A high-precision reconstruction method is presented based on the depth camera to rebuild the model of the aircraft fuel tank. The method achieves the registration by the use of multi-view point clouds， and the registration process includes two stages： Coarse registration and fine registration. We first establish the point pair features （PPFs） on two point clouds as the global descriptor. Then， we deploy a voting method on a set of coarse poses to obtain the one which has the highest score as the initial pose. Based on the initial pose， we construct a joint photometric and geometric objective function， in order to align the colored point clouds using iterative closest point （ICP）. Finally， we apply the above process on the adjacent frame point clouds and optimize the reconstruction results with graph optimization， so as to achieve the reconstruction of the complete 3D model of the aircraft fuel tank. The method achieves high accurate registration than other state-of-the-art approaches. In addition， the method is robust when encountering the repetitive features and structures， in which the traditional ICP may lack stability under this situation.