The accurate indoor positioning is of significant social and commercial value. Pedestrian dead reckoning (PDR) has excellent autonomy properties, but its accumulation errors over time in indoor inertial positioning represents a significant limitation. To address these issues, we propose a novel step length model. Initially, the model utilizes the previous three steps and the difference between the maximum and minimum values of the acceleration at the step to obtain a coarse step length estimation. Subsequently, it employs the walking frequency and the acceleration variance as constraints to rectify the rough step length, with the objective of achieving a corrected step length. Finally, some extend experiments have been conducted in three scenarios. The experimental results demonstrate that the proposed method can improve the cumulative error.