It is required to improve reliability and economical efficiency of nuclear power plant (NPP) at the same time. There is a room to achieve this goal by introducing state-of-art information technology into maintenance work of NPP. In this study, a maintenance support system using Augmented Reality (AR) has been proposed to support water system isolation task that is one of the important works during a periodic maintenance of NPP. In order to apply the AR to the water system isolation task, a tracking method that measures position and orientation of workers in real time is indispensable. The existing tracking methods applied in some AR applications, however, does not fulfill the requirements from the viewpoint of accuracy, stability and availability when using in NPP. In this study, therefore, a new tracking method that combines plural tracking methods has been developed. Concretely, the existing tracking methods employing vision sensor and inertial sensors were improved and combined into one tracking method. Concerning to the method employing vision sensor, a tracking method which recognizes artificial markers and natural features at the same time has been developed to decrease the number of markers which need to be placed in working environment in advance. Concerning to the method employing inertial sensors, a new drift cancel method employing two same sensors placed symmetrically has been developed. These improved tracking methods were combined into one tracking method and evaluated by some experiments.