This paper provided a method to improve the accuracy of infrared temperature measurement on pavement by an accelerated K-means clustering separation interference signal based on the mean-standard deviation for initial central，using the temperature difference between the road surface and the vehicle，the standard deviation of the vehicle signals，and the duration of the interference signal to separate vehicle interference, low-speed obstacles, and environmental mutations data and retain environmental mutation data. The recognition rate of this method for interference signals is 95.6% on sunny days, 90.9% on cloudy days, and 78.2% on cloudy and rainy days, which effectively improves the accuracy of on-surface infrared temperature measurement on the road surface.