A Modified Road Centerlines Search Method from Remote Sensing Images

Abstract

Aiming at the sensitivity of the road centerline extraction algorithm using directional texture to the disturbance in the images, a modified method for road centerlines on highresolution remote sensing images is proposed based on the directional texture and Kalman Filter. After the initial center points of the road are obtained by directional texture matching, Kalman Filter combined with priori information and observation information of the road center points is applied to track the accurate road center points iteratively. Multiple experiments are designed to verify the reliability and robustness of the algorithm, showing that it can reduce the covering impact of vehicles, trees and shadow on road extraction in high-resolution images with relatively strong robustness and flexibility. The average position deviation is 1.9 pixels, and the average position deviation error is 1.7 pixels.