Robot Vision Group
National Laboratory of Pattern Recognition
Institute of Automation, Chinese Academy of Sciences
(Supproted by Natural Science Foundation of China, National Romote Sensing Center of China)
Given a set of overlapping photographs of a scene shot from airplane (or cheap Unmanned Aerial Vehicle, UAV), we can automatically create partial or full panoramas. All the steps of our panorama generation are fully automated, including image matching, dense 3D point reconstruction, stitching, alignment, and color correction. The key feature of our system is to use our 3D reconstruction system to generate a dense but faithful 3D point cloud ( played traditional DEM role) by which the quality of the generated panorama is greatly improved. Worthy of mention is that our system needs neither ground control points nor the pose information of the airlpane.
Key features of our panorama generation:
The following is the pipeline of our method:
You can download the bigger size image from here.[image]