The accuracy requirements for indoor pedestrian navigation are steadily increasing. Traditional algorithms in inertial navigation systems face issues such as cumulative errors and unclear heading research, significantly impeding the application and development of inertial navigation. In response to the accumulation of errors in traditional strap-down algorithms, we propose a gait detection method based on the generalized likelihood ratio test to more effectively identify footstep stationary states. By combining Kalman filtering with sliding window, the zero velocity correction method corrects the cumulative error issue in the inertial measurement unit of the inertial system, thus addressing the problem of low pedestrian walking accuracy in navigation. Experimental results indicate that the zero velocity correction and sliding window approach can reduce endpoint positioning errors to less than 2%, providing accurate and continuous positioning information.