Surface Recovery: Fusion of Image and Point Cloud

The point cloud of the laser scanner is a rich source of information for high level tasks in computer vision such as traffic understanding. However, cost-effective laser scanners provide noisy and low resolution point cloud and they are prone to systematic errors. In this paper, we propose two surface recovery approaches based on geometry and brightness of the surface. The proposed approaches are tested in the realistic outdoor scenarios and the results show that both approaches have superior performance over the-state-of-art methods.