Aiming at the in-motion initial alignment problem of the strapdown inertial navigation system when the attitude changes greatly during the erection process, a non-Gaussian Lie group Bayesian filter based on maximum correntropy is developed in this paper. Firstly, this method overcomes the limitation of the first-order linear approximation in the Kalman method by establishing the initial attitude probability distribution model in the Lie group space and shows superior performance when encountering rapid attitude changes and large angle errors. Secondly, by extending the definition of correntropy to the Lie group space, this method derives the update equation based on Bayesian principles to process the state-dependent non-Gaussian noise in-motion carrier alignment model and enhance robustness against sensor outliers. Results from the turntable experiment indicate that the proposed method outperforms the existing approaches in alignment time and accuracy, demonstrating excellent performance in the dynamic alignment erection process of SINS.