This invention discloses a
pedestrian inertial localization method based on a Kolmogorov-Arnold network. First, the
raw data collected by the
inertial measurement unit (IMU) of a smartphone is preprocessed. Sensor errors are eliminated through bias compensation, attitude calculation, and coordinate
system transformation, and a unified inertial
time series sample is constructed using a
sliding time window. Second, a recurrent Kolmogorov-Arnold network unit is constructed, and a parameterized B-spline
activation function is introduced into the recurrent structure to enhance the model's ability to express complex inertial
time series nonlinear features. Third, a gated memory mechanism is introduced into the recurrent unit to achieve selective updating of
time series information and long-term dependency modeling, thereby improving the model's stability in long-term walking scenarios. Subsequently, the high-dimensional features output by the network are mapped to two-dimensional instantaneous velocities, and the continuous motion trajectory of the
pedestrian is reconstructed through numerical integration. Finally, the model is trained using
supervised learning, and the performance of the localization results is evaluated using a trajectory error index. This invention can achieve high-precision
pedestrian localization relying solely on smartphone inertial data, exhibiting good localization accuracy, robustness, and practical deployment capability.