Authors | Alireza Safdarinezhad, Mehdi Mokhtarzade and Mohammad Javad Valadan Zoej |
---|---|
Journal | remote sensing |
Serial number | 8 |
Volume number | 6 |
IF | 4.118 |
Paper Type | Full Paper |
Published At | 2016-6-3 |
Journal Grade | ISI |
Journal Type | Typographic |
Journal Country | Switzerland |
Abstract
The automatic registration of LiDAR data and optical images, which are heterogeneous data sources, has been a major research challenge in recent years. In this paper, a novel hierarchical method is proposed in which the least amount of interaction of a skilled operator is required. Thereby, two shadow extraction schemes, one from LiDAR and the other from high-resolution satellite images, were used, and the obtained 2D shadow maps were then considered as prospective matching entities. Taken as the base, the reconstructed LiDAR shadows were transformed to image shadows using a four-step hierarchical method starting from a coarse 2D registration model and leading to a fine 3D registration model. In the first step, a general matching was performed in the frequency domain that yielded a rough 2D similarity model that related the LiDAR and image shadow masks. This model was further improved by modeling and compensating for the local geometric distortions that existed between the two heterogeneous data sources. In the third step, shadow masks, which were organized as segmented matched patches, were the subjects of a coinciding procedure that resulted in a coarse 3D registration model. In the last hierarchical step, that model was ultimately reinforced via a precise matching between the LiDAR and image edges. The evaluation results, which were conducted on six datasets and from different relative and absolute aspects, demonstrated the efficiency of the proposed method, which had a very promising accuracy on the order of one pixel
tags: 3D registrations; shadows; LiDAR; HRSI (High Resolution Satellite Imagery); automatic matching