In sensor localization, ultrasound based distance measurements will have large errors if line-of-sight paths are blocked between pairwise sensors. Because these outliers, once mixed together with other correct distance measurements, are difficult for localization algorithms to identify, we propose to exclude the outliers in the first step of distance measurement. Although the distance between pairwise sensors measured by ultrasound can have multiple values due to the multipath effect, our experiments find that a sensor's incorrect position, estimated from the distance measurement along a reflected path instead of a straight line, is always mirrored to the sensor's correct position. Based on this phenomena, we propose to use mobile beacons to measure the distance between pairwise sensors from multiple perspectives and filter incorrect values through a statistical approach. Our performance evaluation shows that the proposed algorithm can achieve better localization results than previous approaches in an indoor environment where multipath effects cannot be avoided