of-Flight (ToF) and Angle-of-Arrival (AoA) measurements are transformed into pseudopositions for clustering and localization. An Integrated Completed Likelihood (ICL)-based Gaussian Mixture Model (GMM) estimates the unknown number of victims while mitigating over-splitting. Simulation results based on the Third Generation Partnership Project (3GPP) Urban Micro Street Canyon (UMiSC) channel model show up to 22% localization-error reduction and approximately 90% success within a 1.5 m threshold under NLOS conditions.