Height irregularity is a significant type of track irregularity that adversely affects the smooth operation of trains. Currently, the track irregularity detection primarily relies on periodic inspection by track inspection vehicles, which cannot continuously assess track conditions. To address this, this paper proposed an intelligent estimation algorithm of track height irregularity based on vehicle dynamic responses. Through an analysis of the effect of height irregularity on vehicle dynamics, a CNN-LSTM-based track irregularity estimation model is developed. Using vertical wheelset acceleration as the model input, the algorithm rapidly estimates track height irregularities. The effectiveness and robustness of this algorithm are comprehensively discussed through various experimental cases, which shows that the algorithm achieves accurate estimations with a MAE of 1.1700.