Abstract: The self-localization of autonomous underwater vehicles(AUVs) and target localization are the fundamental technologies for underwater operations, which can provide real-time position estimation of AUVs and targets, respectively. A distributed multi-model localization algorithm is designed to address the common issues of variable motion states during the task execution process. It enhances the adaptability of localization process to variable motion states and improves the collaborative localization accuracy through the interaction calculation and motion matching between different models. A three-dimensional target localization algorithm based on attitude estimation is designed to address the cumulative errors caused by the non-cooperative nature of the targets and delayed position prediction. The real-time geometric relationship between each AUV and the target is obtained through azimuth measurement, and an estimation strategy for target attitude is constructed, which improves the real-time and accuracy during the target localization prediction process. The advantages of the proposed algorithm in model adaptability, localization precision, etc. are verified through simulation experiments.