Abstract:

Automatic road extraction from remote sensing images in urban area has been a very challenging task due to the complexity of the scene, especially in the occluded or shadowed areas. This paper proposes an integrated method to fuse LiDAR data and high resolution imagery for automatic extraction of road centrelines. Firstly the LiDAR point cloud is filtered to get the ground points whose intensity data is used to detect initial road centrelines and key points of the roads. A road model is then built on the intensity and dispersion of the ground points as well as spectral information obtained from the high resolution image. Based on the model, the dynamic programming algorithm is applied to find the optimal road centrelines linking the key points which are selected by evaluation. The experimental results indicate its effectiveness in automatic road extraction in urban and complex scenes.

Key words: Road extraction, LIDAR, Remote sensing imagery, Information fusion, Dynamic programming