In view of the susceptibility of wifi signals to environmental fluctuations and multipath phenomena, the accuracy of sensor modules in pedestrian dead reckoning (PDR) systems is limited and there is cumulative error. An improved unscented Kalman filter (UKF) with adaptive adjustment factor is proposed. The fusion algorithm uses the difference between the theoretical covariance and the actual covariance of the residual rk as a condition to introduce the adjustment factor ρ to adjust the state vector and the observed covariance to adjust the Kalman gain parameter, which increases the system robustness and improves the accuracy. In the experimental environment, the experimental results show that the proposed algorithm has an average error of 1.35m and a maximum error of 2.23m. The probability of positioning position error within 1.5m is about 80%, and the probability of being within 1.5m compared to the standard UKF algorithm is about 22%. The proposed algorithm is increased by about 58%.