The non-line-of-sight error in the terminal can lower the accuracy of the trilateral positioning algorithm. To address this issue, an adaptive trilateral positioning algorithm is used. By introducing the mean difference as the standard of base station selection, the algorithm selects the three base stations with the smallest mean distance difference and improves the reliability of the base station to label distance information. In the three-sided positioning algorithm, the Euclidean distance is used to decide the position relationship between the two circles, and the radius of the circle with a high error possibility is corrected by combining the distance information. Then, the circle's radius is iteratively updated. The distance between the intersection points of two circles is set to 10 cm as the iteration threshold, and the center of the circumscribed circle at the intersection point of three circles is used as the label's position. Simulation experiments are used to validate the algorithm's effectiveness after adding Gaussian noise to approximate the non-line-of-sight error. Under 10% Gaussian noise, the tag positioning error is 10.5 cm. Finally, an actual passenger positioning system experiment is constructed to verify that the tag positioning accuracy of the proposed algorithm is less than 10 cm, and position error of the proposed algorithm is 13% lower than that of the weighted centroid method in X, Y, and Z directions.