A Robust Turn Detection Algorithm Based on Periodic Signal Identification

Turn detection can be widely used in location navigation, user behavior identification, and scene awareness. Accurate real-time identification of pedestrian turns contributes to improving the accuracy of positioning navigation accuracy and scene awareness. Considering the problem of the insufficient accuracy and robustness of the existing turn detection algorithm, this paper proposes a turn detection algorithm based on periodic signal recognition, which effectively solves the misjudgment caused by the periodic swing of user walking. The algorithm collects the acceleration sensor and the gyro sensor data in real time and calculates the angular velocity of the rotation in the vertical direction by multiplying the vertical gravitational acceleration unit vector by the gyro sensor data. In this paper, fast Fourier transform (FFT) is used to identify and eliminate periodic interfering signals generated by user walking, only keeping the non-periodic signal generated by user walking (such as turning). By integrating the vertical angular velocity data in the sliding window to obtain the angle change value of the user walking within a certain period of time, and the threshold value is compared to realize the accurate identification of the turning behavior. In order to adapt to the user’s turn size and the turn speed, the algorithm also proposes a corner detection method based on multi-sliding window. The experiment results show that the proposed algorithm has higher accuracy and lower power consumption than other algorithms which based on GPS and electronic compass. The algorithm can realize the accurate identification of the turning behavior of the mobile phone in various positions of the user, and has good robustness. The accuracy of turning recognition can reach 93% and the average power consumption is about 60 mW.