The aim of this study was to develop and experimentally validate an algorithm for automatic selection of filter frequency characteristics and detection threshold in order to enhance the accuracy and reliability of gait phase detection. This challenge is crucial not only for objective rehabilitation and monitoring of motor activity, but also for sports analytics, ergonomics, gaming and engineering applications, as well as studies of human locomotion. An automated approach for optimizing the parameters of a gait phase detector based on data from a three-axis foot-mounted accelerometer is presented. This work implements an evolutionary artificial intelligence algorithm that mimics natural selection processes, providing automatic search for the optimal gait phase detector parameters by minimizing the error between the trajectory obtained from inertial measurement units and the reference (optical) trajectory acquired using an OptiTrack system. Details are provided regarding the formation and evolution of the parameter population, design of the objective function, and drift compensation methods utilized during acceleration integration. Experiments involving walking along a closed square path confirmed the high accuracy and robustness of the proposed method: the match between the optimized and reference trajectories demonstrates the practical applicability of the approach for precise gait reconstruction under different conditions. The proposed methodology is easily adaptable to individual movement characteristics and can be integrated into modern wearable sensor systems for a wide range of scientific and applied tasks