This paper presents a nonlinear error model based on the quaternion for the rapid transfer alignment of the inertial navigation system (INS). It allows the large initial misalignment uncertainty. Then, the unscented Kalman filter (UKF) is designed to achieve the nonlinear filtering based on the proposed model, and utilizes the difference in velocity and attitude between the slave and master INS as the measurement variables. This paper analyzes and compares the misalignment estimation error and convergence rate of the proposed algorithm with the rapid alignment prototype (RAP) and the velocity-only matching algorithm. The results of simulation suggest that the proposed algorithm could achieve the same alignment performance, not limiting the initial attitude error, as the rapid alignment prototype to do when the misalignment is small. The convergence rate of the azimuth misalignment using the proposed algorithm is rapider than using the velocity matching algorithm for large heading uncertainty