Ultra-wideband (UWB) technology applied in indoor localization is susceptible to non-line-of-sight, which will lead to a decrease in localization accuracy. To address this problem, a new UWB localization method is proposed. The coordinates of the labels are initially estimated using the intersection classification method. This approach is then combined with an adaptive covariance Kalman filter to optimize the estimated coordinates and ultimately reduce positioning errors. The intersection classification process involves defining circles with base stations as centers and the distances between the base stations and the tag as radii. The number of intersections between the base station circles is classified, and various methods such as line intersections, weighted circles, and weighted centroids are employed to calculate the tag′s initial position, referred to as the rough coordinates. To further refine the positioning, residuals are introduced to adjust the process noise and measurement noise parameters in the Kalman filter. Additionally, a two-stage forgetting factor is incorporated to update the covariance matrix. The rough coordinates serve as input to the adaptive covariance Kalman filter, which then produces the optimized tag coordinates. Experimental results demonstrate that method effectively reduces the maximum positioning error to 14.2 cm, with an average error of 7.65 cm and a total error variance of 2.47 cm. These improvements significantly enhance the accuracy and robustness of UWB-based indoor positioning systems, meeting the stringent demands of indoor localization applications.