Abstract: Objectives: Aiming at the problems of non-line-of-sight (NLOS) error recognition and missed recognition in ultra wide band (UWB) positioning. Methods: A robust adaptive filtering algorithm based on sliding window variance detection and innovation detection is proposed. Based on the innovation robust adaptive algorithm, the sliding window variance detection combined with the innovation detection method is used to reduce the NLOS false recognition and missed recognition rate under the model disturbance state. Furthermore, the variance detection method is optimized by distance smoothing and distance updating, which solves the problem of detection degradation of variance detection. Results: The results of real experiments show that the improved algorithm achieves a positioning accuracy of 0.073m in the line-of-sight environment. In the personnel occlusion environment, the algorithm attains an accuracy of 0.077m, which improves by 40.3%, 33.6%, and 28.7% compared to the least-squares, Kalman filter, and innovation robust adaptive filtering algorithms, respectively. In more severe NLOS environments such as pillar occlusion and underground parking garages, the positioning accuracy is 0.125m. Compared to the least-squares, Kalman filter, and innovation robust adaptive filtering algorithms, the accuracy is improved by more than 80%, 73%, and 36% respectively. Additionally, compared to the innovation robust adaptive filtering algorithm, the NLOS false recognition rate is reduced by more than 38%. Conclusions: The algorithm can meet the high-precision positioning requirements in complex indoor environments.