Multi-feature combination method for point cloud intensity feature image and UAV optical image matching

LiDAR point clouds and optical images are two widely used geospatial data. The fusion of LiDAR point clouds and optical images can take full advantage of these two types of data. Since LiDAR point clouds and optical images vary in dimension (3D vs. 2D), spectral (near-infrared vs. visible) and data acquisition principles ( time of flight vs. perspective projection), the fusion of LiDAR point clouds and optical images is challenging. This paper deals with the registration of LiDAR point clouds and optical images. Feature point-based matching methods with different feature detector and descriptor combinations are evaluated, and find that different combinations affect the matching performance greatly. Among the evaluated 112 combinations, FAST-SIFT and AGAST-SIFT combinations have the best matching performance. Besides, to remove the large amount mismatches in the matching results, the paper proposed a template and RANSAC based mismatch removal algorithm. The experimental results show that the proposed mismatch removal algorithm greatly improved the matching success rate and the correct matching rate.